Grant M. Mussman, MD, MHSA

Children with medical complexity (CMC) have complex and chronic health conditions that often involve multiple organ systems and severely affect cognitive and physical functioning. Although the prevalence of CMC is low (1% of all children), they account for nearly one‐fifth of all pediatric admissions and one‐half of all hospital days and charges in the United States.[1] Over the last decade, CMC have had a particularly large and increasing impact in tertiary‐care children's hospitals.[1, 2] The Institute of Medicine has identified CMC as a priority population for a revised healthcare system.[3]

Medical homes, hospitals, health plans, states, federal agencies, and others are striving to reduce excessive hospital use in CMC because of its high cost.[4, 5, 6] Containing length of stay (LOS)an increasingly used indicator of the time sensitiveness and efficiency of hospital careis a common aim across these initiatives. CMC have longer hospitalizations than children without medical complexity. Speculated reasons for this are that CMC tend to have (1) higher severity of acute illnesses (eg, pneumonia, cellulitis), (2) prolonged recovery time in the hospital, and (3) higher risk of adverse events in the hospital. Moreover, hospital clinicians caring for CMC often find it difficult to determine discharge readiness, given that many CMC do not return to a completely healthy baseline.[7]

Little is known about long LOS in CMC, including which CMC have the highest risk of experiencing such stays and which stays might have the greatest opportunity to be shortened. Patient characteristics associated with prolonged length of stay have been studied extensively for many pediatric conditions (eg, asthma).[8, 9, 10, 11, 12, 13, 14] However, most of these studies excluded CMC. Therefore, the objectives of this study were to examine (1) the prevalence of long LOS in CMC, (2) patient characteristics associated with long LOS, and (3) hospital‐to‐hospital variation in prevalence of long LOS hospitalizations.

METHODS

RESULTS

Study Population

There were 954,018 hospitalizations of 217,163 CMC at 44 children's hospitals included for analysis. Forty‐seven percent of hospitalizations were for females, 49.4% for non‐Hispanic whites, and 61.1% for children with government insurance. Fifteen percent (n = 142,082) had a long LOS of 10 days. The median (IQR) LOS of hospitalizations <10 days versus 10 days were 2 (IQR, 14) and 16 days (IQR, 1226), respectively. Long LOS hospitalizations accounted for 61.1% (3.7 million) hospital days and 61.8% ($13.7 billion) of total hospitalization costs for all CMC in the cohort (Table 1).

Table 1.Demographic, Clinical, and Hospitalization Characteristics of Hospitalized Children With Medical Complexity by Length of Stay*

Characteristic	Overall (n = 954,018)	Length of Stay
		<10 Days (n = 811,936)	10 Days (n = 142,082)
NOTE: Abbreviations: IQR, interquartile range. *All comparisons were significant at the P < 0.001 level.
Age at admission, y, %
<1	14.6	12.7	25.7
14	27.1	27.9	22.4
59	20.1	21.0	14.9
1018	33.6	34.0	31.7
18+	4.6	4.4	5.4
Gender, %
Female	47.0	46.9	47.5
Race/ethnicity, %
Non‐Hispanic white	49.4	49.4	49.4
Non‐Hispanic black	23.1	23.8	19.3
Hispanic	18.2	17.8	20.4
Asian	2.0	1.9	2.3
Other	7.4	7.1	8.6
Complex chronic condition, %
Any	79.5	77.3	91.8
Technology assistance	37.1	34.1	54.2
Gastrointestinal	30.0	27.2	45.9
Neuromuscular	28.2	27.7	30.9
Cardiovascular	16.8	14.5	29.9
Respiratory	14.1	11.5	29.4
Congenital/genetic defect	17.2	16.7	20.2
Metabolic	9.9	8.9	15.4
Renal	10.1	9.5	13.8
Hematology/emmmunodeficiency	11.7	12.0	10.0
Neonatal	3.8	3.1	7.7
Transplantation	4.5	4.2	6.7
Clinical risk group, %
Chronic condition in 2 systems	68.4	71.2	53.9
Catastrophic chronic condition	31.4	28.8	46.1
Distance from hospital to home residence in miles, median [IQR]	16.2 [7.440.4]	15.8 [7.338.7]	19.1 [8.552.6]
Transferred from outside hospital (%)	6.5	5.3	13.6
Admitted for surgery, %	23.4	20.7	38.7
Use of intensive care, %	19.6	14.9	46.5
Discharge disposition, %
Home	91.2	92.9	81.4
Home healthcare	4.5	3.5	9.9
Other	2.9	2.6	4.5
Postacute care facility	1.1	0.8	3.1
Died	0.4	0.3	1.1
Payor, %
Government	61.1	60.6	63.5
Private	33.2	33.6	30.9
Other	5.7	5.7	5.7
Hospital resource use
Median length of stay [IQR]	3 [16]	2 [14]	16 [1226]
Median hospital cost [IQR]	$8,144 [$4,122$18,447]	$6,689 [$3,685$12,395]	$49,207 [$29,444$95,738]
Total hospital cost, $, billions	$22.2	$8.5	$13.7

Demographics and Clinical Characteristics of Children With and Without Long LOS

Compared with hospitalized CMC with LOS <10 days, a higher percentage of hospitalizations with LOS 10 days were CMC age <1 year (25.7% vs 12.7%, P < 0.001) and Hispanic (20.4% vs 17.8%, P < 0.001). CMC hospitalizations with a long LOS had a higher percentage of any CCC (91.8% vs 77.3%, P < 0.001); the most common CCCs were gastrointestinal (45.9%), neuromuscular (30.9%), and cardiovascular (29.9%). Hospitalizations of CMC with a long LOS had a higher percentage of a catastrophic chronic condition (46.1% vs 28.8%, P < 0.001) and technology dependence (46.1% vs 28.8%, P < 0.001) (Table 1).

Hospitalization Characteristics of Children With and Without Long LOS

Compared with hospitalizations of CMC with LOS <10 days, hospitalizations of CMC with a long LOS more often involved transfer in from another hospital at admission (13.6% vs 5.3%, P < 0.001). CMC hospital stays with a long LOS more often involved surgery (38.7% vs 20.7%, P < 0.001) and use of intensive care (46.5% vs 14.9%; P < 0.001). A higher percentage of CMC with long LOS were discharged with home health services (9.9% vs 3.5%; P < 0.001) (Table 1).

The most common admitting diagnoses and CCCs for hospitalizations of CMC with long LOS are presented in Table 2. The two most prevalent APR‐DRGs in CMC hospitalizations lasting 10 days or longer were cystic fibrosis (10.7%) and respiratory system disease with ventilator support (5.5%). The two most common chronic condition characteristics represented among long CMC hospitalizations were gastrointestinal devices (eg, gastrostomy tube) (39.7%) and heart and great vessel malformations (eg, tetralogy of Fallot) (12.8%). The 5 most common CCC subcategories, as listed in Table 2, account for nearly 100% of the patients with long LOS hospitalizations.

Table 2.Most Common Reasons for Admission and Specific Complex Chronic Conditions for Hospitalized Children With Medical Complexity Who Had Length of Stay 10 Days

NOTE: *Reason for admission identified using All‐Patient Refined Diagnosis‐Related Groups. Complex chronic conditions identified using Feudtner and colleagues set of International Classification of Diseases, 9th Revision, Clinical Modification codes. Gastrointestinal devices include gastrostomy, gastrojejunostomy, colostomy. Respiratory devices include tracheostomy, noninvasive positive pressure, ventilator.
Most common reason for admission*
Cystic fibrosis	10.7%
Respiratory system diagnosis with ventilator support 96+ hours	5.5%
Malfunction, reaction, and complication of cardiac or vascular device or procedure	2.8%
Craniotomy except for trauma	2.6%
Major small and large bowel procedures	2.3%
Most common complex chronic condition
Gastrointestinal devices	39.7%
Heart and great vessel malformations	12.8%
Cystic fibrosis	12.5%
Dysrhythmias	11.0%
Respiratory devices	10.7%

Multivariable Analysis of Characteristics Associated With Long LOS

In multivariable analysis, the highest likelihood of long LOS was experienced by children who received care in the ICU (odds ratio [OR]: 3.5, 95% confidence interval [CI]: 3.43.5), who had a respiratory CCC (OR: 2.7, 95% CI: 2.62.7), and who were transferred from another acute care hospital at admission (OR: 2.1, 95% CI: 2.0, 2.1). The likelihood of long LOS was also higher in children <1 year of age (OR: 1.2, 95% CI: 1.21.3), and Hispanic children (OR: 1.1, 95% CI 1.0‐1.10) (Table 3). Similar multivariable findings were observed in sensitivity analysis using the 75th percentile of LOS (4 days) as the model outcome.

Table 3.Multivariable Analysis of the Likelihood of Long Length of Stay 10 Days

Characteristic	Odds Ratio (95% CI) of LOS 10 Days	P Value
NOTE: Abbreviations: CI, confidence interval; LOS, length of stay.
Use of intensive care	3.5 (3.4‐3.5)	<0.001
Transfer from another acute‐care hospital	2.1 (2.0‐2.1)	<0.001
Procedure/surgery	1.8 (1.8‐1.9)	<0.001
Complex chronic condition
Respiratory	2.7 (2.6‐2.7)	<0.001
Gastrointestinal	1.8 (1.8‐1.8)	<0.001
Metabolic	1.7 (1.7‐1.7)	<0.001
Cardiovascular	1.6 (1.5‐1.6)	<0.001
Neonatal	1.5 (1.5‐1.5)	<0.001
Renal	1.4 (1.4‐1.4)	<0.001
Transplant	1.4 (1.4‐1.4)	<0.001
Hematology and immunodeficiency	1.3 (1.3‐1.3)	<0.001
Technology assistance	1.1 (1.1, 1.1)	<0.001
Neuromuscular	0.9 (0.9‐0.9)	<0.001
Congenital or genetic defect	0.8 (0.8‐0.8)	<0.001
Age at admission, y
<1	1.2 (1.2‐1.3)	<0.001
14	0.5 (0.5‐0.5)	<0.001
59	0.6 (0.6‐0.6)	<0.001
1018	0.9 (0.9‐0.9)	<0.001
18+	Reference
Male	0.9 (0.9‐0.9)	<0.001
Race/ethnicity
Non‐Hispanic black	0.9 (0.9‐0.9)	<0.001
Hispanic	1.1 (1.0‐1.1)	<0.001
Asian	1.0 (1.0‐1.1)	0.3
Other	1.1 (1.1‐1.1)	<0.001
Non‐Hispanic white	Reference
Payor
Private	0.9 (0.8 0.9)	<0.001
Other	1.0 (1.0‐1.0)	0.4
Government	Reference
Season
Spring	1.0 (1.0 1.0)	<0.001
Summer	0.9 (0.9‐0.9)	<0.001
Fall	1.0 (0.9‐1.0)	<0.001
Winter	Reference

Variation in the Prevalence of Long LOS Across Children's Hospitals

After controlling for demographic, clinical, and hospital characteristics associated with long LOS, there was significant (P < 0.001) variation in the prevalence of long LOS for CMC across children's hospitals in the cohort (range, 10.3%21.8%) (Figure 1). Twelve (27%) hospitals had a significantly (P < 0.001) higher prevalence of long LOS for their hospitalized CMC, compared to the mean. Eighteen (41%) had a significantly (P < 0.001) lower prevalence of long LOS for their hospitalized CMC. There was also significant variation across hospitals with respect to cost, with 49.7% to 73.7% of all hospital costs of CMC attributed to long LOS hospitalizations. Finally, there was indirect correlation with the prevalence of LOS across hospitals and the hospitals' 30‐day readmission rate ( = 0.3; P = 0.04). As the prevalence of long LOS increased, the readmission rate decreased.

jhm2633-fig-0001-m.png

DISCUSSION

The main findings from this study suggest that a small percentage of CMC experiencing long LOS account for the majority of hospital bed days and cost of all hospitalized CMC in children's hospitals. The likelihood of long LOS varies significantly by CMC's age, race/ethnicity, and payor as well as by type and number of chronic conditions. Among CMC with long LOS, the use of gastrointestinal devices such as gastrostomy tubes, as well as congenital heart disease, were highly prevalent. In multivariable analysis, the characteristics most strongly associated with LOS 10 days were use of the ICU, respiratory complex chronic condition, and transfer from another medical facility at admission. After adjusting for these factors, there was significant variation in the prevalence of LOS 10 days for CMC across children's hospitals.

Although it is well known that CMC as a whole have a major impact on resource use in children's hospitals, this study reveals that 15% of hospitalizations of CMC account for 62% of all hospital costs of CMC. That is, a small fraction of hospitalizations of CMC is largely responsible for the significant financial impact of hospital resource use. To date, most clinical efforts and policies striving to reduce hospital use in CMC have focused on avoiding readmissions or index hospital admissions entirely, rather than improving the efficiency of hospital care after admission occurs.[23, 24, 25, 26] In the adult population, the impact of long LOS on hospital costs has been recognized, and several Medicare incentive programs have focused on in‐hospital timeliness and efficiency. As a result, LOS in Medicare beneficiaries has decreased dramatically over the past 2 decades.[27, 28, 29, 30] Optimizing the efficiency of hospital care for CMC may be an important goal to pursue, especially with precedent set in the adult literature.

Perhaps the substantial variation across hospitals in the prevalence of long LOS in CMC indicates opportunity to improve the efficiency of their inpatient care. This variation was not due to differences across hospitals' case mix of CMC. Further investigation is needed to determine how much of it is due to differences in quality of care. Clinical practice guidelines for hospital treatment of common illnesses usually exclude CMC. In our clinical experience across 9 children's hospitals, we have experienced varying approaches to setting discharge goals (ie, parameters on how healthy the child needs to be to ensure a successful hospital discharge) for CMC.[31] When the goals are absent or not clearly articulated, they can contribute to a prolonged hospitalization. Some families of CMC report significant issues when working with pediatric hospital staff to assess their child's discharge readiness.[7, 32, 33] In addition, there is significant variation across states and regions in access to and quality of post‐discharge health services (eg, home nursing, postacute care, durable medical equipment).[34, 35] In some areas, many CMC are not actively involved with their primary care physician.[5] These issues might also influence the ability of some children's hospitals to efficiently discharge CMC to a safe and supportive post‐discharge environment. Further examination of hospital outliersthose with the lowest and highest percentage of CMC hospitalizations with long LOSmay reveal opportunities to identify and spread best practices.

The demographic and clinical factors associated with long LOS in the present study, including age, ICU use, and transfer from another hospital, might help hospitals target which CMC have the greatest risk for experiencing long LOS. We found that infants age <1 year had longer LOS when compared with older children. Similar to our findings, younger‐aged children hospitalized with bronchiolitis have longer LOS.[36] Certainly, infants with medical complexity, in general, are a high‐acuity population with the potential for rapid clinical deterioration during an acute illness. Prolonged hospitalization for treatment and stabilization may be expected for many of them. Additional investigation is warranted to examine ICU use in CMC, and whether ICU admission or duration can be safely prevented or abbreviated. Opportunities to assess the quality of transfers into children's hospitals of CMC admitted to outside hospitals may be necessary. A study of pediatric burn patients reported that patients initially stabilized at a facility that was not a burn center and subsequently transferred to a burn center had a longer LOS than patients solely treated at a designated burn center.[37] Furthermore, events during transport itself may adversely impact the stability of an already fragile patient. Interventions to optimize the quality of care provided by transport teams have resulted in decreased LOS at the receiving hospital.[38]

This study's findings should be considered in the context of several limitations. Absent a gold‐standard definition of long LOS, we used the distribution of LOS across patients to inform our methods; LOS at the 90th percentile was selected as long. Although our sensitivity analysis using LOS at the 75th percentile produced similar findings, other cut points in LOS might be associated with different results. The study is not positioned to determine how much of the reported LOS was excessive, unnecessary, or preventable. The study findings may not generalize to types of hospitals not contained in PHIS (eg, nonchildren's hospitals and community hospitals). We did not focus on the impact of a new diagnosis (eg, new chronic illness) or acute in‐hospital event (eg, nosocomial infection) on prolonged LOS; future studies should investigate these clinical events with LOS.

PHIS does not contain information regarding characteristics that could influence LOS, including the children's social and familial attributes, transportation availability, home equipment needs, and local availability of postacute care facilities. Moreover, PHIS does not contain information about the hospital discharge procedures, process, or personnel across hospitals, which could influence LOS. Future studies on prolonged LOS should consider assessing this information. Because of the large sample size of hospitalizations included, the statistical power for the analyses was strong, rendering it possible that some findings that were statistically significant might have modest clinical significance (eg, relationship of Hispanic ethnicity with prolonged LOS). We could not determine why a positive correlation was not observed between hospitals' long LOS prevalence and their percentage of cost associated with long LOS; future studies should investigate the reasons for this finding.

Despite these limitations, the findings of the present study highlight the significance of long LOS in hospitalized CMC. These long hospitalizations account for a significant proportion of all hospital costs for this important population of children. The prevalence of long LOS for CMC varies considerably across children's hospitals, even after accounting for the case mix. Efforts to curtail hospital resource use and costs for CMC may benefit from focus on long LOS.

Children with medical complexity (CMC) have complex and chronic health conditions that often involve multiple organ systems and severely affect cognitive and physical functioning. Although the prevalence of CMC is low (1% of all children), they account for nearly one‐fifth of all pediatric admissions and one‐half of all hospital days and charges in the United States.[1] Over the last decade, CMC have had a particularly large and increasing impact in tertiary‐care children's hospitals.[1, 2] The Institute of Medicine has identified CMC as a priority population for a revised healthcare system.[3]

Medical homes, hospitals, health plans, states, federal agencies, and others are striving to reduce excessive hospital use in CMC because of its high cost.[4, 5, 6] Containing length of stay (LOS)an increasingly used indicator of the time sensitiveness and efficiency of hospital careis a common aim across these initiatives. CMC have longer hospitalizations than children without medical complexity. Speculated reasons for this are that CMC tend to have (1) higher severity of acute illnesses (eg, pneumonia, cellulitis), (2) prolonged recovery time in the hospital, and (3) higher risk of adverse events in the hospital. Moreover, hospital clinicians caring for CMC often find it difficult to determine discharge readiness, given that many CMC do not return to a completely healthy baseline.[7]

Little is known about long LOS in CMC, including which CMC have the highest risk of experiencing such stays and which stays might have the greatest opportunity to be shortened. Patient characteristics associated with prolonged length of stay have been studied extensively for many pediatric conditions (eg, asthma).[8, 9, 10, 11, 12, 13, 14] However, most of these studies excluded CMC. Therefore, the objectives of this study were to examine (1) the prevalence of long LOS in CMC, (2) patient characteristics associated with long LOS, and (3) hospital‐to‐hospital variation in prevalence of long LOS hospitalizations.

METHODS

RESULTS

Study Population

There were 954,018 hospitalizations of 217,163 CMC at 44 children's hospitals included for analysis. Forty‐seven percent of hospitalizations were for females, 49.4% for non‐Hispanic whites, and 61.1% for children with government insurance. Fifteen percent (n = 142,082) had a long LOS of 10 days. The median (IQR) LOS of hospitalizations <10 days versus 10 days were 2 (IQR, 14) and 16 days (IQR, 1226), respectively. Long LOS hospitalizations accounted for 61.1% (3.7 million) hospital days and 61.8% ($13.7 billion) of total hospitalization costs for all CMC in the cohort (Table 1).

Table 1.Demographic, Clinical, and Hospitalization Characteristics of Hospitalized Children With Medical Complexity by Length of Stay*

Characteristic	Overall (n = 954,018)	Length of Stay
		<10 Days (n = 811,936)	10 Days (n = 142,082)
NOTE: Abbreviations: IQR, interquartile range. *All comparisons were significant at the P < 0.001 level.
Age at admission, y, %
<1	14.6	12.7	25.7
14	27.1	27.9	22.4
59	20.1	21.0	14.9
1018	33.6	34.0	31.7
18+	4.6	4.4	5.4
Gender, %
Female	47.0	46.9	47.5
Race/ethnicity, %
Non‐Hispanic white	49.4	49.4	49.4
Non‐Hispanic black	23.1	23.8	19.3
Hispanic	18.2	17.8	20.4
Asian	2.0	1.9	2.3
Other	7.4	7.1	8.6
Complex chronic condition, %
Any	79.5	77.3	91.8
Technology assistance	37.1	34.1	54.2
Gastrointestinal	30.0	27.2	45.9
Neuromuscular	28.2	27.7	30.9
Cardiovascular	16.8	14.5	29.9
Respiratory	14.1	11.5	29.4
Congenital/genetic defect	17.2	16.7	20.2
Metabolic	9.9	8.9	15.4
Renal	10.1	9.5	13.8
Hematology/emmmunodeficiency	11.7	12.0	10.0
Neonatal	3.8	3.1	7.7
Transplantation	4.5	4.2	6.7
Clinical risk group, %
Chronic condition in 2 systems	68.4	71.2	53.9
Catastrophic chronic condition	31.4	28.8	46.1
Distance from hospital to home residence in miles, median [IQR]	16.2 [7.440.4]	15.8 [7.338.7]	19.1 [8.552.6]
Transferred from outside hospital (%)	6.5	5.3	13.6
Admitted for surgery, %	23.4	20.7	38.7
Use of intensive care, %	19.6	14.9	46.5
Discharge disposition, %
Home	91.2	92.9	81.4
Home healthcare	4.5	3.5	9.9
Other	2.9	2.6	4.5
Postacute care facility	1.1	0.8	3.1
Died	0.4	0.3	1.1
Payor, %
Government	61.1	60.6	63.5
Private	33.2	33.6	30.9
Other	5.7	5.7	5.7
Hospital resource use
Median length of stay [IQR]	3 [16]	2 [14]	16 [1226]
Median hospital cost [IQR]	$8,144 [$4,122$18,447]	$6,689 [$3,685$12,395]	$49,207 [$29,444$95,738]
Total hospital cost, $, billions	$22.2	$8.5	$13.7

Demographics and Clinical Characteristics of Children With and Without Long LOS

Compared with hospitalized CMC with LOS <10 days, a higher percentage of hospitalizations with LOS 10 days were CMC age <1 year (25.7% vs 12.7%, P < 0.001) and Hispanic (20.4% vs 17.8%, P < 0.001). CMC hospitalizations with a long LOS had a higher percentage of any CCC (91.8% vs 77.3%, P < 0.001); the most common CCCs were gastrointestinal (45.9%), neuromuscular (30.9%), and cardiovascular (29.9%). Hospitalizations of CMC with a long LOS had a higher percentage of a catastrophic chronic condition (46.1% vs 28.8%, P < 0.001) and technology dependence (46.1% vs 28.8%, P < 0.001) (Table 1).

Hospitalization Characteristics of Children With and Without Long LOS

Compared with hospitalizations of CMC with LOS <10 days, hospitalizations of CMC with a long LOS more often involved transfer in from another hospital at admission (13.6% vs 5.3%, P < 0.001). CMC hospital stays with a long LOS more often involved surgery (38.7% vs 20.7%, P < 0.001) and use of intensive care (46.5% vs 14.9%; P < 0.001). A higher percentage of CMC with long LOS were discharged with home health services (9.9% vs 3.5%; P < 0.001) (Table 1).

The most common admitting diagnoses and CCCs for hospitalizations of CMC with long LOS are presented in Table 2. The two most prevalent APR‐DRGs in CMC hospitalizations lasting 10 days or longer were cystic fibrosis (10.7%) and respiratory system disease with ventilator support (5.5%). The two most common chronic condition characteristics represented among long CMC hospitalizations were gastrointestinal devices (eg, gastrostomy tube) (39.7%) and heart and great vessel malformations (eg, tetralogy of Fallot) (12.8%). The 5 most common CCC subcategories, as listed in Table 2, account for nearly 100% of the patients with long LOS hospitalizations.

Table 2.Most Common Reasons for Admission and Specific Complex Chronic Conditions for Hospitalized Children With Medical Complexity Who Had Length of Stay 10 Days

NOTE: *Reason for admission identified using All‐Patient Refined Diagnosis‐Related Groups. Complex chronic conditions identified using Feudtner and colleagues set of International Classification of Diseases, 9th Revision, Clinical Modification codes. Gastrointestinal devices include gastrostomy, gastrojejunostomy, colostomy. Respiratory devices include tracheostomy, noninvasive positive pressure, ventilator.
Most common reason for admission*
Cystic fibrosis	10.7%
Respiratory system diagnosis with ventilator support 96+ hours	5.5%
Malfunction, reaction, and complication of cardiac or vascular device or procedure	2.8%
Craniotomy except for trauma	2.6%
Major small and large bowel procedures	2.3%
Most common complex chronic condition
Gastrointestinal devices	39.7%
Heart and great vessel malformations	12.8%
Cystic fibrosis	12.5%
Dysrhythmias	11.0%
Respiratory devices	10.7%

Multivariable Analysis of Characteristics Associated With Long LOS

In multivariable analysis, the highest likelihood of long LOS was experienced by children who received care in the ICU (odds ratio [OR]: 3.5, 95% confidence interval [CI]: 3.43.5), who had a respiratory CCC (OR: 2.7, 95% CI: 2.62.7), and who were transferred from another acute care hospital at admission (OR: 2.1, 95% CI: 2.0, 2.1). The likelihood of long LOS was also higher in children <1 year of age (OR: 1.2, 95% CI: 1.21.3), and Hispanic children (OR: 1.1, 95% CI 1.0‐1.10) (Table 3). Similar multivariable findings were observed in sensitivity analysis using the 75th percentile of LOS (4 days) as the model outcome.

Table 3.Multivariable Analysis of the Likelihood of Long Length of Stay 10 Days

Characteristic	Odds Ratio (95% CI) of LOS 10 Days	P Value
NOTE: Abbreviations: CI, confidence interval; LOS, length of stay.
Use of intensive care	3.5 (3.4‐3.5)	<0.001
Transfer from another acute‐care hospital	2.1 (2.0‐2.1)	<0.001
Procedure/surgery	1.8 (1.8‐1.9)	<0.001
Complex chronic condition
Respiratory	2.7 (2.6‐2.7)	<0.001
Gastrointestinal	1.8 (1.8‐1.8)	<0.001
Metabolic	1.7 (1.7‐1.7)	<0.001
Cardiovascular	1.6 (1.5‐1.6)	<0.001
Neonatal	1.5 (1.5‐1.5)	<0.001
Renal	1.4 (1.4‐1.4)	<0.001
Transplant	1.4 (1.4‐1.4)	<0.001
Hematology and immunodeficiency	1.3 (1.3‐1.3)	<0.001
Technology assistance	1.1 (1.1, 1.1)	<0.001
Neuromuscular	0.9 (0.9‐0.9)	<0.001
Congenital or genetic defect	0.8 (0.8‐0.8)	<0.001
Age at admission, y
<1	1.2 (1.2‐1.3)	<0.001
14	0.5 (0.5‐0.5)	<0.001
59	0.6 (0.6‐0.6)	<0.001
1018	0.9 (0.9‐0.9)	<0.001
18+	Reference
Male	0.9 (0.9‐0.9)	<0.001
Race/ethnicity
Non‐Hispanic black	0.9 (0.9‐0.9)	<0.001
Hispanic	1.1 (1.0‐1.1)	<0.001
Asian	1.0 (1.0‐1.1)	0.3
Other	1.1 (1.1‐1.1)	<0.001
Non‐Hispanic white	Reference
Payor
Private	0.9 (0.8 0.9)	<0.001
Other	1.0 (1.0‐1.0)	0.4
Government	Reference
Season
Spring	1.0 (1.0 1.0)	<0.001
Summer	0.9 (0.9‐0.9)	<0.001
Fall	1.0 (0.9‐1.0)	<0.001
Winter	Reference

Variation in the Prevalence of Long LOS Across Children's Hospitals

After controlling for demographic, clinical, and hospital characteristics associated with long LOS, there was significant (P < 0.001) variation in the prevalence of long LOS for CMC across children's hospitals in the cohort (range, 10.3%21.8%) (Figure 1). Twelve (27%) hospitals had a significantly (P < 0.001) higher prevalence of long LOS for their hospitalized CMC, compared to the mean. Eighteen (41%) had a significantly (P < 0.001) lower prevalence of long LOS for their hospitalized CMC. There was also significant variation across hospitals with respect to cost, with 49.7% to 73.7% of all hospital costs of CMC attributed to long LOS hospitalizations. Finally, there was indirect correlation with the prevalence of LOS across hospitals and the hospitals' 30‐day readmission rate ( = 0.3; P = 0.04). As the prevalence of long LOS increased, the readmission rate decreased.

jhm2633-fig-0001-m.png

DISCUSSION

The main findings from this study suggest that a small percentage of CMC experiencing long LOS account for the majority of hospital bed days and cost of all hospitalized CMC in children's hospitals. The likelihood of long LOS varies significantly by CMC's age, race/ethnicity, and payor as well as by type and number of chronic conditions. Among CMC with long LOS, the use of gastrointestinal devices such as gastrostomy tubes, as well as congenital heart disease, were highly prevalent. In multivariable analysis, the characteristics most strongly associated with LOS 10 days were use of the ICU, respiratory complex chronic condition, and transfer from another medical facility at admission. After adjusting for these factors, there was significant variation in the prevalence of LOS 10 days for CMC across children's hospitals.

Although it is well known that CMC as a whole have a major impact on resource use in children's hospitals, this study reveals that 15% of hospitalizations of CMC account for 62% of all hospital costs of CMC. That is, a small fraction of hospitalizations of CMC is largely responsible for the significant financial impact of hospital resource use. To date, most clinical efforts and policies striving to reduce hospital use in CMC have focused on avoiding readmissions or index hospital admissions entirely, rather than improving the efficiency of hospital care after admission occurs.[23, 24, 25, 26] In the adult population, the impact of long LOS on hospital costs has been recognized, and several Medicare incentive programs have focused on in‐hospital timeliness and efficiency. As a result, LOS in Medicare beneficiaries has decreased dramatically over the past 2 decades.[27, 28, 29, 30] Optimizing the efficiency of hospital care for CMC may be an important goal to pursue, especially with precedent set in the adult literature.

Perhaps the substantial variation across hospitals in the prevalence of long LOS in CMC indicates opportunity to improve the efficiency of their inpatient care. This variation was not due to differences across hospitals' case mix of CMC. Further investigation is needed to determine how much of it is due to differences in quality of care. Clinical practice guidelines for hospital treatment of common illnesses usually exclude CMC. In our clinical experience across 9 children's hospitals, we have experienced varying approaches to setting discharge goals (ie, parameters on how healthy the child needs to be to ensure a successful hospital discharge) for CMC.[31] When the goals are absent or not clearly articulated, they can contribute to a prolonged hospitalization. Some families of CMC report significant issues when working with pediatric hospital staff to assess their child's discharge readiness.[7, 32, 33] In addition, there is significant variation across states and regions in access to and quality of post‐discharge health services (eg, home nursing, postacute care, durable medical equipment).[34, 35] In some areas, many CMC are not actively involved with their primary care physician.[5] These issues might also influence the ability of some children's hospitals to efficiently discharge CMC to a safe and supportive post‐discharge environment. Further examination of hospital outliersthose with the lowest and highest percentage of CMC hospitalizations with long LOSmay reveal opportunities to identify and spread best practices.

The demographic and clinical factors associated with long LOS in the present study, including age, ICU use, and transfer from another hospital, might help hospitals target which CMC have the greatest risk for experiencing long LOS. We found that infants age <1 year had longer LOS when compared with older children. Similar to our findings, younger‐aged children hospitalized with bronchiolitis have longer LOS.[36] Certainly, infants with medical complexity, in general, are a high‐acuity population with the potential for rapid clinical deterioration during an acute illness. Prolonged hospitalization for treatment and stabilization may be expected for many of them. Additional investigation is warranted to examine ICU use in CMC, and whether ICU admission or duration can be safely prevented or abbreviated. Opportunities to assess the quality of transfers into children's hospitals of CMC admitted to outside hospitals may be necessary. A study of pediatric burn patients reported that patients initially stabilized at a facility that was not a burn center and subsequently transferred to a burn center had a longer LOS than patients solely treated at a designated burn center.[37] Furthermore, events during transport itself may adversely impact the stability of an already fragile patient. Interventions to optimize the quality of care provided by transport teams have resulted in decreased LOS at the receiving hospital.[38]

This study's findings should be considered in the context of several limitations. Absent a gold‐standard definition of long LOS, we used the distribution of LOS across patients to inform our methods; LOS at the 90th percentile was selected as long. Although our sensitivity analysis using LOS at the 75th percentile produced similar findings, other cut points in LOS might be associated with different results. The study is not positioned to determine how much of the reported LOS was excessive, unnecessary, or preventable. The study findings may not generalize to types of hospitals not contained in PHIS (eg, nonchildren's hospitals and community hospitals). We did not focus on the impact of a new diagnosis (eg, new chronic illness) or acute in‐hospital event (eg, nosocomial infection) on prolonged LOS; future studies should investigate these clinical events with LOS.

PHIS does not contain information regarding characteristics that could influence LOS, including the children's social and familial attributes, transportation availability, home equipment needs, and local availability of postacute care facilities. Moreover, PHIS does not contain information about the hospital discharge procedures, process, or personnel across hospitals, which could influence LOS. Future studies on prolonged LOS should consider assessing this information. Because of the large sample size of hospitalizations included, the statistical power for the analyses was strong, rendering it possible that some findings that were statistically significant might have modest clinical significance (eg, relationship of Hispanic ethnicity with prolonged LOS). We could not determine why a positive correlation was not observed between hospitals' long LOS prevalence and their percentage of cost associated with long LOS; future studies should investigate the reasons for this finding.

Despite these limitations, the findings of the present study highlight the significance of long LOS in hospitalized CMC. These long hospitalizations account for a significant proportion of all hospital costs for this important population of children. The prevalence of long LOS for CMC varies considerably across children's hospitals, even after accounting for the case mix. Efforts to curtail hospital resource use and costs for CMC may benefit from focus on long LOS.

Children with medical complexity (CMC) have complex and chronic health conditions that often involve multiple organ systems and severely affect cognitive and physical functioning. Although the prevalence of CMC is low (1% of all children), they account for nearly one‐fifth of all pediatric admissions and one‐half of all hospital days and charges in the United States.[1] Over the last decade, CMC have had a particularly large and increasing impact in tertiary‐care children's hospitals.[1, 2] The Institute of Medicine has identified CMC as a priority population for a revised healthcare system.[3]

Medical homes, hospitals, health plans, states, federal agencies, and others are striving to reduce excessive hospital use in CMC because of its high cost.[4, 5, 6] Containing length of stay (LOS)an increasingly used indicator of the time sensitiveness and efficiency of hospital careis a common aim across these initiatives. CMC have longer hospitalizations than children without medical complexity. Speculated reasons for this are that CMC tend to have (1) higher severity of acute illnesses (eg, pneumonia, cellulitis), (2) prolonged recovery time in the hospital, and (3) higher risk of adverse events in the hospital. Moreover, hospital clinicians caring for CMC often find it difficult to determine discharge readiness, given that many CMC do not return to a completely healthy baseline.[7]

Little is known about long LOS in CMC, including which CMC have the highest risk of experiencing such stays and which stays might have the greatest opportunity to be shortened. Patient characteristics associated with prolonged length of stay have been studied extensively for many pediatric conditions (eg, asthma).[8, 9, 10, 11, 12, 13, 14] However, most of these studies excluded CMC. Therefore, the objectives of this study were to examine (1) the prevalence of long LOS in CMC, (2) patient characteristics associated with long LOS, and (3) hospital‐to‐hospital variation in prevalence of long LOS hospitalizations.

METHODS

RESULTS

Study Population

There were 954,018 hospitalizations of 217,163 CMC at 44 children's hospitals included for analysis. Forty‐seven percent of hospitalizations were for females, 49.4% for non‐Hispanic whites, and 61.1% for children with government insurance. Fifteen percent (n = 142,082) had a long LOS of 10 days. The median (IQR) LOS of hospitalizations <10 days versus 10 days were 2 (IQR, 14) and 16 days (IQR, 1226), respectively. Long LOS hospitalizations accounted for 61.1% (3.7 million) hospital days and 61.8% ($13.7 billion) of total hospitalization costs for all CMC in the cohort (Table 1).

Table 1.Demographic, Clinical, and Hospitalization Characteristics of Hospitalized Children With Medical Complexity by Length of Stay*

Characteristic	Overall (n = 954,018)	Length of Stay
		<10 Days (n = 811,936)	10 Days (n = 142,082)
NOTE: Abbreviations: IQR, interquartile range. *All comparisons were significant at the P < 0.001 level.
Age at admission, y, %
<1	14.6	12.7	25.7
14	27.1	27.9	22.4
59	20.1	21.0	14.9
1018	33.6	34.0	31.7
18+	4.6	4.4	5.4
Gender, %
Female	47.0	46.9	47.5
Race/ethnicity, %
Non‐Hispanic white	49.4	49.4	49.4
Non‐Hispanic black	23.1	23.8	19.3
Hispanic	18.2	17.8	20.4
Asian	2.0	1.9	2.3
Other	7.4	7.1	8.6
Complex chronic condition, %
Any	79.5	77.3	91.8
Technology assistance	37.1	34.1	54.2
Gastrointestinal	30.0	27.2	45.9
Neuromuscular	28.2	27.7	30.9
Cardiovascular	16.8	14.5	29.9
Respiratory	14.1	11.5	29.4
Congenital/genetic defect	17.2	16.7	20.2
Metabolic	9.9	8.9	15.4
Renal	10.1	9.5	13.8
Hematology/emmmunodeficiency	11.7	12.0	10.0
Neonatal	3.8	3.1	7.7
Transplantation	4.5	4.2	6.7
Clinical risk group, %
Chronic condition in 2 systems	68.4	71.2	53.9
Catastrophic chronic condition	31.4	28.8	46.1
Distance from hospital to home residence in miles, median [IQR]	16.2 [7.440.4]	15.8 [7.338.7]	19.1 [8.552.6]
Transferred from outside hospital (%)	6.5	5.3	13.6
Admitted for surgery, %	23.4	20.7	38.7
Use of intensive care, %	19.6	14.9	46.5
Discharge disposition, %
Home	91.2	92.9	81.4
Home healthcare	4.5	3.5	9.9
Other	2.9	2.6	4.5
Postacute care facility	1.1	0.8	3.1
Died	0.4	0.3	1.1
Payor, %
Government	61.1	60.6	63.5
Private	33.2	33.6	30.9
Other	5.7	5.7	5.7
Hospital resource use
Median length of stay [IQR]	3 [16]	2 [14]	16 [1226]
Median hospital cost [IQR]	$8,144 [$4,122$18,447]	$6,689 [$3,685$12,395]	$49,207 [$29,444$95,738]
Total hospital cost, $, billions	$22.2	$8.5	$13.7

Demographics and Clinical Characteristics of Children With and Without Long LOS

Compared with hospitalized CMC with LOS <10 days, a higher percentage of hospitalizations with LOS 10 days were CMC age <1 year (25.7% vs 12.7%, P < 0.001) and Hispanic (20.4% vs 17.8%, P < 0.001). CMC hospitalizations with a long LOS had a higher percentage of any CCC (91.8% vs 77.3%, P < 0.001); the most common CCCs were gastrointestinal (45.9%), neuromuscular (30.9%), and cardiovascular (29.9%). Hospitalizations of CMC with a long LOS had a higher percentage of a catastrophic chronic condition (46.1% vs 28.8%, P < 0.001) and technology dependence (46.1% vs 28.8%, P < 0.001) (Table 1).

Hospitalization Characteristics of Children With and Without Long LOS

Compared with hospitalizations of CMC with LOS <10 days, hospitalizations of CMC with a long LOS more often involved transfer in from another hospital at admission (13.6% vs 5.3%, P < 0.001). CMC hospital stays with a long LOS more often involved surgery (38.7% vs 20.7%, P < 0.001) and use of intensive care (46.5% vs 14.9%; P < 0.001). A higher percentage of CMC with long LOS were discharged with home health services (9.9% vs 3.5%; P < 0.001) (Table 1).

The most common admitting diagnoses and CCCs for hospitalizations of CMC with long LOS are presented in Table 2. The two most prevalent APR‐DRGs in CMC hospitalizations lasting 10 days or longer were cystic fibrosis (10.7%) and respiratory system disease with ventilator support (5.5%). The two most common chronic condition characteristics represented among long CMC hospitalizations were gastrointestinal devices (eg, gastrostomy tube) (39.7%) and heart and great vessel malformations (eg, tetralogy of Fallot) (12.8%). The 5 most common CCC subcategories, as listed in Table 2, account for nearly 100% of the patients with long LOS hospitalizations.

Table 2.Most Common Reasons for Admission and Specific Complex Chronic Conditions for Hospitalized Children With Medical Complexity Who Had Length of Stay 10 Days

NOTE: *Reason for admission identified using All‐Patient Refined Diagnosis‐Related Groups. Complex chronic conditions identified using Feudtner and colleagues set of International Classification of Diseases, 9th Revision, Clinical Modification codes. Gastrointestinal devices include gastrostomy, gastrojejunostomy, colostomy. Respiratory devices include tracheostomy, noninvasive positive pressure, ventilator.
Most common reason for admission*
Cystic fibrosis	10.7%
Respiratory system diagnosis with ventilator support 96+ hours	5.5%
Malfunction, reaction, and complication of cardiac or vascular device or procedure	2.8%
Craniotomy except for trauma	2.6%
Major small and large bowel procedures	2.3%
Most common complex chronic condition
Gastrointestinal devices	39.7%
Heart and great vessel malformations	12.8%
Cystic fibrosis	12.5%
Dysrhythmias	11.0%
Respiratory devices	10.7%

Multivariable Analysis of Characteristics Associated With Long LOS

In multivariable analysis, the highest likelihood of long LOS was experienced by children who received care in the ICU (odds ratio [OR]: 3.5, 95% confidence interval [CI]: 3.43.5), who had a respiratory CCC (OR: 2.7, 95% CI: 2.62.7), and who were transferred from another acute care hospital at admission (OR: 2.1, 95% CI: 2.0, 2.1). The likelihood of long LOS was also higher in children <1 year of age (OR: 1.2, 95% CI: 1.21.3), and Hispanic children (OR: 1.1, 95% CI 1.0‐1.10) (Table 3). Similar multivariable findings were observed in sensitivity analysis using the 75th percentile of LOS (4 days) as the model outcome.

Table 3.Multivariable Analysis of the Likelihood of Long Length of Stay 10 Days

Characteristic	Odds Ratio (95% CI) of LOS 10 Days	P Value
NOTE: Abbreviations: CI, confidence interval; LOS, length of stay.
Use of intensive care	3.5 (3.4‐3.5)	<0.001
Transfer from another acute‐care hospital	2.1 (2.0‐2.1)	<0.001
Procedure/surgery	1.8 (1.8‐1.9)	<0.001
Complex chronic condition
Respiratory	2.7 (2.6‐2.7)	<0.001
Gastrointestinal	1.8 (1.8‐1.8)	<0.001
Metabolic	1.7 (1.7‐1.7)	<0.001
Cardiovascular	1.6 (1.5‐1.6)	<0.001
Neonatal	1.5 (1.5‐1.5)	<0.001
Renal	1.4 (1.4‐1.4)	<0.001
Transplant	1.4 (1.4‐1.4)	<0.001
Hematology and immunodeficiency	1.3 (1.3‐1.3)	<0.001
Technology assistance	1.1 (1.1, 1.1)	<0.001
Neuromuscular	0.9 (0.9‐0.9)	<0.001
Congenital or genetic defect	0.8 (0.8‐0.8)	<0.001
Age at admission, y
<1	1.2 (1.2‐1.3)	<0.001
14	0.5 (0.5‐0.5)	<0.001
59	0.6 (0.6‐0.6)	<0.001
1018	0.9 (0.9‐0.9)	<0.001
18+	Reference
Male	0.9 (0.9‐0.9)	<0.001
Race/ethnicity
Non‐Hispanic black	0.9 (0.9‐0.9)	<0.001
Hispanic	1.1 (1.0‐1.1)	<0.001
Asian	1.0 (1.0‐1.1)	0.3
Other	1.1 (1.1‐1.1)	<0.001
Non‐Hispanic white	Reference
Payor
Private	0.9 (0.8 0.9)	<0.001
Other	1.0 (1.0‐1.0)	0.4
Government	Reference
Season
Spring	1.0 (1.0 1.0)	<0.001
Summer	0.9 (0.9‐0.9)	<0.001
Fall	1.0 (0.9‐1.0)	<0.001
Winter	Reference

Variation in the Prevalence of Long LOS Across Children's Hospitals

After controlling for demographic, clinical, and hospital characteristics associated with long LOS, there was significant (P < 0.001) variation in the prevalence of long LOS for CMC across children's hospitals in the cohort (range, 10.3%21.8%) (Figure 1). Twelve (27%) hospitals had a significantly (P < 0.001) higher prevalence of long LOS for their hospitalized CMC, compared to the mean. Eighteen (41%) had a significantly (P < 0.001) lower prevalence of long LOS for their hospitalized CMC. There was also significant variation across hospitals with respect to cost, with 49.7% to 73.7% of all hospital costs of CMC attributed to long LOS hospitalizations. Finally, there was indirect correlation with the prevalence of LOS across hospitals and the hospitals' 30‐day readmission rate ( = 0.3; P = 0.04). As the prevalence of long LOS increased, the readmission rate decreased.

jhm2633-fig-0001-m.png

DISCUSSION

The main findings from this study suggest that a small percentage of CMC experiencing long LOS account for the majority of hospital bed days and cost of all hospitalized CMC in children's hospitals. The likelihood of long LOS varies significantly by CMC's age, race/ethnicity, and payor as well as by type and number of chronic conditions. Among CMC with long LOS, the use of gastrointestinal devices such as gastrostomy tubes, as well as congenital heart disease, were highly prevalent. In multivariable analysis, the characteristics most strongly associated with LOS 10 days were use of the ICU, respiratory complex chronic condition, and transfer from another medical facility at admission. After adjusting for these factors, there was significant variation in the prevalence of LOS 10 days for CMC across children's hospitals.

Although it is well known that CMC as a whole have a major impact on resource use in children's hospitals, this study reveals that 15% of hospitalizations of CMC account for 62% of all hospital costs of CMC. That is, a small fraction of hospitalizations of CMC is largely responsible for the significant financial impact of hospital resource use. To date, most clinical efforts and policies striving to reduce hospital use in CMC have focused on avoiding readmissions or index hospital admissions entirely, rather than improving the efficiency of hospital care after admission occurs.[23, 24, 25, 26] In the adult population, the impact of long LOS on hospital costs has been recognized, and several Medicare incentive programs have focused on in‐hospital timeliness and efficiency. As a result, LOS in Medicare beneficiaries has decreased dramatically over the past 2 decades.[27, 28, 29, 30] Optimizing the efficiency of hospital care for CMC may be an important goal to pursue, especially with precedent set in the adult literature.

Perhaps the substantial variation across hospitals in the prevalence of long LOS in CMC indicates opportunity to improve the efficiency of their inpatient care. This variation was not due to differences across hospitals' case mix of CMC. Further investigation is needed to determine how much of it is due to differences in quality of care. Clinical practice guidelines for hospital treatment of common illnesses usually exclude CMC. In our clinical experience across 9 children's hospitals, we have experienced varying approaches to setting discharge goals (ie, parameters on how healthy the child needs to be to ensure a successful hospital discharge) for CMC.[31] When the goals are absent or not clearly articulated, they can contribute to a prolonged hospitalization. Some families of CMC report significant issues when working with pediatric hospital staff to assess their child's discharge readiness.[7, 32, 33] In addition, there is significant variation across states and regions in access to and quality of post‐discharge health services (eg, home nursing, postacute care, durable medical equipment).[34, 35] In some areas, many CMC are not actively involved with their primary care physician.[5] These issues might also influence the ability of some children's hospitals to efficiently discharge CMC to a safe and supportive post‐discharge environment. Further examination of hospital outliersthose with the lowest and highest percentage of CMC hospitalizations with long LOSmay reveal opportunities to identify and spread best practices.

The demographic and clinical factors associated with long LOS in the present study, including age, ICU use, and transfer from another hospital, might help hospitals target which CMC have the greatest risk for experiencing long LOS. We found that infants age <1 year had longer LOS when compared with older children. Similar to our findings, younger‐aged children hospitalized with bronchiolitis have longer LOS.[36] Certainly, infants with medical complexity, in general, are a high‐acuity population with the potential for rapid clinical deterioration during an acute illness. Prolonged hospitalization for treatment and stabilization may be expected for many of them. Additional investigation is warranted to examine ICU use in CMC, and whether ICU admission or duration can be safely prevented or abbreviated. Opportunities to assess the quality of transfers into children's hospitals of CMC admitted to outside hospitals may be necessary. A study of pediatric burn patients reported that patients initially stabilized at a facility that was not a burn center and subsequently transferred to a burn center had a longer LOS than patients solely treated at a designated burn center.[37] Furthermore, events during transport itself may adversely impact the stability of an already fragile patient. Interventions to optimize the quality of care provided by transport teams have resulted in decreased LOS at the receiving hospital.[38]

This study's findings should be considered in the context of several limitations. Absent a gold‐standard definition of long LOS, we used the distribution of LOS across patients to inform our methods; LOS at the 90th percentile was selected as long. Although our sensitivity analysis using LOS at the 75th percentile produced similar findings, other cut points in LOS might be associated with different results. The study is not positioned to determine how much of the reported LOS was excessive, unnecessary, or preventable. The study findings may not generalize to types of hospitals not contained in PHIS (eg, nonchildren's hospitals and community hospitals). We did not focus on the impact of a new diagnosis (eg, new chronic illness) or acute in‐hospital event (eg, nosocomial infection) on prolonged LOS; future studies should investigate these clinical events with LOS.

PHIS does not contain information regarding characteristics that could influence LOS, including the children's social and familial attributes, transportation availability, home equipment needs, and local availability of postacute care facilities. Moreover, PHIS does not contain information about the hospital discharge procedures, process, or personnel across hospitals, which could influence LOS. Future studies on prolonged LOS should consider assessing this information. Because of the large sample size of hospitalizations included, the statistical power for the analyses was strong, rendering it possible that some findings that were statistically significant might have modest clinical significance (eg, relationship of Hispanic ethnicity with prolonged LOS). We could not determine why a positive correlation was not observed between hospitals' long LOS prevalence and their percentage of cost associated with long LOS; future studies should investigate the reasons for this finding.

Despite these limitations, the findings of the present study highlight the significance of long LOS in hospitalized CMC. These long hospitalizations account for a significant proportion of all hospital costs for this important population of children. The prevalence of long LOS for CMC varies considerably across children's hospitals, even after accounting for the case mix. Efforts to curtail hospital resource use and costs for CMC may benefit from focus on long LOS.

Timely and reliable communication of important data between hospital‐based physicians and primary care physicians is critical for prevention of medical adverse events.[1, 2] Extrapolation from high‐performance organizations outside of medicine suggests that verbal communication is an important component of patient handoffs.[3, 4] Though the Joint Commission does not mandate verbal communication during handoffs per se, stipulating instead that handoff participants have an opportunity to ask and respond to questions,[5] there is some evidence that primary care providers prefer verbal handoffs at least for certain patients such as those with medical complexity.[6] Verbal communication offers the receiver the opportunity to ask questions, but in practice, 2‐way verbal communication is often difficult to achieve at hospital discharge.

At our institution, hospital medicine (HM) physicians serve as the primary inpatient providers for nearly 90% of all general pediatric admissions. When the HM service was established, primary care physicians (PCPs) and HM physicians together agreed upon an expectation for verbal, physician‐to‐physician communication at the time of discharge. Discharge communication is provided by either residents or attendings depending on the facility. A telephone operator service called Physician Priority Link (PPL) was made available to facilitate this communication. The PPL service is staffed 24/7 by operators whose only responsibilities are to connect providers inside and outside the institution. By utilizing this service, PCPs could respond in a nonemergent fashion to discharge phone calls.

Over the last several years, PCPs have observed high variation in the reliability of discharge communication phone calls. A review of PPL phone records in 2009 showed that only 52% of HM discharges had a record of a call initiated to the PCP on the day of discharge. The overall goal of this improvement project was to improve the completion of verbal handoffs from HM physicians (residents or attendings) to PCPs. The specific aim of the project was to increase the proportion of completed verbal handoffs from on‐call residents or attendings to PCPs within 24 hours of discharge to more than 90% within 18 months.

METHODS

Planning the Intervention

The project was championed by physicians on the HM service and supported by a chief resident, PPL administrators, and 2 information technology analysts.

At the onset of the project, the team mapped the process for completing a discharge call to the PCPs, conducted a modified failure mode and effects analysis,[7, 8] and examined the key drivers used to prioritize interventions (Figure 1). Through the modified failure modes effect analysis, the team was able to identify system issues that led to unsuccessful communication: failure of call initiation, absence of an identified PCP, long wait times on hold, failure of PCP to call back, and failure of the call to be documented. These failure modes informed the key drivers to achieving the study aim. Figure 2 depicts the final key drivers, which were revised through testing and learning.

jhm2392-fig-0001-m.png

jhm2392-fig-0002-m.png

Interventions to Standardization the Communication Process

To facilitate initiation of calls to PCPs at hospital discharge, the improvement team created a standard process using the PPL service (Figure 3). All patients discharged from the HM service were included in the process. Discharging physicians (who were usually but not always residents, depending on the facility), were instructed to call the PPL operator at the time of discharge. The PPL operator would then page the patient's PCP. It was the responsibility of the discharging physician to identify a PCP prior to discharge. Instances where no PCP was identified were counted as process failures because no phone call could be made. The expectation for the PCPs was that they would return the page within 20 minutes. PPL operators would then page back to the discharging physician to connect the 2 parties with the expectation that the discharging physician respond within 2 to 4 minutes to the PPL operator's page. Standardization of all calls through PPL allowed efficient tracking of incomplete calls and operators to reattempt calls that were not completed. This process also shifted the burden of following up on incomplete calls to PPL. The use of PPL to make the connection also allowed the physician to complete other work while awaiting a call back from the PCP.

jhm2392-fig-0003-m.png

RESULTS

From July 2011 to March 2014, there were 6313 discharges from the HM service. The process measure (percentage of calls initiated) improved from 50% to 97% after 4 interventions (Figure 4). Data for the outcome measure (percentage of calls completed) were collected starting in August 2012, shortly after linking the EHR discharge order to the discharge call. Over the first 8 weeks, our median was 80%, which increased to a median of 93% (Figure 5). These results were sustained for 18 months.

jhm2392-fig-0004-m.png

jhm2392-fig-0005-m.png

Several key interventions were identified that were critical to achievement of our goal. Standardization of the communication process through PPL was temporally associated with a shift in the median rate of call initiation from 52% to 72%. Use of the discharge order to initiate discharge communication was associated with an increase from 72% to 97%. Finally, the percentage of completed verbal handoffs increased to more than 93% following batching of phone calls to PCPs at specific times during the day.

DISCUSSION

We used improvement and reliability science methods to implement a successful process for improving verbal handoffs from HM physicians to PCPs within 24 hours of discharge to 93%. This result has been sustained for 18 months.

Utilization of the PPL call center for flexible call facilitation along with support for data analysis and leveraging the EHR to automate the process increased reliability, leading to rapid improvement. Prior to mandating the use of PPL to connect discharging physicians with PCPs, the exact rate of successful handoffs in our institution was not known. We do know, however, that only 52% of calls were initiated, so clearly a large gap was present prior to our improvement work. Data collection from the PPL system was automated so that accurate, timely, and sustainable data could be provided, greatly aiding improvement efforts. Flexibility in call‐back timing was also crucial, because coordinating the availability of PCPs and discharging physicians is often challenging. The EHR‐initiated process for discharge communication was a key intervention, and improvement of our process measure to 97% performance was associated with this implementation. Two final interventions: (1) assignment of responsibility for communication to a team pager held by a designated resident and (2) batching of calls to specific times streamlined the EHR‐initiated process and were associated with achievement of our main outcome goal of >90% completed verbal communication.

There are several reports of successful interventions to improve receipt or content of discharge summaries by PCPs following hospital discharge available in the literature.[14, 15, 16, 17, 18, 19, 20] Recently, Shen et al. reported on the success of a multisite improvement collaborative involving pediatric hospitalist programs at community hospitals whose aim was to improve the timely documentation of communication directed at PCPs.[21] In their report, all 7 hospital sites that participated in the collaborative for more than 4 months were able to demonstrate substantial improvement in documentation of some form of communication directed at PCPs (whether by e‐mail, fax, or telephone call), from a baseline of approximately 50% to more than 90%. A limitation of their study was that they were unable to document whether PCPs had received any information or by what method. A recent survey of PCPs by Sheu et al. indicated that for many discharges, information in addition to that present in the EHR was desirable to ensure a safe transition of care.[6] Two‐way communication, such as with a phone call, allows for senders to verify information receipt and for receivers to ask questions to ensure complete information. To our knowledge, there have been no previous reports describing processes for improving verbal communication between hospitalist services and PCPs at discharge.

It may be that use of the call system allowed PCPs to return phone calls regarding discharges at convenient stopping points in their day while allowing discharging physicians to initiate a call without having to wait on hold. Interestingly, though we anticipated the need for additional PPL resources during the course of this improvement, the final process was efficient enough that PPL did not require additional staffing to accommodate the higher call volume.

A key insight during our implementation was that relying on the EHR to initiate every discharge communication created disruption of resident workflow due to disregard of patient, resident, and PCP factors. This was reflected by the improvement in call initiation (our process measure) following this intervention, whereas at the same time call completion (our outcome measure) remained below goal. To achieve our goal of completing verbal communication required a process that was highly reliable yet flexible enough to allow discharging physicians to complete the call in the unpredictable environment of inpatient care. Ultimately, this was achieved by allowing discharging physicians to initiate the process when convenient, and allowing for the EHR‐initiated process to function as a backup strategy to identify and mitigate failures of initiation.

An important limitation of our study was the lack of PCPs on the improvement team, likely making the success of the project more difficult than it might have been. For example, during the study we did not measure the time PCPs spent on hold or how many reattempts were needed to complete the communication loop. Immediately following the completion of our study, it became apparent that physicians returning calls for our own institution's primary care clinic were experiencing regular workflow interruptions and occasional hold times more than 20 minutes, necessitating ongoing further work to determine the root causes and solutions to these problems. Though this work is ongoing, average PCP hold times measured from a sample of call reviews in 2013 to 2014 was 3 minutes and 15 seconds.

This study has several other limitations. We were unable to account for phone calls to PCPs initiated outside of the new process. It may be that PCPs were called more than 52% of the time at baseline due to noncompliance with the new protocol. Also, we only have data for call completion starting after implementation of the link between the discharge order and the discharge phone call, making the baseline appear artificially high and precluding any analysis of how earlier interventions affected our outcome metric. Communication with PCPs should ideally occur prior to discharge. An important limitation of our process is that calls could occur several hours after discharge between an on‐call resident and an on‐call outpatient physician rather than between the PCP and the discharging resident, limiting appropriate information exchange. Though verbal discharge communication is a desirable goal for many reasons, the current project did not focus on the quality of the call or the information that was transmitted to the PCP. Additionally, direct attending‐to‐attending communication may be valuable with medically or socially complex discharges, but we did not have a process to facilitate this. We also did not measure what effect our new process had on outcomes such as quality of patient and family transition from hospital or physician satisfaction. The existence of programs similar to our PPL subspecialty referral line may be limited to large institutions. However, it should be noted that although some internal resource reallocation was necessary within PPL, no actual staffing increases were required despite a large increase in call volume. It may be that any hospital operator system could be adapted for this purpose with modest additional resources. Finally, although our EHR system is widely utilized, there are many competing systems in the market, and our intervention required utilization of EHR capabilities that may not be present in all systems. However, our EHR intervention utilized existing functionality and did not require modification of the system.

This project focused on discharge phone calls to primary care physicians for patients hospitalized on the hospital medicine service. Because communication with the PCP should ideally occur prior to discharge, future work will include identifying a more proximal trigger than the discharge order to which to link the EHR trigger for discharge communication. Other next steps to improve handoff effectiveness and optimize the efficiency of our process include identifying essential information that should be transmitted to the primary care physician at the time of the phone call, developing processes to ensure communication of this information, measuring PCP satisfaction with this communication, and measuring the impact on patient outcomes. Finally, though expert opinion indicates that verbal handoffs may have safety advantages over nonverbal handoffs, studies comparing the safety and efficacy of verbal versus nonverbal handoffs at patient discharge are lacking. Studies establishing the relative efficacy and safety of verbal versus nonverbal handoffs at hospital discharge are needed. Knowledge gained from these activities could inform future projects centered on the spread of the process to other hospital services and/or other hospitals.

CONCLUSION

We increased the percentage of calls initiated to PCPs at patient discharge from 52% to 97% and the percentage of calls completed between HM physicians and PCPs to 93% through the use of a standardized discharge communication process coupled with a basic EHR messaging functionality. The results of this study may be of interest for further testing and adaptation for any institution with an electronic healthcare system.

Disclosure: Nothing to report.

Timely and reliable communication of important data between hospital‐based physicians and primary care physicians is critical for prevention of medical adverse events.[1, 2] Extrapolation from high‐performance organizations outside of medicine suggests that verbal communication is an important component of patient handoffs.[3, 4] Though the Joint Commission does not mandate verbal communication during handoffs per se, stipulating instead that handoff participants have an opportunity to ask and respond to questions,[5] there is some evidence that primary care providers prefer verbal handoffs at least for certain patients such as those with medical complexity.[6] Verbal communication offers the receiver the opportunity to ask questions, but in practice, 2‐way verbal communication is often difficult to achieve at hospital discharge.

At our institution, hospital medicine (HM) physicians serve as the primary inpatient providers for nearly 90% of all general pediatric admissions. When the HM service was established, primary care physicians (PCPs) and HM physicians together agreed upon an expectation for verbal, physician‐to‐physician communication at the time of discharge. Discharge communication is provided by either residents or attendings depending on the facility. A telephone operator service called Physician Priority Link (PPL) was made available to facilitate this communication. The PPL service is staffed 24/7 by operators whose only responsibilities are to connect providers inside and outside the institution. By utilizing this service, PCPs could respond in a nonemergent fashion to discharge phone calls.

Over the last several years, PCPs have observed high variation in the reliability of discharge communication phone calls. A review of PPL phone records in 2009 showed that only 52% of HM discharges had a record of a call initiated to the PCP on the day of discharge. The overall goal of this improvement project was to improve the completion of verbal handoffs from HM physicians (residents or attendings) to PCPs. The specific aim of the project was to increase the proportion of completed verbal handoffs from on‐call residents or attendings to PCPs within 24 hours of discharge to more than 90% within 18 months.

METHODS

Planning the Intervention

The project was championed by physicians on the HM service and supported by a chief resident, PPL administrators, and 2 information technology analysts.

At the onset of the project, the team mapped the process for completing a discharge call to the PCPs, conducted a modified failure mode and effects analysis,[7, 8] and examined the key drivers used to prioritize interventions (Figure 1). Through the modified failure modes effect analysis, the team was able to identify system issues that led to unsuccessful communication: failure of call initiation, absence of an identified PCP, long wait times on hold, failure of PCP to call back, and failure of the call to be documented. These failure modes informed the key drivers to achieving the study aim. Figure 2 depicts the final key drivers, which were revised through testing and learning.

jhm2392-fig-0001-m.png

jhm2392-fig-0002-m.png

Interventions to Standardization the Communication Process

To facilitate initiation of calls to PCPs at hospital discharge, the improvement team created a standard process using the PPL service (Figure 3). All patients discharged from the HM service were included in the process. Discharging physicians (who were usually but not always residents, depending on the facility), were instructed to call the PPL operator at the time of discharge. The PPL operator would then page the patient's PCP. It was the responsibility of the discharging physician to identify a PCP prior to discharge. Instances where no PCP was identified were counted as process failures because no phone call could be made. The expectation for the PCPs was that they would return the page within 20 minutes. PPL operators would then page back to the discharging physician to connect the 2 parties with the expectation that the discharging physician respond within 2 to 4 minutes to the PPL operator's page. Standardization of all calls through PPL allowed efficient tracking of incomplete calls and operators to reattempt calls that were not completed. This process also shifted the burden of following up on incomplete calls to PPL. The use of PPL to make the connection also allowed the physician to complete other work while awaiting a call back from the PCP.

jhm2392-fig-0003-m.png

RESULTS

From July 2011 to March 2014, there were 6313 discharges from the HM service. The process measure (percentage of calls initiated) improved from 50% to 97% after 4 interventions (Figure 4). Data for the outcome measure (percentage of calls completed) were collected starting in August 2012, shortly after linking the EHR discharge order to the discharge call. Over the first 8 weeks, our median was 80%, which increased to a median of 93% (Figure 5). These results were sustained for 18 months.

jhm2392-fig-0004-m.png

jhm2392-fig-0005-m.png

Several key interventions were identified that were critical to achievement of our goal. Standardization of the communication process through PPL was temporally associated with a shift in the median rate of call initiation from 52% to 72%. Use of the discharge order to initiate discharge communication was associated with an increase from 72% to 97%. Finally, the percentage of completed verbal handoffs increased to more than 93% following batching of phone calls to PCPs at specific times during the day.

DISCUSSION

We used improvement and reliability science methods to implement a successful process for improving verbal handoffs from HM physicians to PCPs within 24 hours of discharge to 93%. This result has been sustained for 18 months.

Utilization of the PPL call center for flexible call facilitation along with support for data analysis and leveraging the EHR to automate the process increased reliability, leading to rapid improvement. Prior to mandating the use of PPL to connect discharging physicians with PCPs, the exact rate of successful handoffs in our institution was not known. We do know, however, that only 52% of calls were initiated, so clearly a large gap was present prior to our improvement work. Data collection from the PPL system was automated so that accurate, timely, and sustainable data could be provided, greatly aiding improvement efforts. Flexibility in call‐back timing was also crucial, because coordinating the availability of PCPs and discharging physicians is often challenging. The EHR‐initiated process for discharge communication was a key intervention, and improvement of our process measure to 97% performance was associated with this implementation. Two final interventions: (1) assignment of responsibility for communication to a team pager held by a designated resident and (2) batching of calls to specific times streamlined the EHR‐initiated process and were associated with achievement of our main outcome goal of >90% completed verbal communication.

There are several reports of successful interventions to improve receipt or content of discharge summaries by PCPs following hospital discharge available in the literature.[14, 15, 16, 17, 18, 19, 20] Recently, Shen et al. reported on the success of a multisite improvement collaborative involving pediatric hospitalist programs at community hospitals whose aim was to improve the timely documentation of communication directed at PCPs.[21] In their report, all 7 hospital sites that participated in the collaborative for more than 4 months were able to demonstrate substantial improvement in documentation of some form of communication directed at PCPs (whether by e‐mail, fax, or telephone call), from a baseline of approximately 50% to more than 90%. A limitation of their study was that they were unable to document whether PCPs had received any information or by what method. A recent survey of PCPs by Sheu et al. indicated that for many discharges, information in addition to that present in the EHR was desirable to ensure a safe transition of care.[6] Two‐way communication, such as with a phone call, allows for senders to verify information receipt and for receivers to ask questions to ensure complete information. To our knowledge, there have been no previous reports describing processes for improving verbal communication between hospitalist services and PCPs at discharge.

It may be that use of the call system allowed PCPs to return phone calls regarding discharges at convenient stopping points in their day while allowing discharging physicians to initiate a call without having to wait on hold. Interestingly, though we anticipated the need for additional PPL resources during the course of this improvement, the final process was efficient enough that PPL did not require additional staffing to accommodate the higher call volume.

A key insight during our implementation was that relying on the EHR to initiate every discharge communication created disruption of resident workflow due to disregard of patient, resident, and PCP factors. This was reflected by the improvement in call initiation (our process measure) following this intervention, whereas at the same time call completion (our outcome measure) remained below goal. To achieve our goal of completing verbal communication required a process that was highly reliable yet flexible enough to allow discharging physicians to complete the call in the unpredictable environment of inpatient care. Ultimately, this was achieved by allowing discharging physicians to initiate the process when convenient, and allowing for the EHR‐initiated process to function as a backup strategy to identify and mitigate failures of initiation.

An important limitation of our study was the lack of PCPs on the improvement team, likely making the success of the project more difficult than it might have been. For example, during the study we did not measure the time PCPs spent on hold or how many reattempts were needed to complete the communication loop. Immediately following the completion of our study, it became apparent that physicians returning calls for our own institution's primary care clinic were experiencing regular workflow interruptions and occasional hold times more than 20 minutes, necessitating ongoing further work to determine the root causes and solutions to these problems. Though this work is ongoing, average PCP hold times measured from a sample of call reviews in 2013 to 2014 was 3 minutes and 15 seconds.

This study has several other limitations. We were unable to account for phone calls to PCPs initiated outside of the new process. It may be that PCPs were called more than 52% of the time at baseline due to noncompliance with the new protocol. Also, we only have data for call completion starting after implementation of the link between the discharge order and the discharge phone call, making the baseline appear artificially high and precluding any analysis of how earlier interventions affected our outcome metric. Communication with PCPs should ideally occur prior to discharge. An important limitation of our process is that calls could occur several hours after discharge between an on‐call resident and an on‐call outpatient physician rather than between the PCP and the discharging resident, limiting appropriate information exchange. Though verbal discharge communication is a desirable goal for many reasons, the current project did not focus on the quality of the call or the information that was transmitted to the PCP. Additionally, direct attending‐to‐attending communication may be valuable with medically or socially complex discharges, but we did not have a process to facilitate this. We also did not measure what effect our new process had on outcomes such as quality of patient and family transition from hospital or physician satisfaction. The existence of programs similar to our PPL subspecialty referral line may be limited to large institutions. However, it should be noted that although some internal resource reallocation was necessary within PPL, no actual staffing increases were required despite a large increase in call volume. It may be that any hospital operator system could be adapted for this purpose with modest additional resources. Finally, although our EHR system is widely utilized, there are many competing systems in the market, and our intervention required utilization of EHR capabilities that may not be present in all systems. However, our EHR intervention utilized existing functionality and did not require modification of the system.

This project focused on discharge phone calls to primary care physicians for patients hospitalized on the hospital medicine service. Because communication with the PCP should ideally occur prior to discharge, future work will include identifying a more proximal trigger than the discharge order to which to link the EHR trigger for discharge communication. Other next steps to improve handoff effectiveness and optimize the efficiency of our process include identifying essential information that should be transmitted to the primary care physician at the time of the phone call, developing processes to ensure communication of this information, measuring PCP satisfaction with this communication, and measuring the impact on patient outcomes. Finally, though expert opinion indicates that verbal handoffs may have safety advantages over nonverbal handoffs, studies comparing the safety and efficacy of verbal versus nonverbal handoffs at patient discharge are lacking. Studies establishing the relative efficacy and safety of verbal versus nonverbal handoffs at hospital discharge are needed. Knowledge gained from these activities could inform future projects centered on the spread of the process to other hospital services and/or other hospitals.

CONCLUSION

We increased the percentage of calls initiated to PCPs at patient discharge from 52% to 97% and the percentage of calls completed between HM physicians and PCPs to 93% through the use of a standardized discharge communication process coupled with a basic EHR messaging functionality. The results of this study may be of interest for further testing and adaptation for any institution with an electronic healthcare system.

Disclosure: Nothing to report.

Timely and reliable communication of important data between hospital‐based physicians and primary care physicians is critical for prevention of medical adverse events.[1, 2] Extrapolation from high‐performance organizations outside of medicine suggests that verbal communication is an important component of patient handoffs.[3, 4] Though the Joint Commission does not mandate verbal communication during handoffs per se, stipulating instead that handoff participants have an opportunity to ask and respond to questions,[5] there is some evidence that primary care providers prefer verbal handoffs at least for certain patients such as those with medical complexity.[6] Verbal communication offers the receiver the opportunity to ask questions, but in practice, 2‐way verbal communication is often difficult to achieve at hospital discharge.

At our institution, hospital medicine (HM) physicians serve as the primary inpatient providers for nearly 90% of all general pediatric admissions. When the HM service was established, primary care physicians (PCPs) and HM physicians together agreed upon an expectation for verbal, physician‐to‐physician communication at the time of discharge. Discharge communication is provided by either residents or attendings depending on the facility. A telephone operator service called Physician Priority Link (PPL) was made available to facilitate this communication. The PPL service is staffed 24/7 by operators whose only responsibilities are to connect providers inside and outside the institution. By utilizing this service, PCPs could respond in a nonemergent fashion to discharge phone calls.

Over the last several years, PCPs have observed high variation in the reliability of discharge communication phone calls. A review of PPL phone records in 2009 showed that only 52% of HM discharges had a record of a call initiated to the PCP on the day of discharge. The overall goal of this improvement project was to improve the completion of verbal handoffs from HM physicians (residents or attendings) to PCPs. The specific aim of the project was to increase the proportion of completed verbal handoffs from on‐call residents or attendings to PCPs within 24 hours of discharge to more than 90% within 18 months.

METHODS

Planning the Intervention

The project was championed by physicians on the HM service and supported by a chief resident, PPL administrators, and 2 information technology analysts.

At the onset of the project, the team mapped the process for completing a discharge call to the PCPs, conducted a modified failure mode and effects analysis,[7, 8] and examined the key drivers used to prioritize interventions (Figure 1). Through the modified failure modes effect analysis, the team was able to identify system issues that led to unsuccessful communication: failure of call initiation, absence of an identified PCP, long wait times on hold, failure of PCP to call back, and failure of the call to be documented. These failure modes informed the key drivers to achieving the study aim. Figure 2 depicts the final key drivers, which were revised through testing and learning.

jhm2392-fig-0001-m.png

jhm2392-fig-0002-m.png

Interventions to Standardization the Communication Process

To facilitate initiation of calls to PCPs at hospital discharge, the improvement team created a standard process using the PPL service (Figure 3). All patients discharged from the HM service were included in the process. Discharging physicians (who were usually but not always residents, depending on the facility), were instructed to call the PPL operator at the time of discharge. The PPL operator would then page the patient's PCP. It was the responsibility of the discharging physician to identify a PCP prior to discharge. Instances where no PCP was identified were counted as process failures because no phone call could be made. The expectation for the PCPs was that they would return the page within 20 minutes. PPL operators would then page back to the discharging physician to connect the 2 parties with the expectation that the discharging physician respond within 2 to 4 minutes to the PPL operator's page. Standardization of all calls through PPL allowed efficient tracking of incomplete calls and operators to reattempt calls that were not completed. This process also shifted the burden of following up on incomplete calls to PPL. The use of PPL to make the connection also allowed the physician to complete other work while awaiting a call back from the PCP.

jhm2392-fig-0003-m.png

RESULTS

From July 2011 to March 2014, there were 6313 discharges from the HM service. The process measure (percentage of calls initiated) improved from 50% to 97% after 4 interventions (Figure 4). Data for the outcome measure (percentage of calls completed) were collected starting in August 2012, shortly after linking the EHR discharge order to the discharge call. Over the first 8 weeks, our median was 80%, which increased to a median of 93% (Figure 5). These results were sustained for 18 months.

jhm2392-fig-0004-m.png

jhm2392-fig-0005-m.png

Several key interventions were identified that were critical to achievement of our goal. Standardization of the communication process through PPL was temporally associated with a shift in the median rate of call initiation from 52% to 72%. Use of the discharge order to initiate discharge communication was associated with an increase from 72% to 97%. Finally, the percentage of completed verbal handoffs increased to more than 93% following batching of phone calls to PCPs at specific times during the day.

DISCUSSION

We used improvement and reliability science methods to implement a successful process for improving verbal handoffs from HM physicians to PCPs within 24 hours of discharge to 93%. This result has been sustained for 18 months.

Utilization of the PPL call center for flexible call facilitation along with support for data analysis and leveraging the EHR to automate the process increased reliability, leading to rapid improvement. Prior to mandating the use of PPL to connect discharging physicians with PCPs, the exact rate of successful handoffs in our institution was not known. We do know, however, that only 52% of calls were initiated, so clearly a large gap was present prior to our improvement work. Data collection from the PPL system was automated so that accurate, timely, and sustainable data could be provided, greatly aiding improvement efforts. Flexibility in call‐back timing was also crucial, because coordinating the availability of PCPs and discharging physicians is often challenging. The EHR‐initiated process for discharge communication was a key intervention, and improvement of our process measure to 97% performance was associated with this implementation. Two final interventions: (1) assignment of responsibility for communication to a team pager held by a designated resident and (2) batching of calls to specific times streamlined the EHR‐initiated process and were associated with achievement of our main outcome goal of >90% completed verbal communication.

There are several reports of successful interventions to improve receipt or content of discharge summaries by PCPs following hospital discharge available in the literature.[14, 15, 16, 17, 18, 19, 20] Recently, Shen et al. reported on the success of a multisite improvement collaborative involving pediatric hospitalist programs at community hospitals whose aim was to improve the timely documentation of communication directed at PCPs.[21] In their report, all 7 hospital sites that participated in the collaborative for more than 4 months were able to demonstrate substantial improvement in documentation of some form of communication directed at PCPs (whether by e‐mail, fax, or telephone call), from a baseline of approximately 50% to more than 90%. A limitation of their study was that they were unable to document whether PCPs had received any information or by what method. A recent survey of PCPs by Sheu et al. indicated that for many discharges, information in addition to that present in the EHR was desirable to ensure a safe transition of care.[6] Two‐way communication, such as with a phone call, allows for senders to verify information receipt and for receivers to ask questions to ensure complete information. To our knowledge, there have been no previous reports describing processes for improving verbal communication between hospitalist services and PCPs at discharge.

It may be that use of the call system allowed PCPs to return phone calls regarding discharges at convenient stopping points in their day while allowing discharging physicians to initiate a call without having to wait on hold. Interestingly, though we anticipated the need for additional PPL resources during the course of this improvement, the final process was efficient enough that PPL did not require additional staffing to accommodate the higher call volume.

A key insight during our implementation was that relying on the EHR to initiate every discharge communication created disruption of resident workflow due to disregard of patient, resident, and PCP factors. This was reflected by the improvement in call initiation (our process measure) following this intervention, whereas at the same time call completion (our outcome measure) remained below goal. To achieve our goal of completing verbal communication required a process that was highly reliable yet flexible enough to allow discharging physicians to complete the call in the unpredictable environment of inpatient care. Ultimately, this was achieved by allowing discharging physicians to initiate the process when convenient, and allowing for the EHR‐initiated process to function as a backup strategy to identify and mitigate failures of initiation.

An important limitation of our study was the lack of PCPs on the improvement team, likely making the success of the project more difficult than it might have been. For example, during the study we did not measure the time PCPs spent on hold or how many reattempts were needed to complete the communication loop. Immediately following the completion of our study, it became apparent that physicians returning calls for our own institution's primary care clinic were experiencing regular workflow interruptions and occasional hold times more than 20 minutes, necessitating ongoing further work to determine the root causes and solutions to these problems. Though this work is ongoing, average PCP hold times measured from a sample of call reviews in 2013 to 2014 was 3 minutes and 15 seconds.

This study has several other limitations. We were unable to account for phone calls to PCPs initiated outside of the new process. It may be that PCPs were called more than 52% of the time at baseline due to noncompliance with the new protocol. Also, we only have data for call completion starting after implementation of the link between the discharge order and the discharge phone call, making the baseline appear artificially high and precluding any analysis of how earlier interventions affected our outcome metric. Communication with PCPs should ideally occur prior to discharge. An important limitation of our process is that calls could occur several hours after discharge between an on‐call resident and an on‐call outpatient physician rather than between the PCP and the discharging resident, limiting appropriate information exchange. Though verbal discharge communication is a desirable goal for many reasons, the current project did not focus on the quality of the call or the information that was transmitted to the PCP. Additionally, direct attending‐to‐attending communication may be valuable with medically or socially complex discharges, but we did not have a process to facilitate this. We also did not measure what effect our new process had on outcomes such as quality of patient and family transition from hospital or physician satisfaction. The existence of programs similar to our PPL subspecialty referral line may be limited to large institutions. However, it should be noted that although some internal resource reallocation was necessary within PPL, no actual staffing increases were required despite a large increase in call volume. It may be that any hospital operator system could be adapted for this purpose with modest additional resources. Finally, although our EHR system is widely utilized, there are many competing systems in the market, and our intervention required utilization of EHR capabilities that may not be present in all systems. However, our EHR intervention utilized existing functionality and did not require modification of the system.

This project focused on discharge phone calls to primary care physicians for patients hospitalized on the hospital medicine service. Because communication with the PCP should ideally occur prior to discharge, future work will include identifying a more proximal trigger than the discharge order to which to link the EHR trigger for discharge communication. Other next steps to improve handoff effectiveness and optimize the efficiency of our process include identifying essential information that should be transmitted to the primary care physician at the time of the phone call, developing processes to ensure communication of this information, measuring PCP satisfaction with this communication, and measuring the impact on patient outcomes. Finally, though expert opinion indicates that verbal handoffs may have safety advantages over nonverbal handoffs, studies comparing the safety and efficacy of verbal versus nonverbal handoffs at patient discharge are lacking. Studies establishing the relative efficacy and safety of verbal versus nonverbal handoffs at hospital discharge are needed. Knowledge gained from these activities could inform future projects centered on the spread of the process to other hospital services and/or other hospitals.

CONCLUSION

We increased the percentage of calls initiated to PCPs at patient discharge from 52% to 97% and the percentage of calls completed between HM physicians and PCPs to 93% through the use of a standardized discharge communication process coupled with a basic EHR messaging functionality. The results of this study may be of interest for further testing and adaptation for any institution with an electronic healthcare system.

Disclosure: Nothing to report.