Joshua Lee, MD

Study Design

We developed, implemented, and refined a VTE prevention protocol and examined the impact of our efforts. We observed adult inpatients on a longitudinal basis for the prevalence of adequate VTE prophylaxis and for the incidence of HA VTE throughout a 36‐month period from calendar year 2005 through 2007, and performed a retrospective analysis for any potential adverse effects of increased VTE prophylaxis. The project adhered to the HIPAA requirements for privacy involving health‐related data from human research participants. The study was approved by the Institutional Review Board of the University of California, San Diego, which waived the requirement for individual patient informed consent.

We included all hospitalized adult patients (medical and surgical services) at our medical center in our observations and interventions, including patients of all ethnic groups, geriatric patients, prisoners, and the socially and economically disadvantaged in our population. Exclusion criteria were age under 14 years, and hospitalization on Psychiatry or Obstetrics/Gynecology services.

Development of a VTE Risk‐assessment Model and VTE Prevention Protocol

A core multidisciplinary team with hospitalists, pulmonary critical care VTE experts, pharmacists, nurses, and information specialists was formed. After gaining administrative support for standardization, we worked with medical staff leaders to gain consensus on a VTE prevention protocol for all medical and surgical areas from mid‐2005 through mid‐2006. The VTE prevention protocol included the elements of VTE risk stratification, definitions of adequate VTE prevention measures linked to the level of VTE risk, and definitions for contraindications to pharmacologic prophylactic measures. We piloted risk‐assessment model (RAM) drafts for ease of use and clarity, using rapid cycle feedback from pharmacy residents, house staff, and medical staff attending physicians. Models often cited in the literature15, 18 that include point‐based scoring of VTE risk factors (with prophylaxis choices hinging on the additive sum of scoring) were rejected based on the pilot experience.

We adopted a simple model with 3 levels of VTE risk that could be completed by the physician in seconds, and then proceeded to integrate this RAM into standardized data collection instruments and eventually (April 2006) into a computerized provider order entry (CPOE) order set (Siemmens Invision v26). Each level of VTE risk was firmly linked to a menu of acceptable prophylaxis options (Table 1). Simple text cues were used to define risk assessment, with more exhaustive listings of risk factors being relegated to accessible reference tables.

Intermittent pneumatic compression devices were endorsed as an adjunct in all patients in the highest risk level, and as the primary method in patients with contraindications to pharmacologic prophylaxis. Aspirin was deemed an inappropriate choice for VTE prophylaxis. Subcutaneous unfractionated or low‐molecular‐weight heparin were endorsed as the primary method of prophylaxis for the majority of patients without contraindications.

Integration of the VTE Protocol into Order Sets

An essential strategy for the success of the VTE protocol included integrating guidance for the physician into the flow of patient care, via standardized order sets. The CPOE VTE prevention order set was modular by design, as opposed to a stand alone design. After conferring with appropriate stakeholders, preexisting and nonstandardized prompts for VTE prophylaxis were removed from commonly used order sets, and the standardized module was inserted in its place. This allowed for integration of the standardized VTE prevention module into all admission and transfer order sets, essentially insuring that all patients admitted or transferred within the medical center would be exposed to the protocol. Physicians using a variety of admission and transfer order sets were prompted to select each patient's risk for VTE, and declare the presence or absence of contraindications to pharmacologic prophylaxis. Only the VTE prevention options most appropriate for the patient's VTE and anticoagulation risk profile were presented as the default choice for VTE prophylaxis. Explicit designation of VTE risk level and a prophylaxis choice were presented in a hard stop mechanism, and utilization of these orders was therefore mandatory, not optional. Proper use (such as the proper classification of VTE risk by the ordering physician) was actively monitored on an auditing basis, and order sets were modified occasionally on the basis of subjective and objective feedback.

Assessment of VTE Risk Assessment Interobserver Agreement

Data from 150 randomly selected patients from the audit pool (from late 2005 through mid‐2006) were abstracted by the nurse practitioner in a detailed manner. Five independent reviewers assessed each patient for VTE risk level, and for a determination of whether or not they were on adequate VTE prophylaxis on the day of the audit per protocol. Interobserver agreement was calculated for these parameters using kappa scores.

Prospective Monitoring of Adequate VTE Prophylaxis

A daily medical center inpatient census report of eligible patients in the medical center for >48 hours was downloaded into an Microsoft Excel spreadsheet, with each patient assigned a consecutive number. The Excel random number generator plug‐in function was used to generate a randomly sequenced list of the patients. The research nurse practitioner targeted serial patients on the list for further study, until she accomplished the requisite number of audits each day. The mean number of audits per month declined over the study years as the trends stabilized and as grant funding expired, but remained robust throughout (2005: 107 audits per month; 2006: 80 audits per month; and 2007: 57 audits per month).

The data collected on each patient randomly selected for audit included age, gender, location, service, date and time of review, and date of admission. The audit VTE RAM (identical to the VTE RAM incorporated into the order set), was used to classify each patient's VTE risk as low, moderate, or high. For each audit, we determined if the patient was on an adequate VTE prevention regimen consistent with our protocol, given their VTE risk level, demographics, and absence or presence of contraindications to pharmacologic prophylaxis. All questionable cases were reviewed by at least 2 physicians at weekly meetings with a final consensus determination. Adequacy of the VTE regimen was judged by orders entered on the day of the audit, but we also noted whether or not ordered intermittent compression devices were in place and functioning at the time of the audit.

Prospective (Concurrent) Discovery and Analysis of VTE Cases

The team nurse practitioner used the PACS radiology reporting and archival system (IMPAX version 4.5; AGFA Healthcare Informatics, Greenville, SC) to identify all new diagnoses of VTE, in the process described below.

Procedure codes for following studies were entered into the IMPAX search engine to locate all such exams performed in the previous 1 to 3 days:

• 1

  Ultrasound exams of the neck, upper extremities, and lower extremities;

• 2

  Computed tomography (CT) angiograms of the chest;

• 3

  Ventilation/perfusion nuclear medicine scans; and

• 4

  Pulmonary angiograms.

Negative studies and studies that revealed unchanged chronic thromboses were excluded, while clots with a chronic appearance but no evidence of prior diagnosis were included. Iliofemoral, popliteal, calf vein, subclavian, internal and external jugular vein, and axillary vein thromboses were therefore included, as were all PEs. Less common locations, such as renal vein and cavernous sinus thromboses, were excluded. The improvement/research team exerted no influence over decisions about whether or not testing was done.

Each new case of VTE was then classified as HA VTE or community‐acquired VTE. A new VTE was classified as HA if the diagnosis was first suspected and made in the hospital. A newly diagnosed VTE was also classified as HA if the VTE was suspected in the ambulatory setting, but the patient had been hospitalized within the arbitrary window of the preceding 30 days.

Each new diagnosis of HA VTE was reviewed by core members of the multidisciplinary support team. This investigation included a determination of whether the patient was on an adequate VTE prophylaxis regimen at the time of the HA VTE, using the RAM and linked prophylaxis menu described above. The VTE prevention regimen ordered at the time the inpatient developed the HA VTE was classified as adherent or nonadherent to the University of California, San Diego (UCSD) protocol: patients who developed VTE when on suboptimal prophylaxis per protocol were classified as having a potentially preventable case. Potentially iatrogenic precipitants of VTE (such as the presence of a central venous catheter or restraints) were also noted. All data were entered into a Microsoft Access database for ease of retrieval and reporting.

All tests for VTE were performed based on clinical signs and symptoms, rather than routine screening, except for the Trauma and Burn services, which also screen for VTE in high‐risk patients per their established screening protocols.

Statistical Analysis of VTE Prophylaxis and HA VTE Cases

Gender differences between cases of VTE and randomly sampled and audited inpatients were examined by chi‐square analysis, and analysis of variance (ANOVA) was used to examine any age or body mass index (BMI) differences between audits and cases.

The unadjusted risk ratio (RR) for adequate prophylaxis was compared by year, with year 2005 being the baseline (comparison) year, by chi‐square analysis.

The unadjusted RR of HA VTE was calculated by dividing the number of cases found in the calendar year by the hospital census of adult inpatients at risk. For each case, a classification for the type of VTE (PE vs. DVT vs. combinations) was recorded. Cases not receiving adequate prophylaxis were categorized as preventable DVT. Unadjusted RRs were calculated for each year by chi‐square analysis, compared to the baseline (2005) year.

All data were analyzed using Stata (version 10; Stata Corp., College Station, TX). Results for the different analysis were considered significant at P < 0.05.

Retrospective Study of Unintentional Adverse Effects

The increase in anticoagulant use accompanying the introduction of the VTE prophylaxis order set warranted an evaluation of any subsequent rise in related adverse events. A study was done to determine the rates of bleeding and heparin‐induced thrombocytopenia (HIT) before and after the implementation of the VTE prophylaxis order set.

A retrospective analysis was conducted to evaluate outcomes in our inpatients from December 2004 through November 2006, with April to November, 2006 representing the post‐order set implementation time period. Any patient with a discharge diagnosis code of e934.2 (anticoagulant‐related adverse event) was selected for study to identify possible bleeding attributable to pharmacologic VTE prophylaxis. Major or minor bleeding attributable to pharmacologic VTE prophylaxis was defined as a bleed occurring 72 hours after receiving pharmacologic VTE prophylaxis. Major bleeding was defined as cerebrovascular, gastrointestinal, retroperitoneal, or overt bleeding with a decrease in hemoglobin 2 mg/dL with clinical symptoms such as hypotension or hypoxia (not associated with hemodialysis) or transfusion of 2 units of packed red blood cells. Minor bleeding was defined as ecchymosis, epistaxis, hematoma, hematuria, hemoptysis, petechiae, or bleeding without a decrease in hemoglobin 2 g/dL.

Possible cases of HIT were identified by screening for a concomitant secondary thrombocytopenia code (287.4). Chart review was then conducted to determine a causal relationship between the use of pharmacologic VTE prophylaxis and adverse events during the hospital stay. HIT attributable to pharmacologic VTE prophylaxis was determined by assessing if patients developed any of the following clinical criteria after receiving pharmacologic VTE prophylaxis: platelet count <150 10⁹/L or 50% decrease from baseline, with or without an associated venous or arterial thrombosis or other sequelae (skin lesions at injection site, acute systemic reaction) and/or a positive heparin‐induced platelet activation (HIPA) test. In order to receive a diagnosis of HIT, thrombocytopenia must have occurred between days 5 to 15 of heparin therapy, unless existing evidence suggested that the patient developed rapid‐onset HIT or delayed‐onset HIT. Rapid‐onset HIT was defined as an abrupt drop in platelet count upon receiving a heparin product, due to heparin exposure within the previous 100 days. Delayed‐onset HIT was defined as HIT that developed several days after discontinuation of heparin. Other evident causes of thrombocytopenia were ruled out.

Statistical Analysis of Retrospective Study of Unintentional Adverse Effects

Regression analysis with chi‐square and ANOVA were used in the analysis of the demographic data. RRs were calculated for the number of cases coded with an anticoagulant‐related adverse event secondary thrombocytopenia before and after the order set implementation.

Educational Efforts and Feedback

Members of the multidisciplinary team presented information on HA VTE and the VTE prevention protocol at Medical and Surgical grand rounds, teaching rounds, and noon conference, averaging 1 educational session per quarter. Feedback and education was provided to physicians and nursing staff when audits revealed that a patient had inadequate prophylaxis with reference to the protocol standard. In addition, these conversations provided on opportunity to explore reasons for nonadherence with the protocol, confusion regarding the VTE RAM, and other barriers to effective prophylaxis, thereby providing guidance for further protocol revision and educational efforts. We adjusted the order set based on active monitoring of order set use and the audit process.

Interobserver Agreement

The VTE RAM interobserver agreement was assessed on 150 patients with 5 observers as described above. The kappa score for the VTE risk level was 0.81. The kappa score for the judgment of whether the patient was on adequate prophylaxis or not was 0.90.

Impact on Percent of Patients with Adequate Prophylaxis (Longitudinal Audits)

Audits of randomly sampled inpatients occurred longitudinally throughout the study period as described above. Based on the intervention, the percent of patients on adequate prophylaxis improved significantly (P < 0.001) by each calendar year (see Table 3), from a baseline of 58% in 2005 to 78% in 2006 (unadjusted relative benefit = 1.35; 95% confidence interval [CI] = 1.28‐1.43), and 93% in 2007 (unadjusted relative benefit = 1.61; 95% CI = 1.52, 1.69). The improvement seen was more marked in the moderate VTE risk patients when compared to the high VTE risk patients. The percent of audited VTE prophylaxis improved from 53% in calendar year (CY) 2005 to 93% in 2007 (unadjusted relative benefit = 1.75; 95% CI = 1.70‐1.81) in the moderate VTE risk group, while the high VTE risk group improved from 83% to 92% in the same time period (unadjusted relative benefit = 1.11; 95% CI = 0.95‐1.25).

Overall, adequate VTE prophylaxis was present in over 98% of audited patients in the last 6 months of 2007, and this high rate has been sustained throughout 2008. Age, ethnicity, and gender were not associated with differential rates of adequate VTE prophylaxis.

Figure 1 is a timeline of interventions and the impact on the prevalence of adequate VTE prophylaxis. The first 7 to 8 months represent the baseline rate 50% to 55% of VTE prophylaxis. In this baseline period, the improvement team was meeting, but had not yet begun meeting with the large variety of medical and surgical service leaders. Consensus‐building sessions with these leaders in the latter part of 2005 through mid‐2006 correlated with improvement in adequate VTE prophylaxis rates to near 70%. The consensus‐building sessions also prepared these varied services for a go live date of the modular order set that was incorporated into all admit and transfer order sets, often replacing preexisting orders referring to VTE prevention measures. The order set resulted in an improvement to 80% adequate prophylaxis, with the incremental improvement occurring virtually overnight with the go live date at the onset of quarter 2 (Q2) of 2006. Monitoring of the order set use confirmed that it was easy and efficient to use, but also revealed that physicians were at times classifying patients as low VTE risk inaccurately, when they possessed qualities that actually qualified them for moderate risk status by our protocol. We therefore inserted a secondary CPOE screen when patients were categorized as low VTE risk, asking the physician to deny or confirm that the patient had no risk factors that qualified them for moderate risk status. This confirmation screen essentially acted as a reminder to the physician to ask Are you sure this patient does not need VTE prophylaxis? This minor modification of the CPOE order set improved adequate VTE prophylaxis rates to 90%. Finally, we asked nurses to evaluate patients who were not on therapeutic or prophylactic doses of anticoagulants. Patients with VTE risk factors but no obvious contraindications generated a note from the nurse to the doctor, prompting the doctor to reassess VTE risk and potential contraindications. This simple intervention raised the percent of audited patients on adequate VTE prophylaxis to 98% in the last 6 months of 2007.

Figure 1

Description of Prospectively Identified VTE

We identified 748 cases of VTE among patients admitted to the medical center over the 36‐month study period; 387 (52%) were community‐acquired VTE. There were 361 HA cases (48% of total cases) over the same time period. There was no difference in age, gender, or BMI between the community‐acquired and hospital‐related VTE.

Of the 361 HA cases, 199 (55%) occurred on Surgical Services and 162 (45%) occurred on Medical Services; 58 (16%) unique patients had pulmonary emboli, while 303 (84%) patients experienced only DVT. Remarkably, almost one‐third of the DVT occurred in the upper extremities (108 upper extremities, 240 lower extremities), and most (80%) of the upper‐extremity DVT were associated with central venous catheters.

Of 361 HA VTE cases, 292 (81%) occurred in those in the moderate VTE risk category, 69 HA VTE cases occurred in high‐risk category patients (19%), and no VTE occurred in patients in the low‐risk category.

Improvement in HA VTE

HA VTE were identified and each case analyzed on an ongoing basis over the entire 3 year study period, as described above. Table 4 depicts a comparison of HA VTE on a year‐to‐year basis and the impact of the VTE prevention protocol on the incidence of HA VTE. In 2007 (the first full CY after the implementation of the order set) there was a 39% relative risk reduction (RRR) in the risk of experiencing an HA VTE. The reduction in the risk of preventable HA VTE was even more marked (RRR = 86%; 7 preventable VTE in 2007, compared to 44 in baseline year of 2005; RR = 0.14; 95% CI = 0.06‐0.31).

Retrospective Analysis of Impact on HIT and Bleeding

There were no statistically significant differences in the number of cases coded for an anticoagulant‐related bleed or secondary thrombocytopenia (Table 5). Chart review revealed there were 2 cases of minor bleeding attributable to pharmacologic VTE prophylaxis before the order set implementation. There were no cases after implementation. No cases of HIT attributable to pharmacologic VTE prophylaxis were identified in either study period, with all cases being attributed to therapeutic anticoagulation.

Study Design

We developed, implemented, and refined a VTE prevention protocol and examined the impact of our efforts. We observed adult inpatients on a longitudinal basis for the prevalence of adequate VTE prophylaxis and for the incidence of HA VTE throughout a 36‐month period from calendar year 2005 through 2007, and performed a retrospective analysis for any potential adverse effects of increased VTE prophylaxis. The project adhered to the HIPAA requirements for privacy involving health‐related data from human research participants. The study was approved by the Institutional Review Board of the University of California, San Diego, which waived the requirement for individual patient informed consent.

We included all hospitalized adult patients (medical and surgical services) at our medical center in our observations and interventions, including patients of all ethnic groups, geriatric patients, prisoners, and the socially and economically disadvantaged in our population. Exclusion criteria were age under 14 years, and hospitalization on Psychiatry or Obstetrics/Gynecology services.

Development of a VTE Risk‐assessment Model and VTE Prevention Protocol

A core multidisciplinary team with hospitalists, pulmonary critical care VTE experts, pharmacists, nurses, and information specialists was formed. After gaining administrative support for standardization, we worked with medical staff leaders to gain consensus on a VTE prevention protocol for all medical and surgical areas from mid‐2005 through mid‐2006. The VTE prevention protocol included the elements of VTE risk stratification, definitions of adequate VTE prevention measures linked to the level of VTE risk, and definitions for contraindications to pharmacologic prophylactic measures. We piloted risk‐assessment model (RAM) drafts for ease of use and clarity, using rapid cycle feedback from pharmacy residents, house staff, and medical staff attending physicians. Models often cited in the literature15, 18 that include point‐based scoring of VTE risk factors (with prophylaxis choices hinging on the additive sum of scoring) were rejected based on the pilot experience.

We adopted a simple model with 3 levels of VTE risk that could be completed by the physician in seconds, and then proceeded to integrate this RAM into standardized data collection instruments and eventually (April 2006) into a computerized provider order entry (CPOE) order set (Siemmens Invision v26). Each level of VTE risk was firmly linked to a menu of acceptable prophylaxis options (Table 1). Simple text cues were used to define risk assessment, with more exhaustive listings of risk factors being relegated to accessible reference tables.

Intermittent pneumatic compression devices were endorsed as an adjunct in all patients in the highest risk level, and as the primary method in patients with contraindications to pharmacologic prophylaxis. Aspirin was deemed an inappropriate choice for VTE prophylaxis. Subcutaneous unfractionated or low‐molecular‐weight heparin were endorsed as the primary method of prophylaxis for the majority of patients without contraindications.

Integration of the VTE Protocol into Order Sets

An essential strategy for the success of the VTE protocol included integrating guidance for the physician into the flow of patient care, via standardized order sets. The CPOE VTE prevention order set was modular by design, as opposed to a stand alone design. After conferring with appropriate stakeholders, preexisting and nonstandardized prompts for VTE prophylaxis were removed from commonly used order sets, and the standardized module was inserted in its place. This allowed for integration of the standardized VTE prevention module into all admission and transfer order sets, essentially insuring that all patients admitted or transferred within the medical center would be exposed to the protocol. Physicians using a variety of admission and transfer order sets were prompted to select each patient's risk for VTE, and declare the presence or absence of contraindications to pharmacologic prophylaxis. Only the VTE prevention options most appropriate for the patient's VTE and anticoagulation risk profile were presented as the default choice for VTE prophylaxis. Explicit designation of VTE risk level and a prophylaxis choice were presented in a hard stop mechanism, and utilization of these orders was therefore mandatory, not optional. Proper use (such as the proper classification of VTE risk by the ordering physician) was actively monitored on an auditing basis, and order sets were modified occasionally on the basis of subjective and objective feedback.

Assessment of VTE Risk Assessment Interobserver Agreement

Data from 150 randomly selected patients from the audit pool (from late 2005 through mid‐2006) were abstracted by the nurse practitioner in a detailed manner. Five independent reviewers assessed each patient for VTE risk level, and for a determination of whether or not they were on adequate VTE prophylaxis on the day of the audit per protocol. Interobserver agreement was calculated for these parameters using kappa scores.

Prospective Monitoring of Adequate VTE Prophylaxis

A daily medical center inpatient census report of eligible patients in the medical center for >48 hours was downloaded into an Microsoft Excel spreadsheet, with each patient assigned a consecutive number. The Excel random number generator plug‐in function was used to generate a randomly sequenced list of the patients. The research nurse practitioner targeted serial patients on the list for further study, until she accomplished the requisite number of audits each day. The mean number of audits per month declined over the study years as the trends stabilized and as grant funding expired, but remained robust throughout (2005: 107 audits per month; 2006: 80 audits per month; and 2007: 57 audits per month).

The data collected on each patient randomly selected for audit included age, gender, location, service, date and time of review, and date of admission. The audit VTE RAM (identical to the VTE RAM incorporated into the order set), was used to classify each patient's VTE risk as low, moderate, or high. For each audit, we determined if the patient was on an adequate VTE prevention regimen consistent with our protocol, given their VTE risk level, demographics, and absence or presence of contraindications to pharmacologic prophylaxis. All questionable cases were reviewed by at least 2 physicians at weekly meetings with a final consensus determination. Adequacy of the VTE regimen was judged by orders entered on the day of the audit, but we also noted whether or not ordered intermittent compression devices were in place and functioning at the time of the audit.

Prospective (Concurrent) Discovery and Analysis of VTE Cases

The team nurse practitioner used the PACS radiology reporting and archival system (IMPAX version 4.5; AGFA Healthcare Informatics, Greenville, SC) to identify all new diagnoses of VTE, in the process described below.

Procedure codes for following studies were entered into the IMPAX search engine to locate all such exams performed in the previous 1 to 3 days:

• 1

  Ultrasound exams of the neck, upper extremities, and lower extremities;

• 2

  Computed tomography (CT) angiograms of the chest;

• 3

  Ventilation/perfusion nuclear medicine scans; and

• 4

  Pulmonary angiograms.

Negative studies and studies that revealed unchanged chronic thromboses were excluded, while clots with a chronic appearance but no evidence of prior diagnosis were included. Iliofemoral, popliteal, calf vein, subclavian, internal and external jugular vein, and axillary vein thromboses were therefore included, as were all PEs. Less common locations, such as renal vein and cavernous sinus thromboses, were excluded. The improvement/research team exerted no influence over decisions about whether or not testing was done.

Each new case of VTE was then classified as HA VTE or community‐acquired VTE. A new VTE was classified as HA if the diagnosis was first suspected and made in the hospital. A newly diagnosed VTE was also classified as HA if the VTE was suspected in the ambulatory setting, but the patient had been hospitalized within the arbitrary window of the preceding 30 days.

Each new diagnosis of HA VTE was reviewed by core members of the multidisciplinary support team. This investigation included a determination of whether the patient was on an adequate VTE prophylaxis regimen at the time of the HA VTE, using the RAM and linked prophylaxis menu described above. The VTE prevention regimen ordered at the time the inpatient developed the HA VTE was classified as adherent or nonadherent to the University of California, San Diego (UCSD) protocol: patients who developed VTE when on suboptimal prophylaxis per protocol were classified as having a potentially preventable case. Potentially iatrogenic precipitants of VTE (such as the presence of a central venous catheter or restraints) were also noted. All data were entered into a Microsoft Access database for ease of retrieval and reporting.

All tests for VTE were performed based on clinical signs and symptoms, rather than routine screening, except for the Trauma and Burn services, which also screen for VTE in high‐risk patients per their established screening protocols.

Statistical Analysis of VTE Prophylaxis and HA VTE Cases

Gender differences between cases of VTE and randomly sampled and audited inpatients were examined by chi‐square analysis, and analysis of variance (ANOVA) was used to examine any age or body mass index (BMI) differences between audits and cases.

The unadjusted risk ratio (RR) for adequate prophylaxis was compared by year, with year 2005 being the baseline (comparison) year, by chi‐square analysis.

The unadjusted RR of HA VTE was calculated by dividing the number of cases found in the calendar year by the hospital census of adult inpatients at risk. For each case, a classification for the type of VTE (PE vs. DVT vs. combinations) was recorded. Cases not receiving adequate prophylaxis were categorized as preventable DVT. Unadjusted RRs were calculated for each year by chi‐square analysis, compared to the baseline (2005) year.

All data were analyzed using Stata (version 10; Stata Corp., College Station, TX). Results for the different analysis were considered significant at P < 0.05.

Retrospective Study of Unintentional Adverse Effects

The increase in anticoagulant use accompanying the introduction of the VTE prophylaxis order set warranted an evaluation of any subsequent rise in related adverse events. A study was done to determine the rates of bleeding and heparin‐induced thrombocytopenia (HIT) before and after the implementation of the VTE prophylaxis order set.

A retrospective analysis was conducted to evaluate outcomes in our inpatients from December 2004 through November 2006, with April to November, 2006 representing the post‐order set implementation time period. Any patient with a discharge diagnosis code of e934.2 (anticoagulant‐related adverse event) was selected for study to identify possible bleeding attributable to pharmacologic VTE prophylaxis. Major or minor bleeding attributable to pharmacologic VTE prophylaxis was defined as a bleed occurring 72 hours after receiving pharmacologic VTE prophylaxis. Major bleeding was defined as cerebrovascular, gastrointestinal, retroperitoneal, or overt bleeding with a decrease in hemoglobin 2 mg/dL with clinical symptoms such as hypotension or hypoxia (not associated with hemodialysis) or transfusion of 2 units of packed red blood cells. Minor bleeding was defined as ecchymosis, epistaxis, hematoma, hematuria, hemoptysis, petechiae, or bleeding without a decrease in hemoglobin 2 g/dL.

Possible cases of HIT were identified by screening for a concomitant secondary thrombocytopenia code (287.4). Chart review was then conducted to determine a causal relationship between the use of pharmacologic VTE prophylaxis and adverse events during the hospital stay. HIT attributable to pharmacologic VTE prophylaxis was determined by assessing if patients developed any of the following clinical criteria after receiving pharmacologic VTE prophylaxis: platelet count <150 10⁹/L or 50% decrease from baseline, with or without an associated venous or arterial thrombosis or other sequelae (skin lesions at injection site, acute systemic reaction) and/or a positive heparin‐induced platelet activation (HIPA) test. In order to receive a diagnosis of HIT, thrombocytopenia must have occurred between days 5 to 15 of heparin therapy, unless existing evidence suggested that the patient developed rapid‐onset HIT or delayed‐onset HIT. Rapid‐onset HIT was defined as an abrupt drop in platelet count upon receiving a heparin product, due to heparin exposure within the previous 100 days. Delayed‐onset HIT was defined as HIT that developed several days after discontinuation of heparin. Other evident causes of thrombocytopenia were ruled out.

Statistical Analysis of Retrospective Study of Unintentional Adverse Effects

Regression analysis with chi‐square and ANOVA were used in the analysis of the demographic data. RRs were calculated for the number of cases coded with an anticoagulant‐related adverse event secondary thrombocytopenia before and after the order set implementation.

Educational Efforts and Feedback

Members of the multidisciplinary team presented information on HA VTE and the VTE prevention protocol at Medical and Surgical grand rounds, teaching rounds, and noon conference, averaging 1 educational session per quarter. Feedback and education was provided to physicians and nursing staff when audits revealed that a patient had inadequate prophylaxis with reference to the protocol standard. In addition, these conversations provided on opportunity to explore reasons for nonadherence with the protocol, confusion regarding the VTE RAM, and other barriers to effective prophylaxis, thereby providing guidance for further protocol revision and educational efforts. We adjusted the order set based on active monitoring of order set use and the audit process.

Interobserver Agreement

The VTE RAM interobserver agreement was assessed on 150 patients with 5 observers as described above. The kappa score for the VTE risk level was 0.81. The kappa score for the judgment of whether the patient was on adequate prophylaxis or not was 0.90.

Impact on Percent of Patients with Adequate Prophylaxis (Longitudinal Audits)

Audits of randomly sampled inpatients occurred longitudinally throughout the study period as described above. Based on the intervention, the percent of patients on adequate prophylaxis improved significantly (P < 0.001) by each calendar year (see Table 3), from a baseline of 58% in 2005 to 78% in 2006 (unadjusted relative benefit = 1.35; 95% confidence interval [CI] = 1.28‐1.43), and 93% in 2007 (unadjusted relative benefit = 1.61; 95% CI = 1.52, 1.69). The improvement seen was more marked in the moderate VTE risk patients when compared to the high VTE risk patients. The percent of audited VTE prophylaxis improved from 53% in calendar year (CY) 2005 to 93% in 2007 (unadjusted relative benefit = 1.75; 95% CI = 1.70‐1.81) in the moderate VTE risk group, while the high VTE risk group improved from 83% to 92% in the same time period (unadjusted relative benefit = 1.11; 95% CI = 0.95‐1.25).

Overall, adequate VTE prophylaxis was present in over 98% of audited patients in the last 6 months of 2007, and this high rate has been sustained throughout 2008. Age, ethnicity, and gender were not associated with differential rates of adequate VTE prophylaxis.

Figure 1 is a timeline of interventions and the impact on the prevalence of adequate VTE prophylaxis. The first 7 to 8 months represent the baseline rate 50% to 55% of VTE prophylaxis. In this baseline period, the improvement team was meeting, but had not yet begun meeting with the large variety of medical and surgical service leaders. Consensus‐building sessions with these leaders in the latter part of 2005 through mid‐2006 correlated with improvement in adequate VTE prophylaxis rates to near 70%. The consensus‐building sessions also prepared these varied services for a go live date of the modular order set that was incorporated into all admit and transfer order sets, often replacing preexisting orders referring to VTE prevention measures. The order set resulted in an improvement to 80% adequate prophylaxis, with the incremental improvement occurring virtually overnight with the go live date at the onset of quarter 2 (Q2) of 2006. Monitoring of the order set use confirmed that it was easy and efficient to use, but also revealed that physicians were at times classifying patients as low VTE risk inaccurately, when they possessed qualities that actually qualified them for moderate risk status by our protocol. We therefore inserted a secondary CPOE screen when patients were categorized as low VTE risk, asking the physician to deny or confirm that the patient had no risk factors that qualified them for moderate risk status. This confirmation screen essentially acted as a reminder to the physician to ask Are you sure this patient does not need VTE prophylaxis? This minor modification of the CPOE order set improved adequate VTE prophylaxis rates to 90%. Finally, we asked nurses to evaluate patients who were not on therapeutic or prophylactic doses of anticoagulants. Patients with VTE risk factors but no obvious contraindications generated a note from the nurse to the doctor, prompting the doctor to reassess VTE risk and potential contraindications. This simple intervention raised the percent of audited patients on adequate VTE prophylaxis to 98% in the last 6 months of 2007.

Figure 1

Description of Prospectively Identified VTE

We identified 748 cases of VTE among patients admitted to the medical center over the 36‐month study period; 387 (52%) were community‐acquired VTE. There were 361 HA cases (48% of total cases) over the same time period. There was no difference in age, gender, or BMI between the community‐acquired and hospital‐related VTE.

Of the 361 HA cases, 199 (55%) occurred on Surgical Services and 162 (45%) occurred on Medical Services; 58 (16%) unique patients had pulmonary emboli, while 303 (84%) patients experienced only DVT. Remarkably, almost one‐third of the DVT occurred in the upper extremities (108 upper extremities, 240 lower extremities), and most (80%) of the upper‐extremity DVT were associated with central venous catheters.

Of 361 HA VTE cases, 292 (81%) occurred in those in the moderate VTE risk category, 69 HA VTE cases occurred in high‐risk category patients (19%), and no VTE occurred in patients in the low‐risk category.

Improvement in HA VTE

HA VTE were identified and each case analyzed on an ongoing basis over the entire 3 year study period, as described above. Table 4 depicts a comparison of HA VTE on a year‐to‐year basis and the impact of the VTE prevention protocol on the incidence of HA VTE. In 2007 (the first full CY after the implementation of the order set) there was a 39% relative risk reduction (RRR) in the risk of experiencing an HA VTE. The reduction in the risk of preventable HA VTE was even more marked (RRR = 86%; 7 preventable VTE in 2007, compared to 44 in baseline year of 2005; RR = 0.14; 95% CI = 0.06‐0.31).

Retrospective Analysis of Impact on HIT and Bleeding

There were no statistically significant differences in the number of cases coded for an anticoagulant‐related bleed or secondary thrombocytopenia (Table 5). Chart review revealed there were 2 cases of minor bleeding attributable to pharmacologic VTE prophylaxis before the order set implementation. There were no cases after implementation. No cases of HIT attributable to pharmacologic VTE prophylaxis were identified in either study period, with all cases being attributed to therapeutic anticoagulation.

Study Design

We developed, implemented, and refined a VTE prevention protocol and examined the impact of our efforts. We observed adult inpatients on a longitudinal basis for the prevalence of adequate VTE prophylaxis and for the incidence of HA VTE throughout a 36‐month period from calendar year 2005 through 2007, and performed a retrospective analysis for any potential adverse effects of increased VTE prophylaxis. The project adhered to the HIPAA requirements for privacy involving health‐related data from human research participants. The study was approved by the Institutional Review Board of the University of California, San Diego, which waived the requirement for individual patient informed consent.

We included all hospitalized adult patients (medical and surgical services) at our medical center in our observations and interventions, including patients of all ethnic groups, geriatric patients, prisoners, and the socially and economically disadvantaged in our population. Exclusion criteria were age under 14 years, and hospitalization on Psychiatry or Obstetrics/Gynecology services.

Development of a VTE Risk‐assessment Model and VTE Prevention Protocol

A core multidisciplinary team with hospitalists, pulmonary critical care VTE experts, pharmacists, nurses, and information specialists was formed. After gaining administrative support for standardization, we worked with medical staff leaders to gain consensus on a VTE prevention protocol for all medical and surgical areas from mid‐2005 through mid‐2006. The VTE prevention protocol included the elements of VTE risk stratification, definitions of adequate VTE prevention measures linked to the level of VTE risk, and definitions for contraindications to pharmacologic prophylactic measures. We piloted risk‐assessment model (RAM) drafts for ease of use and clarity, using rapid cycle feedback from pharmacy residents, house staff, and medical staff attending physicians. Models often cited in the literature15, 18 that include point‐based scoring of VTE risk factors (with prophylaxis choices hinging on the additive sum of scoring) were rejected based on the pilot experience.

We adopted a simple model with 3 levels of VTE risk that could be completed by the physician in seconds, and then proceeded to integrate this RAM into standardized data collection instruments and eventually (April 2006) into a computerized provider order entry (CPOE) order set (Siemmens Invision v26). Each level of VTE risk was firmly linked to a menu of acceptable prophylaxis options (Table 1). Simple text cues were used to define risk assessment, with more exhaustive listings of risk factors being relegated to accessible reference tables.

Intermittent pneumatic compression devices were endorsed as an adjunct in all patients in the highest risk level, and as the primary method in patients with contraindications to pharmacologic prophylaxis. Aspirin was deemed an inappropriate choice for VTE prophylaxis. Subcutaneous unfractionated or low‐molecular‐weight heparin were endorsed as the primary method of prophylaxis for the majority of patients without contraindications.

Integration of the VTE Protocol into Order Sets

An essential strategy for the success of the VTE protocol included integrating guidance for the physician into the flow of patient care, via standardized order sets. The CPOE VTE prevention order set was modular by design, as opposed to a stand alone design. After conferring with appropriate stakeholders, preexisting and nonstandardized prompts for VTE prophylaxis were removed from commonly used order sets, and the standardized module was inserted in its place. This allowed for integration of the standardized VTE prevention module into all admission and transfer order sets, essentially insuring that all patients admitted or transferred within the medical center would be exposed to the protocol. Physicians using a variety of admission and transfer order sets were prompted to select each patient's risk for VTE, and declare the presence or absence of contraindications to pharmacologic prophylaxis. Only the VTE prevention options most appropriate for the patient's VTE and anticoagulation risk profile were presented as the default choice for VTE prophylaxis. Explicit designation of VTE risk level and a prophylaxis choice were presented in a hard stop mechanism, and utilization of these orders was therefore mandatory, not optional. Proper use (such as the proper classification of VTE risk by the ordering physician) was actively monitored on an auditing basis, and order sets were modified occasionally on the basis of subjective and objective feedback.

Assessment of VTE Risk Assessment Interobserver Agreement

Data from 150 randomly selected patients from the audit pool (from late 2005 through mid‐2006) were abstracted by the nurse practitioner in a detailed manner. Five independent reviewers assessed each patient for VTE risk level, and for a determination of whether or not they were on adequate VTE prophylaxis on the day of the audit per protocol. Interobserver agreement was calculated for these parameters using kappa scores.

Prospective Monitoring of Adequate VTE Prophylaxis

A daily medical center inpatient census report of eligible patients in the medical center for >48 hours was downloaded into an Microsoft Excel spreadsheet, with each patient assigned a consecutive number. The Excel random number generator plug‐in function was used to generate a randomly sequenced list of the patients. The research nurse practitioner targeted serial patients on the list for further study, until she accomplished the requisite number of audits each day. The mean number of audits per month declined over the study years as the trends stabilized and as grant funding expired, but remained robust throughout (2005: 107 audits per month; 2006: 80 audits per month; and 2007: 57 audits per month).

The data collected on each patient randomly selected for audit included age, gender, location, service, date and time of review, and date of admission. The audit VTE RAM (identical to the VTE RAM incorporated into the order set), was used to classify each patient's VTE risk as low, moderate, or high. For each audit, we determined if the patient was on an adequate VTE prevention regimen consistent with our protocol, given their VTE risk level, demographics, and absence or presence of contraindications to pharmacologic prophylaxis. All questionable cases were reviewed by at least 2 physicians at weekly meetings with a final consensus determination. Adequacy of the VTE regimen was judged by orders entered on the day of the audit, but we also noted whether or not ordered intermittent compression devices were in place and functioning at the time of the audit.

Prospective (Concurrent) Discovery and Analysis of VTE Cases

The team nurse practitioner used the PACS radiology reporting and archival system (IMPAX version 4.5; AGFA Healthcare Informatics, Greenville, SC) to identify all new diagnoses of VTE, in the process described below.

Procedure codes for following studies were entered into the IMPAX search engine to locate all such exams performed in the previous 1 to 3 days:

• 1

  Ultrasound exams of the neck, upper extremities, and lower extremities;

• 2

  Computed tomography (CT) angiograms of the chest;

• 3

  Ventilation/perfusion nuclear medicine scans; and

• 4

  Pulmonary angiograms.

Negative studies and studies that revealed unchanged chronic thromboses were excluded, while clots with a chronic appearance but no evidence of prior diagnosis were included. Iliofemoral, popliteal, calf vein, subclavian, internal and external jugular vein, and axillary vein thromboses were therefore included, as were all PEs. Less common locations, such as renal vein and cavernous sinus thromboses, were excluded. The improvement/research team exerted no influence over decisions about whether or not testing was done.

Each new case of VTE was then classified as HA VTE or community‐acquired VTE. A new VTE was classified as HA if the diagnosis was first suspected and made in the hospital. A newly diagnosed VTE was also classified as HA if the VTE was suspected in the ambulatory setting, but the patient had been hospitalized within the arbitrary window of the preceding 30 days.

Each new diagnosis of HA VTE was reviewed by core members of the multidisciplinary support team. This investigation included a determination of whether the patient was on an adequate VTE prophylaxis regimen at the time of the HA VTE, using the RAM and linked prophylaxis menu described above. The VTE prevention regimen ordered at the time the inpatient developed the HA VTE was classified as adherent or nonadherent to the University of California, San Diego (UCSD) protocol: patients who developed VTE when on suboptimal prophylaxis per protocol were classified as having a potentially preventable case. Potentially iatrogenic precipitants of VTE (such as the presence of a central venous catheter or restraints) were also noted. All data were entered into a Microsoft Access database for ease of retrieval and reporting.

All tests for VTE were performed based on clinical signs and symptoms, rather than routine screening, except for the Trauma and Burn services, which also screen for VTE in high‐risk patients per their established screening protocols.

Statistical Analysis of VTE Prophylaxis and HA VTE Cases

Gender differences between cases of VTE and randomly sampled and audited inpatients were examined by chi‐square analysis, and analysis of variance (ANOVA) was used to examine any age or body mass index (BMI) differences between audits and cases.

The unadjusted risk ratio (RR) for adequate prophylaxis was compared by year, with year 2005 being the baseline (comparison) year, by chi‐square analysis.

The unadjusted RR of HA VTE was calculated by dividing the number of cases found in the calendar year by the hospital census of adult inpatients at risk. For each case, a classification for the type of VTE (PE vs. DVT vs. combinations) was recorded. Cases not receiving adequate prophylaxis were categorized as preventable DVT. Unadjusted RRs were calculated for each year by chi‐square analysis, compared to the baseline (2005) year.

All data were analyzed using Stata (version 10; Stata Corp., College Station, TX). Results for the different analysis were considered significant at P < 0.05.

Retrospective Study of Unintentional Adverse Effects

The increase in anticoagulant use accompanying the introduction of the VTE prophylaxis order set warranted an evaluation of any subsequent rise in related adverse events. A study was done to determine the rates of bleeding and heparin‐induced thrombocytopenia (HIT) before and after the implementation of the VTE prophylaxis order set.

A retrospective analysis was conducted to evaluate outcomes in our inpatients from December 2004 through November 2006, with April to November, 2006 representing the post‐order set implementation time period. Any patient with a discharge diagnosis code of e934.2 (anticoagulant‐related adverse event) was selected for study to identify possible bleeding attributable to pharmacologic VTE prophylaxis. Major or minor bleeding attributable to pharmacologic VTE prophylaxis was defined as a bleed occurring 72 hours after receiving pharmacologic VTE prophylaxis. Major bleeding was defined as cerebrovascular, gastrointestinal, retroperitoneal, or overt bleeding with a decrease in hemoglobin 2 mg/dL with clinical symptoms such as hypotension or hypoxia (not associated with hemodialysis) or transfusion of 2 units of packed red blood cells. Minor bleeding was defined as ecchymosis, epistaxis, hematoma, hematuria, hemoptysis, petechiae, or bleeding without a decrease in hemoglobin 2 g/dL.

Possible cases of HIT were identified by screening for a concomitant secondary thrombocytopenia code (287.4). Chart review was then conducted to determine a causal relationship between the use of pharmacologic VTE prophylaxis and adverse events during the hospital stay. HIT attributable to pharmacologic VTE prophylaxis was determined by assessing if patients developed any of the following clinical criteria after receiving pharmacologic VTE prophylaxis: platelet count <150 10⁹/L or 50% decrease from baseline, with or without an associated venous or arterial thrombosis or other sequelae (skin lesions at injection site, acute systemic reaction) and/or a positive heparin‐induced platelet activation (HIPA) test. In order to receive a diagnosis of HIT, thrombocytopenia must have occurred between days 5 to 15 of heparin therapy, unless existing evidence suggested that the patient developed rapid‐onset HIT or delayed‐onset HIT. Rapid‐onset HIT was defined as an abrupt drop in platelet count upon receiving a heparin product, due to heparin exposure within the previous 100 days. Delayed‐onset HIT was defined as HIT that developed several days after discontinuation of heparin. Other evident causes of thrombocytopenia were ruled out.

Statistical Analysis of Retrospective Study of Unintentional Adverse Effects

Regression analysis with chi‐square and ANOVA were used in the analysis of the demographic data. RRs were calculated for the number of cases coded with an anticoagulant‐related adverse event secondary thrombocytopenia before and after the order set implementation.

Educational Efforts and Feedback

Members of the multidisciplinary team presented information on HA VTE and the VTE prevention protocol at Medical and Surgical grand rounds, teaching rounds, and noon conference, averaging 1 educational session per quarter. Feedback and education was provided to physicians and nursing staff when audits revealed that a patient had inadequate prophylaxis with reference to the protocol standard. In addition, these conversations provided on opportunity to explore reasons for nonadherence with the protocol, confusion regarding the VTE RAM, and other barriers to effective prophylaxis, thereby providing guidance for further protocol revision and educational efforts. We adjusted the order set based on active monitoring of order set use and the audit process.

Interobserver Agreement

The VTE RAM interobserver agreement was assessed on 150 patients with 5 observers as described above. The kappa score for the VTE risk level was 0.81. The kappa score for the judgment of whether the patient was on adequate prophylaxis or not was 0.90.

Impact on Percent of Patients with Adequate Prophylaxis (Longitudinal Audits)

Audits of randomly sampled inpatients occurred longitudinally throughout the study period as described above. Based on the intervention, the percent of patients on adequate prophylaxis improved significantly (P < 0.001) by each calendar year (see Table 3), from a baseline of 58% in 2005 to 78% in 2006 (unadjusted relative benefit = 1.35; 95% confidence interval [CI] = 1.28‐1.43), and 93% in 2007 (unadjusted relative benefit = 1.61; 95% CI = 1.52, 1.69). The improvement seen was more marked in the moderate VTE risk patients when compared to the high VTE risk patients. The percent of audited VTE prophylaxis improved from 53% in calendar year (CY) 2005 to 93% in 2007 (unadjusted relative benefit = 1.75; 95% CI = 1.70‐1.81) in the moderate VTE risk group, while the high VTE risk group improved from 83% to 92% in the same time period (unadjusted relative benefit = 1.11; 95% CI = 0.95‐1.25).

Overall, adequate VTE prophylaxis was present in over 98% of audited patients in the last 6 months of 2007, and this high rate has been sustained throughout 2008. Age, ethnicity, and gender were not associated with differential rates of adequate VTE prophylaxis.

Figure 1 is a timeline of interventions and the impact on the prevalence of adequate VTE prophylaxis. The first 7 to 8 months represent the baseline rate 50% to 55% of VTE prophylaxis. In this baseline period, the improvement team was meeting, but had not yet begun meeting with the large variety of medical and surgical service leaders. Consensus‐building sessions with these leaders in the latter part of 2005 through mid‐2006 correlated with improvement in adequate VTE prophylaxis rates to near 70%. The consensus‐building sessions also prepared these varied services for a go live date of the modular order set that was incorporated into all admit and transfer order sets, often replacing preexisting orders referring to VTE prevention measures. The order set resulted in an improvement to 80% adequate prophylaxis, with the incremental improvement occurring virtually overnight with the go live date at the onset of quarter 2 (Q2) of 2006. Monitoring of the order set use confirmed that it was easy and efficient to use, but also revealed that physicians were at times classifying patients as low VTE risk inaccurately, when they possessed qualities that actually qualified them for moderate risk status by our protocol. We therefore inserted a secondary CPOE screen when patients were categorized as low VTE risk, asking the physician to deny or confirm that the patient had no risk factors that qualified them for moderate risk status. This confirmation screen essentially acted as a reminder to the physician to ask Are you sure this patient does not need VTE prophylaxis? This minor modification of the CPOE order set improved adequate VTE prophylaxis rates to 90%. Finally, we asked nurses to evaluate patients who were not on therapeutic or prophylactic doses of anticoagulants. Patients with VTE risk factors but no obvious contraindications generated a note from the nurse to the doctor, prompting the doctor to reassess VTE risk and potential contraindications. This simple intervention raised the percent of audited patients on adequate VTE prophylaxis to 98% in the last 6 months of 2007.

Figure 1

Description of Prospectively Identified VTE

We identified 748 cases of VTE among patients admitted to the medical center over the 36‐month study period; 387 (52%) were community‐acquired VTE. There were 361 HA cases (48% of total cases) over the same time period. There was no difference in age, gender, or BMI between the community‐acquired and hospital‐related VTE.

Of the 361 HA cases, 199 (55%) occurred on Surgical Services and 162 (45%) occurred on Medical Services; 58 (16%) unique patients had pulmonary emboli, while 303 (84%) patients experienced only DVT. Remarkably, almost one‐third of the DVT occurred in the upper extremities (108 upper extremities, 240 lower extremities), and most (80%) of the upper‐extremity DVT were associated with central venous catheters.

Of 361 HA VTE cases, 292 (81%) occurred in those in the moderate VTE risk category, 69 HA VTE cases occurred in high‐risk category patients (19%), and no VTE occurred in patients in the low‐risk category.

Improvement in HA VTE

HA VTE were identified and each case analyzed on an ongoing basis over the entire 3 year study period, as described above. Table 4 depicts a comparison of HA VTE on a year‐to‐year basis and the impact of the VTE prevention protocol on the incidence of HA VTE. In 2007 (the first full CY after the implementation of the order set) there was a 39% relative risk reduction (RRR) in the risk of experiencing an HA VTE. The reduction in the risk of preventable HA VTE was even more marked (RRR = 86%; 7 preventable VTE in 2007, compared to 44 in baseline year of 2005; RR = 0.14; 95% CI = 0.06‐0.31).

Retrospective Analysis of Impact on HIT and Bleeding

There were no statistically significant differences in the number of cases coded for an anticoagulant‐related bleed or secondary thrombocytopenia (Table 5). Chart review revealed there were 2 cases of minor bleeding attributable to pharmacologic VTE prophylaxis before the order set implementation. There were no cases after implementation. No cases of HIT attributable to pharmacologic VTE prophylaxis were identified in either study period, with all cases being attributed to therapeutic anticoagulation.

Interventions

We introduced several interventions and educational efforts throughout the course of our improvement. The 2 key interventions were as follows:

• Structured subcutaneous insulin order sets (November, 2003).

• An insulin management algorithm, described below (May 2005).

Key Intervention #1: Structured Subcutaneous Insulin Order Set Implementation

In November 2003, we introduced a paper‐based structured subcutaneous insulin order set. This order set encouraged the use of scheduled basal and nutritional insulin, provided guidance for monitoring glucose levels, and for insulin dosing. A hypoglycemia protocol and a standardized correction insulin table were embedded in the order set. This set was similar to examples of structured insulin ordering subsequently presented in the literature.9 In a parallel effort, the University of California, San Diego Medical Center (UCSDMC) was developing a computer physician order entry (CPOE) module for our comprehensive clinical information system, Invision (Siemens Medical Systems, Malvern, PA), that heretofore had primarily focused on result review, patient schedule management, and nursing documentation. In anticipation of CPOE and for the purpose of standardization, we removed outdated sliding scale insulin regimens from a variety preexisting order sets and inserted references to the standardized subcutaneous insulin order set in their stead. The medication administration record (MAR) was changed to reflect the basal/nutritional/correction insulin terminology. It became more difficult to order a stand‐alone insulin sliding scale even before CPOE versions became available. The standardized order set was the only preprinted correction scale insulin order available, and ordering physicians have to specifically opt out of basal and nutritional insulin choices to order sliding scale only regimens. Verbal orders for correction dose scales were deemed unacceptable by medical staff committees. Correctional insulin doses could be ordered as a 1‐time order, but the pharmacy rejected ongoing insulin orders that were not entered on the structured form.

We introduced our first standardized CPOE subcutaneous insulin order set in January 2004 at the smaller of our 2 campuses, and subsequently completed full deployment across both campuses in all adult medical‐surgical care areas by September 2004.

The CPOE version, like the paper version that immediately preceded it, encouraged the use of basal/bolus insulin regimens, promoted the terms basal, nutritional or premeal, and adjustment dose insulin in the order sets and the medication administration record, and was mandatory for providers wishing to order anything but a 1‐time order of insulin. Figure 1 depicts a screen shot of the CPOE version. Similar to the paper version, the ordering physician had to specifically opt out of ordering scheduled premeal and basal insulin to order a sliding scale only regimen. The first screen also ensured that appropriate POC glucose monitoring was ordered and endorsed a standing hypoglycemia protocol order. The CPOE version had only a few additional features not possible on paper. Obvious benefits included elimination of unapproved abbreviations and handwriting errors. Nutritional and correction insulin types were forced to be identical. Fundamentally, however, both the paper and online structured ordering experiences had the same degree of control over provider ordering patterns, and there was no increment in guidance for choosing insulin regimens, hence their combined analysis as structured orders.

Figure 1

Key Intervention #2: Insulin Management Algorithm

The structured insulin order set had many advantages, but also had many limitations. Guidance for preferred insulin regimens for patients in different nutritional situations was not inherent in the order set, and all basal and nutritional insulin options were offered as equally acceptable choices. The order set gave very general guidance for insulin dosing, but did not calculate insulin doses or assist in the apportionment of insulin between basal and nutritional components, and guidance for setting a glycemic target or adjusting insulin was lacking.

Recognizing these limitations, we devised an insulin management algorithm to provide guidance incremental to that offered in the order set. In April 2005, 3 hospitalists piloted a paper‐based insulin management algorithm (Figure 2, front; Figure 3, reverse) on their teaching services. This 1‐page algorithm provided guidance on insulin dosing and monitoring, and provided institutionally preferred insulin regimens for patients in different nutritional situations. As an example, of the several acceptable subcutaneous insulin regimens that an eating patient might use in the inpatient setting, we advocated the use of 1 preferred regimen (a relatively peakless, long‐acting basal insulin once a day, along with a rapid acting analog nutritional insulin with each meal). We introduced the concept of a ward glycemic target, provided prompts for diabetes education, and generally recommended discontinuation of oral hypoglycemic agents in the inpatient setting. The hospitalists were introduced to the concepts and the algorithm via 1 of the authors (G.M.) in a 1‐hour session. The algorithm was introduced on each teaching team during routine teaching rounds with a slide set (approximately 15 slides) that outlined the basic principles of insulin dosing, and gave example cases which modeled the proper use of the algorithm. The principles were reinforced on daily patient work rounds as they were applied on inpatients with hyperglycemia. The pilot results on 25 patients, compared to 250 historical control patients, were very promising, with markedly improved glycemic control and no increase in hypoglycemia. We therefore sought to spread the use of the algorithm. In May 2005 the insulin management algorithm and teaching slide set were promoted on all 7 hospitalist‐run services, and the results of the pilot and concepts of the algorithm were shared with a variety of house staff and service leaders in approximately a dozen sessions: educational grand rounds, assorted noon lectures, and subsequently, at new intern orientations. Easy access to the algorithm was assured by providing a link to the file within the CPOE insulin order set.

Figure 2

Figure 3

Measures and Analyses

Insulin Use Patterns

Figure 4 demonstrates the dramatic improvement that took place with the introduction of the structured order set. In the 6 months preceding the introduction of the structured insulin order set (May‐October 2003) 72% of 477 sampled patients with insulin orders were on sliding scale‐only insulin regimens (with no basal insulin), compared to just 26% of 499 patients sampled in the March to August 2004 time period subsequent to order set implementation (P < .0001, chi square statistic). Intermittent monthly checks on insulin use patterns reveal this change has been sustained.

Figure 4

Glycemic Control

A total of 9314 patients with 44,232 monitored patient‐days and over 120,000 POC glucose values were analyzed to assess glycemic control, which was improved with structured insulin orders and improved incrementally with the introduction of the insulin management algorithm.

The percent of patient‐days that were uncontrolled, defined as a monitored day with a mean glucose of 180 mg/dL, was reduced over the 3 time periods (37.8% versus 33.9% versus 30.1%, P < 0.005, Pearson chi square statistic), representing a 21% RR reduction of uncontrolled patient‐days from TP1 versus TP3. Table 2 shows the summary results for glycemic control, including the RR and CIs between the 3 time periods.

In a similar fashion, the percent of patients with uncontrolled patient‐stays (day‐weighted mean glucose 180 mg/dL) was also reduced over the 3 time periods (41.5% versus 37.6% versus 34.2%, P < 0.05, Pearson chi square statistic, with an RR reduction of 16% for TP3:TP1). Figure 5 depicts a statistical process control chart of the percent of patients experiencing uncontrolled patient‐stays over time, and is more effective in displaying the temporal relationship of the interventions with the improved results.

Figure 5

Uncontrolled hyperglycemic days and stays were reduced incrementally from TP3 versus TP2, reflecting the added benefit of the insulin management algorithm, compared to the benefit enjoyed with the structured order set alone.

When the analyses were repeated after excluding patients with fewer than 8 POC glucose readings (Table 3), the findings were similar, but as predicted, the effect was slightly more pronounced, with a 23% relative reduction in uncontrolled patient‐days and a 27% reduction in uncontrolled patient‐stays of TP3 versus TP1.

Hypoglycemia

Table 4 summarizes the results for hypoglycemia and severe hypoglycemia in the study population, and Table 5 summarizes the secondary analyses of hypoglycemia in the subset with at least 8 POC glucose readings.

Strengths and Limitations

Our study has several strengths. The study is large and incorporates all patients with diabetes or hyperglycemia captured by POC glucose testing, and the observation period is long enough that bias from merely being observed is not a factor. We used metrics for glycemic control, hypoglycemia, and insulin use patterns that are of high quality and are generally in line with the Society of Hospital Medicine (SHM) Glycemic Control Task force recommendations,12, 13 and examined data by both patient‐stay and patient‐day.

The increased use of anticipatory physiologic subcutaneous insulin regimens, and the subsequent decline in the use of sliding scale insulin, is the most likely mechanism for improvement. The improvements seen are fairly dramatic for an institution in absolute terms, because inpatient hyperglycemia and hypoglycemia are so common. For example, on an annualized basis for our 400‐bed medical center, these interventions prevent 124 patients from experiencing 208 hypoglycemic days.

Other institutions should be able to replicate our results. We received administrative support to create a multidisciplinary steering committee, but we did not have incremental resources to create a dedicated team for insulin management, mandated endocrinology comanagement or consultations, or manual data collection. In fact, we had only 1 diabetes educator for 400 adult beds at 2 sites, and were relatively underresourced in this area by community standards. There was some time and expense in creating the glycemic control reports, but all of the glucose data collected were part of normal care, and the data retrieval became automated.

The main limitation of this study lies in the observational study design. There were multiple interventions in addition to structured insulin orders and the insulin management algorithm, and these educational and organizational changes undoubtedly also contributed to the overall success of our program. Since we did not perform a randomized controlled trial, the reader might reasonably question if the structured order sets and insulin management algorithm were actually the cause of the improvement seen, as opposed to these ancillary efforts or secular change. However, there are several factors that make this unlikely. First, the study population was well‐defined, having diabetes or documented hyperglycemia in all 3 time periods. Second, the demographics remained constant or actually worked against improvement trends, since the markers of patient acuity suggest increased patient acuity over the observation period. Third, the temporal relationship of the improvement to the introduction of our key interventions, as viewed on statistical process control charts shown in Figure 5, strongly suggest a causal relationship. This temporal relationship was consistently observed no matter how we chose to define uncontrolled hyperglycemia, and was also seen on hypoglycemia control charts. We view the ancillary interventions (such as educational efforts) as necessary, but not sufficient, in and of themselves, to effect major improvement.

We did not analyze the impact of the improved glycemic control on patient outcomes. In the absence of a randomized controlled trial design, controlling for the various confounders is a challenging task. Also, it is likely that not all hypoglycemic events were attributable to inpatient glycemic control regimens, though the secondary analysis probably eliminated many hypoglycemia admissions.

Lessons Learned: Implications from our study

We agree with the American Association of Clinical Endocrinologists (AACE)/American Diabetes Association (ADA)2 and the SHM Glycemic Control Task Force12 about the essential elements needed for successful implementation of inpatient glycemic control programs:

• An appropriate level of administrative support.

• Formation of a multidisciplinary steering committee to drive the development of initiatives, empowered to enact changes.

• Assessment of current processes, quality of care, and barriers to practice change.

• Development and implementation of interventions, including standardized order sets, protocols, policies, and algorithms with associated educational programs.

• Metrics for evaluation of glycemic control, hypoglycemia, insulin use patterns, and other aspects of care.

Metrics to follow hypoglycemia are extremely important. The voluntary reporting on insulin‐induced hypoglycemia fluctuated widely over the course of our project. These fluctuations did not correlate well with the more objective and accurate measures we followed, and this objective data was very helpful in reducing the fear of hypoglycemia, and spreading the wider use of basal bolus insulin regimens. We strongly recommend that improvement teams formulate and follow measures of glycemic control, hypoglycemia, and insulin use, similar to those outlined in the SHM Glycemic Control Improvement Guide12 and the SHM Glycemic Control Task Force summary on glucometrics.13

Although we introduced our structured insulin order set first, with a long lag time until we introduced the insulin management algorithm, we advocate a different approach for institutions grappling with these issues. This approach is well‐described by the SHM Glycemic Control Task Force.14 An insulin management algorithm should be crafted first, integrating guidance for insulin dosing, preferred insulin regimens for different nutritional situations, a glycemic target, insulin dosing adjustment, glucose monitoring, and prompts for ordering a glycosylated hemoglobin (A1c) level. Next, the order set and the supporting educational programs should integrate this guidance as much as possible, making the key guidance available at the point of patient care.

This guidance was available in our algorithm but was not inherent in the structured insulin orders described in this report, and all basal and nutritional insulin options were offered as equally acceptable choices. This version did not calculate insulin doses or assist in the apportionment of insulin between basal and nutritional components. Only a single adjustment dose scale was offered, leaving appropriate modifications up to the end user, and from a usability standpoint, our CPOE insulin orders lacked dynamic flexibility (revising a single insulin required discontinuing all prior orders and reentering all orders). These limitations have subsequently been addressed with Version 2 of our CPOE insulin orders, and the details will soon be available in the literature.15

We are now exploring further improvement with concurrent identification and intervention of hyperglycemic patients that are not on physiologic insulin regimens or not meeting glycemic targets, and implementing protocols addressing the transition from infusion insulin.

Interventions

We introduced several interventions and educational efforts throughout the course of our improvement. The 2 key interventions were as follows:

• Structured subcutaneous insulin order sets (November, 2003).

• An insulin management algorithm, described below (May 2005).

Key Intervention #1: Structured Subcutaneous Insulin Order Set Implementation

In November 2003, we introduced a paper‐based structured subcutaneous insulin order set. This order set encouraged the use of scheduled basal and nutritional insulin, provided guidance for monitoring glucose levels, and for insulin dosing. A hypoglycemia protocol and a standardized correction insulin table were embedded in the order set. This set was similar to examples of structured insulin ordering subsequently presented in the literature.9 In a parallel effort, the University of California, San Diego Medical Center (UCSDMC) was developing a computer physician order entry (CPOE) module for our comprehensive clinical information system, Invision (Siemens Medical Systems, Malvern, PA), that heretofore had primarily focused on result review, patient schedule management, and nursing documentation. In anticipation of CPOE and for the purpose of standardization, we removed outdated sliding scale insulin regimens from a variety preexisting order sets and inserted references to the standardized subcutaneous insulin order set in their stead. The medication administration record (MAR) was changed to reflect the basal/nutritional/correction insulin terminology. It became more difficult to order a stand‐alone insulin sliding scale even before CPOE versions became available. The standardized order set was the only preprinted correction scale insulin order available, and ordering physicians have to specifically opt out of basal and nutritional insulin choices to order sliding scale only regimens. Verbal orders for correction dose scales were deemed unacceptable by medical staff committees. Correctional insulin doses could be ordered as a 1‐time order, but the pharmacy rejected ongoing insulin orders that were not entered on the structured form.

We introduced our first standardized CPOE subcutaneous insulin order set in January 2004 at the smaller of our 2 campuses, and subsequently completed full deployment across both campuses in all adult medical‐surgical care areas by September 2004.

The CPOE version, like the paper version that immediately preceded it, encouraged the use of basal/bolus insulin regimens, promoted the terms basal, nutritional or premeal, and adjustment dose insulin in the order sets and the medication administration record, and was mandatory for providers wishing to order anything but a 1‐time order of insulin. Figure 1 depicts a screen shot of the CPOE version. Similar to the paper version, the ordering physician had to specifically opt out of ordering scheduled premeal and basal insulin to order a sliding scale only regimen. The first screen also ensured that appropriate POC glucose monitoring was ordered and endorsed a standing hypoglycemia protocol order. The CPOE version had only a few additional features not possible on paper. Obvious benefits included elimination of unapproved abbreviations and handwriting errors. Nutritional and correction insulin types were forced to be identical. Fundamentally, however, both the paper and online structured ordering experiences had the same degree of control over provider ordering patterns, and there was no increment in guidance for choosing insulin regimens, hence their combined analysis as structured orders.

Figure 1

Key Intervention #2: Insulin Management Algorithm

The structured insulin order set had many advantages, but also had many limitations. Guidance for preferred insulin regimens for patients in different nutritional situations was not inherent in the order set, and all basal and nutritional insulin options were offered as equally acceptable choices. The order set gave very general guidance for insulin dosing, but did not calculate insulin doses or assist in the apportionment of insulin between basal and nutritional components, and guidance for setting a glycemic target or adjusting insulin was lacking.

Recognizing these limitations, we devised an insulin management algorithm to provide guidance incremental to that offered in the order set. In April 2005, 3 hospitalists piloted a paper‐based insulin management algorithm (Figure 2, front; Figure 3, reverse) on their teaching services. This 1‐page algorithm provided guidance on insulin dosing and monitoring, and provided institutionally preferred insulin regimens for patients in different nutritional situations. As an example, of the several acceptable subcutaneous insulin regimens that an eating patient might use in the inpatient setting, we advocated the use of 1 preferred regimen (a relatively peakless, long‐acting basal insulin once a day, along with a rapid acting analog nutritional insulin with each meal). We introduced the concept of a ward glycemic target, provided prompts for diabetes education, and generally recommended discontinuation of oral hypoglycemic agents in the inpatient setting. The hospitalists were introduced to the concepts and the algorithm via 1 of the authors (G.M.) in a 1‐hour session. The algorithm was introduced on each teaching team during routine teaching rounds with a slide set (approximately 15 slides) that outlined the basic principles of insulin dosing, and gave example cases which modeled the proper use of the algorithm. The principles were reinforced on daily patient work rounds as they were applied on inpatients with hyperglycemia. The pilot results on 25 patients, compared to 250 historical control patients, were very promising, with markedly improved glycemic control and no increase in hypoglycemia. We therefore sought to spread the use of the algorithm. In May 2005 the insulin management algorithm and teaching slide set were promoted on all 7 hospitalist‐run services, and the results of the pilot and concepts of the algorithm were shared with a variety of house staff and service leaders in approximately a dozen sessions: educational grand rounds, assorted noon lectures, and subsequently, at new intern orientations. Easy access to the algorithm was assured by providing a link to the file within the CPOE insulin order set.

Figure 2

Figure 3

Measures and Analyses

Insulin Use Patterns

Figure 4 demonstrates the dramatic improvement that took place with the introduction of the structured order set. In the 6 months preceding the introduction of the structured insulin order set (May‐October 2003) 72% of 477 sampled patients with insulin orders were on sliding scale‐only insulin regimens (with no basal insulin), compared to just 26% of 499 patients sampled in the March to August 2004 time period subsequent to order set implementation (P < .0001, chi square statistic). Intermittent monthly checks on insulin use patterns reveal this change has been sustained.

Figure 4

Glycemic Control

A total of 9314 patients with 44,232 monitored patient‐days and over 120,000 POC glucose values were analyzed to assess glycemic control, which was improved with structured insulin orders and improved incrementally with the introduction of the insulin management algorithm.

The percent of patient‐days that were uncontrolled, defined as a monitored day with a mean glucose of 180 mg/dL, was reduced over the 3 time periods (37.8% versus 33.9% versus 30.1%, P < 0.005, Pearson chi square statistic), representing a 21% RR reduction of uncontrolled patient‐days from TP1 versus TP3. Table 2 shows the summary results for glycemic control, including the RR and CIs between the 3 time periods.

In a similar fashion, the percent of patients with uncontrolled patient‐stays (day‐weighted mean glucose 180 mg/dL) was also reduced over the 3 time periods (41.5% versus 37.6% versus 34.2%, P < 0.05, Pearson chi square statistic, with an RR reduction of 16% for TP3:TP1). Figure 5 depicts a statistical process control chart of the percent of patients experiencing uncontrolled patient‐stays over time, and is more effective in displaying the temporal relationship of the interventions with the improved results.

Figure 5

Uncontrolled hyperglycemic days and stays were reduced incrementally from TP3 versus TP2, reflecting the added benefit of the insulin management algorithm, compared to the benefit enjoyed with the structured order set alone.

When the analyses were repeated after excluding patients with fewer than 8 POC glucose readings (Table 3), the findings were similar, but as predicted, the effect was slightly more pronounced, with a 23% relative reduction in uncontrolled patient‐days and a 27% reduction in uncontrolled patient‐stays of TP3 versus TP1.

Hypoglycemia

Table 4 summarizes the results for hypoglycemia and severe hypoglycemia in the study population, and Table 5 summarizes the secondary analyses of hypoglycemia in the subset with at least 8 POC glucose readings.

Strengths and Limitations

Our study has several strengths. The study is large and incorporates all patients with diabetes or hyperglycemia captured by POC glucose testing, and the observation period is long enough that bias from merely being observed is not a factor. We used metrics for glycemic control, hypoglycemia, and insulin use patterns that are of high quality and are generally in line with the Society of Hospital Medicine (SHM) Glycemic Control Task force recommendations,12, 13 and examined data by both patient‐stay and patient‐day.

The increased use of anticipatory physiologic subcutaneous insulin regimens, and the subsequent decline in the use of sliding scale insulin, is the most likely mechanism for improvement. The improvements seen are fairly dramatic for an institution in absolute terms, because inpatient hyperglycemia and hypoglycemia are so common. For example, on an annualized basis for our 400‐bed medical center, these interventions prevent 124 patients from experiencing 208 hypoglycemic days.

Other institutions should be able to replicate our results. We received administrative support to create a multidisciplinary steering committee, but we did not have incremental resources to create a dedicated team for insulin management, mandated endocrinology comanagement or consultations, or manual data collection. In fact, we had only 1 diabetes educator for 400 adult beds at 2 sites, and were relatively underresourced in this area by community standards. There was some time and expense in creating the glycemic control reports, but all of the glucose data collected were part of normal care, and the data retrieval became automated.

The main limitation of this study lies in the observational study design. There were multiple interventions in addition to structured insulin orders and the insulin management algorithm, and these educational and organizational changes undoubtedly also contributed to the overall success of our program. Since we did not perform a randomized controlled trial, the reader might reasonably question if the structured order sets and insulin management algorithm were actually the cause of the improvement seen, as opposed to these ancillary efforts or secular change. However, there are several factors that make this unlikely. First, the study population was well‐defined, having diabetes or documented hyperglycemia in all 3 time periods. Second, the demographics remained constant or actually worked against improvement trends, since the markers of patient acuity suggest increased patient acuity over the observation period. Third, the temporal relationship of the improvement to the introduction of our key interventions, as viewed on statistical process control charts shown in Figure 5, strongly suggest a causal relationship. This temporal relationship was consistently observed no matter how we chose to define uncontrolled hyperglycemia, and was also seen on hypoglycemia control charts. We view the ancillary interventions (such as educational efforts) as necessary, but not sufficient, in and of themselves, to effect major improvement.

We did not analyze the impact of the improved glycemic control on patient outcomes. In the absence of a randomized controlled trial design, controlling for the various confounders is a challenging task. Also, it is likely that not all hypoglycemic events were attributable to inpatient glycemic control regimens, though the secondary analysis probably eliminated many hypoglycemia admissions.

Lessons Learned: Implications from our study

We agree with the American Association of Clinical Endocrinologists (AACE)/American Diabetes Association (ADA)2 and the SHM Glycemic Control Task Force12 about the essential elements needed for successful implementation of inpatient glycemic control programs:

• An appropriate level of administrative support.

• Formation of a multidisciplinary steering committee to drive the development of initiatives, empowered to enact changes.

• Assessment of current processes, quality of care, and barriers to practice change.

• Development and implementation of interventions, including standardized order sets, protocols, policies, and algorithms with associated educational programs.

• Metrics for evaluation of glycemic control, hypoglycemia, insulin use patterns, and other aspects of care.

Metrics to follow hypoglycemia are extremely important. The voluntary reporting on insulin‐induced hypoglycemia fluctuated widely over the course of our project. These fluctuations did not correlate well with the more objective and accurate measures we followed, and this objective data was very helpful in reducing the fear of hypoglycemia, and spreading the wider use of basal bolus insulin regimens. We strongly recommend that improvement teams formulate and follow measures of glycemic control, hypoglycemia, and insulin use, similar to those outlined in the SHM Glycemic Control Improvement Guide12 and the SHM Glycemic Control Task Force summary on glucometrics.13

Although we introduced our structured insulin order set first, with a long lag time until we introduced the insulin management algorithm, we advocate a different approach for institutions grappling with these issues. This approach is well‐described by the SHM Glycemic Control Task Force.14 An insulin management algorithm should be crafted first, integrating guidance for insulin dosing, preferred insulin regimens for different nutritional situations, a glycemic target, insulin dosing adjustment, glucose monitoring, and prompts for ordering a glycosylated hemoglobin (A1c) level. Next, the order set and the supporting educational programs should integrate this guidance as much as possible, making the key guidance available at the point of patient care.

This guidance was available in our algorithm but was not inherent in the structured insulin orders described in this report, and all basal and nutritional insulin options were offered as equally acceptable choices. This version did not calculate insulin doses or assist in the apportionment of insulin between basal and nutritional components. Only a single adjustment dose scale was offered, leaving appropriate modifications up to the end user, and from a usability standpoint, our CPOE insulin orders lacked dynamic flexibility (revising a single insulin required discontinuing all prior orders and reentering all orders). These limitations have subsequently been addressed with Version 2 of our CPOE insulin orders, and the details will soon be available in the literature.15

We are now exploring further improvement with concurrent identification and intervention of hyperglycemic patients that are not on physiologic insulin regimens or not meeting glycemic targets, and implementing protocols addressing the transition from infusion insulin.

Interventions

We introduced several interventions and educational efforts throughout the course of our improvement. The 2 key interventions were as follows:

• Structured subcutaneous insulin order sets (November, 2003).

• An insulin management algorithm, described below (May 2005).

Key Intervention #1: Structured Subcutaneous Insulin Order Set Implementation

In November 2003, we introduced a paper‐based structured subcutaneous insulin order set. This order set encouraged the use of scheduled basal and nutritional insulin, provided guidance for monitoring glucose levels, and for insulin dosing. A hypoglycemia protocol and a standardized correction insulin table were embedded in the order set. This set was similar to examples of structured insulin ordering subsequently presented in the literature.9 In a parallel effort, the University of California, San Diego Medical Center (UCSDMC) was developing a computer physician order entry (CPOE) module for our comprehensive clinical information system, Invision (Siemens Medical Systems, Malvern, PA), that heretofore had primarily focused on result review, patient schedule management, and nursing documentation. In anticipation of CPOE and for the purpose of standardization, we removed outdated sliding scale insulin regimens from a variety preexisting order sets and inserted references to the standardized subcutaneous insulin order set in their stead. The medication administration record (MAR) was changed to reflect the basal/nutritional/correction insulin terminology. It became more difficult to order a stand‐alone insulin sliding scale even before CPOE versions became available. The standardized order set was the only preprinted correction scale insulin order available, and ordering physicians have to specifically opt out of basal and nutritional insulin choices to order sliding scale only regimens. Verbal orders for correction dose scales were deemed unacceptable by medical staff committees. Correctional insulin doses could be ordered as a 1‐time order, but the pharmacy rejected ongoing insulin orders that were not entered on the structured form.

We introduced our first standardized CPOE subcutaneous insulin order set in January 2004 at the smaller of our 2 campuses, and subsequently completed full deployment across both campuses in all adult medical‐surgical care areas by September 2004.

The CPOE version, like the paper version that immediately preceded it, encouraged the use of basal/bolus insulin regimens, promoted the terms basal, nutritional or premeal, and adjustment dose insulin in the order sets and the medication administration record, and was mandatory for providers wishing to order anything but a 1‐time order of insulin. Figure 1 depicts a screen shot of the CPOE version. Similar to the paper version, the ordering physician had to specifically opt out of ordering scheduled premeal and basal insulin to order a sliding scale only regimen. The first screen also ensured that appropriate POC glucose monitoring was ordered and endorsed a standing hypoglycemia protocol order. The CPOE version had only a few additional features not possible on paper. Obvious benefits included elimination of unapproved abbreviations and handwriting errors. Nutritional and correction insulin types were forced to be identical. Fundamentally, however, both the paper and online structured ordering experiences had the same degree of control over provider ordering patterns, and there was no increment in guidance for choosing insulin regimens, hence their combined analysis as structured orders.

Figure 1

Key Intervention #2: Insulin Management Algorithm

The structured insulin order set had many advantages, but also had many limitations. Guidance for preferred insulin regimens for patients in different nutritional situations was not inherent in the order set, and all basal and nutritional insulin options were offered as equally acceptable choices. The order set gave very general guidance for insulin dosing, but did not calculate insulin doses or assist in the apportionment of insulin between basal and nutritional components, and guidance for setting a glycemic target or adjusting insulin was lacking.

Recognizing these limitations, we devised an insulin management algorithm to provide guidance incremental to that offered in the order set. In April 2005, 3 hospitalists piloted a paper‐based insulin management algorithm (Figure 2, front; Figure 3, reverse) on their teaching services. This 1‐page algorithm provided guidance on insulin dosing and monitoring, and provided institutionally preferred insulin regimens for patients in different nutritional situations. As an example, of the several acceptable subcutaneous insulin regimens that an eating patient might use in the inpatient setting, we advocated the use of 1 preferred regimen (a relatively peakless, long‐acting basal insulin once a day, along with a rapid acting analog nutritional insulin with each meal). We introduced the concept of a ward glycemic target, provided prompts for diabetes education, and generally recommended discontinuation of oral hypoglycemic agents in the inpatient setting. The hospitalists were introduced to the concepts and the algorithm via 1 of the authors (G.M.) in a 1‐hour session. The algorithm was introduced on each teaching team during routine teaching rounds with a slide set (approximately 15 slides) that outlined the basic principles of insulin dosing, and gave example cases which modeled the proper use of the algorithm. The principles were reinforced on daily patient work rounds as they were applied on inpatients with hyperglycemia. The pilot results on 25 patients, compared to 250 historical control patients, were very promising, with markedly improved glycemic control and no increase in hypoglycemia. We therefore sought to spread the use of the algorithm. In May 2005 the insulin management algorithm and teaching slide set were promoted on all 7 hospitalist‐run services, and the results of the pilot and concepts of the algorithm were shared with a variety of house staff and service leaders in approximately a dozen sessions: educational grand rounds, assorted noon lectures, and subsequently, at new intern orientations. Easy access to the algorithm was assured by providing a link to the file within the CPOE insulin order set.

Figure 2

Figure 3

Measures and Analyses

Insulin Use Patterns

Figure 4 demonstrates the dramatic improvement that took place with the introduction of the structured order set. In the 6 months preceding the introduction of the structured insulin order set (May‐October 2003) 72% of 477 sampled patients with insulin orders were on sliding scale‐only insulin regimens (with no basal insulin), compared to just 26% of 499 patients sampled in the March to August 2004 time period subsequent to order set implementation (P < .0001, chi square statistic). Intermittent monthly checks on insulin use patterns reveal this change has been sustained.

Figure 4

Glycemic Control

A total of 9314 patients with 44,232 monitored patient‐days and over 120,000 POC glucose values were analyzed to assess glycemic control, which was improved with structured insulin orders and improved incrementally with the introduction of the insulin management algorithm.

The percent of patient‐days that were uncontrolled, defined as a monitored day with a mean glucose of 180 mg/dL, was reduced over the 3 time periods (37.8% versus 33.9% versus 30.1%, P < 0.005, Pearson chi square statistic), representing a 21% RR reduction of uncontrolled patient‐days from TP1 versus TP3. Table 2 shows the summary results for glycemic control, including the RR and CIs between the 3 time periods.

In a similar fashion, the percent of patients with uncontrolled patient‐stays (day‐weighted mean glucose 180 mg/dL) was also reduced over the 3 time periods (41.5% versus 37.6% versus 34.2%, P < 0.05, Pearson chi square statistic, with an RR reduction of 16% for TP3:TP1). Figure 5 depicts a statistical process control chart of the percent of patients experiencing uncontrolled patient‐stays over time, and is more effective in displaying the temporal relationship of the interventions with the improved results.

Figure 5

Uncontrolled hyperglycemic days and stays were reduced incrementally from TP3 versus TP2, reflecting the added benefit of the insulin management algorithm, compared to the benefit enjoyed with the structured order set alone.

When the analyses were repeated after excluding patients with fewer than 8 POC glucose readings (Table 3), the findings were similar, but as predicted, the effect was slightly more pronounced, with a 23% relative reduction in uncontrolled patient‐days and a 27% reduction in uncontrolled patient‐stays of TP3 versus TP1.

Hypoglycemia

Table 4 summarizes the results for hypoglycemia and severe hypoglycemia in the study population, and Table 5 summarizes the secondary analyses of hypoglycemia in the subset with at least 8 POC glucose readings.

Strengths and Limitations

Our study has several strengths. The study is large and incorporates all patients with diabetes or hyperglycemia captured by POC glucose testing, and the observation period is long enough that bias from merely being observed is not a factor. We used metrics for glycemic control, hypoglycemia, and insulin use patterns that are of high quality and are generally in line with the Society of Hospital Medicine (SHM) Glycemic Control Task force recommendations,12, 13 and examined data by both patient‐stay and patient‐day.

The increased use of anticipatory physiologic subcutaneous insulin regimens, and the subsequent decline in the use of sliding scale insulin, is the most likely mechanism for improvement. The improvements seen are fairly dramatic for an institution in absolute terms, because inpatient hyperglycemia and hypoglycemia are so common. For example, on an annualized basis for our 400‐bed medical center, these interventions prevent 124 patients from experiencing 208 hypoglycemic days.

Other institutions should be able to replicate our results. We received administrative support to create a multidisciplinary steering committee, but we did not have incremental resources to create a dedicated team for insulin management, mandated endocrinology comanagement or consultations, or manual data collection. In fact, we had only 1 diabetes educator for 400 adult beds at 2 sites, and were relatively underresourced in this area by community standards. There was some time and expense in creating the glycemic control reports, but all of the glucose data collected were part of normal care, and the data retrieval became automated.

The main limitation of this study lies in the observational study design. There were multiple interventions in addition to structured insulin orders and the insulin management algorithm, and these educational and organizational changes undoubtedly also contributed to the overall success of our program. Since we did not perform a randomized controlled trial, the reader might reasonably question if the structured order sets and insulin management algorithm were actually the cause of the improvement seen, as opposed to these ancillary efforts or secular change. However, there are several factors that make this unlikely. First, the study population was well‐defined, having diabetes or documented hyperglycemia in all 3 time periods. Second, the demographics remained constant or actually worked against improvement trends, since the markers of patient acuity suggest increased patient acuity over the observation period. Third, the temporal relationship of the improvement to the introduction of our key interventions, as viewed on statistical process control charts shown in Figure 5, strongly suggest a causal relationship. This temporal relationship was consistently observed no matter how we chose to define uncontrolled hyperglycemia, and was also seen on hypoglycemia control charts. We view the ancillary interventions (such as educational efforts) as necessary, but not sufficient, in and of themselves, to effect major improvement.

We did not analyze the impact of the improved glycemic control on patient outcomes. In the absence of a randomized controlled trial design, controlling for the various confounders is a challenging task. Also, it is likely that not all hypoglycemic events were attributable to inpatient glycemic control regimens, though the secondary analysis probably eliminated many hypoglycemia admissions.

Lessons Learned: Implications from our study

We agree with the American Association of Clinical Endocrinologists (AACE)/American Diabetes Association (ADA)2 and the SHM Glycemic Control Task Force12 about the essential elements needed for successful implementation of inpatient glycemic control programs:

• An appropriate level of administrative support.

• Formation of a multidisciplinary steering committee to drive the development of initiatives, empowered to enact changes.

• Assessment of current processes, quality of care, and barriers to practice change.

• Development and implementation of interventions, including standardized order sets, protocols, policies, and algorithms with associated educational programs.

• Metrics for evaluation of glycemic control, hypoglycemia, insulin use patterns, and other aspects of care.

Metrics to follow hypoglycemia are extremely important. The voluntary reporting on insulin‐induced hypoglycemia fluctuated widely over the course of our project. These fluctuations did not correlate well with the more objective and accurate measures we followed, and this objective data was very helpful in reducing the fear of hypoglycemia, and spreading the wider use of basal bolus insulin regimens. We strongly recommend that improvement teams formulate and follow measures of glycemic control, hypoglycemia, and insulin use, similar to those outlined in the SHM Glycemic Control Improvement Guide12 and the SHM Glycemic Control Task Force summary on glucometrics.13

Although we introduced our structured insulin order set first, with a long lag time until we introduced the insulin management algorithm, we advocate a different approach for institutions grappling with these issues. This approach is well‐described by the SHM Glycemic Control Task Force.14 An insulin management algorithm should be crafted first, integrating guidance for insulin dosing, preferred insulin regimens for different nutritional situations, a glycemic target, insulin dosing adjustment, glucose monitoring, and prompts for ordering a glycosylated hemoglobin (A1c) level. Next, the order set and the supporting educational programs should integrate this guidance as much as possible, making the key guidance available at the point of patient care.

This guidance was available in our algorithm but was not inherent in the structured insulin orders described in this report, and all basal and nutritional insulin options were offered as equally acceptable choices. This version did not calculate insulin doses or assist in the apportionment of insulin between basal and nutritional components. Only a single adjustment dose scale was offered, leaving appropriate modifications up to the end user, and from a usability standpoint, our CPOE insulin orders lacked dynamic flexibility (revising a single insulin required discontinuing all prior orders and reentering all orders). These limitations have subsequently been addressed with Version 2 of our CPOE insulin orders, and the details will soon be available in the literature.15

We are now exploring further improvement with concurrent identification and intervention of hyperglycemic patients that are not on physiologic insulin regimens or not meeting glycemic targets, and implementing protocols addressing the transition from infusion insulin.