User login
SEATTLE – Kaiser Permanente Northern California has developed an EHR program that automatically flags patients at high risk for HIV.
The idea was to come up with a way to help clinicians focus their pre-exposure prophylaxis (PrEP) outreach on the people who need it most. PrEP prevents HIV, but often “it’s difficult for providers to identify patients who are at risk. Prediction models using EHR data can identify patients who are at high risk but not using PrEP,” said study lead Julia Marcus, PhD, an assistant professor in the department of population medicine at Harvard Medical School, Boston.
She proved that assertion in a presentation at the Conference on Retroviruses & Opportunistic Infections.
The Kaiser program uses 44 variables routinely collected in EHRs from five categories: demographics, social history, lab data, medication use, and diagnoses. Specific variables include living in a zip code with high HIV incidence; men who have sex with men (MSM); black race; urine tests positive for cocaine or methadone; use of erectile dysfunction medications; and diagnoses of depression, anal warts, and other conditions.
The development cohort included 3,143,963 Kaiser members from 2007 to 2014 with 2 or more years of enrollment; at least one outpatient visit; no prior PrEP use; and no HIV diagnosis.
There were 701 incident HIV cases; the model did a good job at predicting them, with a C-statistic of 0.86 (95% confidence interval, 0.85-0.87). A score of 1.0 would be perfect prediction, and 0.5 no predictive value. Previous efforts at HIV prediction – relying generally on just MSM status and STD history – have C-statistics of around 0.6; prediction models commonly used for cardiovascular and other diseases often have C-statistics of around 0.7, Dr. Marcus explained.
The model was validated in 606,701 Kaiser members during 2015-2017. The validation cohort was slightly younger than the development cohort, with a mean age of 37 versus 45 years, and slightly more diverse, with fewer white patients, 44% versus 52%. Both cohorts had slightly more women than men.
There were 83 new HIV diagnoses in the validation cohort. The C-statistic for HIV prediction was 0.84 (95% CI, 0.8-0.89). The model predicted 32 of 69 (46%) incident HIV cases among men tagged as high risk – at least a 0.2% chance of contracting HIV within 3 years – or very high risk, a 1% chance or higher, which is more than 50 times the risk among the general population. Relying on just MSM and STD history predicted 32% of cases.
Overall, “our model identified nearly half of new HIV cases among males by flagging only 2% of the general population. The results suggest our model would perform well if implemented today. You could replicate our approach in any health system with an EHR. Our specific variables may not translate to every setting, but any health care system can develop this model based on the EHR data they do have,” Dr. Marcus said.
“You could embed this in any EHR system and have it update in real time to flag providers to do a sexual history and talk with patients about PrEP,” she said.
The next step is a pilot project at Kaiser Permanente San Francisco to evaluate the impact on PrEP prescribing and HIV incidence. The model failed to predict 14 incident HIV cases among women in the validation cohort, a problem that also needs to be addressed.
The work was funded by the National Institutes of Health and Kaiser Permanente. Dr. Marcus didn’t have any relevant disclosures.
SOURCE: Marcus JL et al. CROI 2019, Abstract 105.
SEATTLE – Kaiser Permanente Northern California has developed an EHR program that automatically flags patients at high risk for HIV.
The idea was to come up with a way to help clinicians focus their pre-exposure prophylaxis (PrEP) outreach on the people who need it most. PrEP prevents HIV, but often “it’s difficult for providers to identify patients who are at risk. Prediction models using EHR data can identify patients who are at high risk but not using PrEP,” said study lead Julia Marcus, PhD, an assistant professor in the department of population medicine at Harvard Medical School, Boston.
She proved that assertion in a presentation at the Conference on Retroviruses & Opportunistic Infections.
The Kaiser program uses 44 variables routinely collected in EHRs from five categories: demographics, social history, lab data, medication use, and diagnoses. Specific variables include living in a zip code with high HIV incidence; men who have sex with men (MSM); black race; urine tests positive for cocaine or methadone; use of erectile dysfunction medications; and diagnoses of depression, anal warts, and other conditions.
The development cohort included 3,143,963 Kaiser members from 2007 to 2014 with 2 or more years of enrollment; at least one outpatient visit; no prior PrEP use; and no HIV diagnosis.
There were 701 incident HIV cases; the model did a good job at predicting them, with a C-statistic of 0.86 (95% confidence interval, 0.85-0.87). A score of 1.0 would be perfect prediction, and 0.5 no predictive value. Previous efforts at HIV prediction – relying generally on just MSM status and STD history – have C-statistics of around 0.6; prediction models commonly used for cardiovascular and other diseases often have C-statistics of around 0.7, Dr. Marcus explained.
The model was validated in 606,701 Kaiser members during 2015-2017. The validation cohort was slightly younger than the development cohort, with a mean age of 37 versus 45 years, and slightly more diverse, with fewer white patients, 44% versus 52%. Both cohorts had slightly more women than men.
There were 83 new HIV diagnoses in the validation cohort. The C-statistic for HIV prediction was 0.84 (95% CI, 0.8-0.89). The model predicted 32 of 69 (46%) incident HIV cases among men tagged as high risk – at least a 0.2% chance of contracting HIV within 3 years – or very high risk, a 1% chance or higher, which is more than 50 times the risk among the general population. Relying on just MSM and STD history predicted 32% of cases.
Overall, “our model identified nearly half of new HIV cases among males by flagging only 2% of the general population. The results suggest our model would perform well if implemented today. You could replicate our approach in any health system with an EHR. Our specific variables may not translate to every setting, but any health care system can develop this model based on the EHR data they do have,” Dr. Marcus said.
“You could embed this in any EHR system and have it update in real time to flag providers to do a sexual history and talk with patients about PrEP,” she said.
The next step is a pilot project at Kaiser Permanente San Francisco to evaluate the impact on PrEP prescribing and HIV incidence. The model failed to predict 14 incident HIV cases among women in the validation cohort, a problem that also needs to be addressed.
The work was funded by the National Institutes of Health and Kaiser Permanente. Dr. Marcus didn’t have any relevant disclosures.
SOURCE: Marcus JL et al. CROI 2019, Abstract 105.
SEATTLE – Kaiser Permanente Northern California has developed an EHR program that automatically flags patients at high risk for HIV.
The idea was to come up with a way to help clinicians focus their pre-exposure prophylaxis (PrEP) outreach on the people who need it most. PrEP prevents HIV, but often “it’s difficult for providers to identify patients who are at risk. Prediction models using EHR data can identify patients who are at high risk but not using PrEP,” said study lead Julia Marcus, PhD, an assistant professor in the department of population medicine at Harvard Medical School, Boston.
She proved that assertion in a presentation at the Conference on Retroviruses & Opportunistic Infections.
The Kaiser program uses 44 variables routinely collected in EHRs from five categories: demographics, social history, lab data, medication use, and diagnoses. Specific variables include living in a zip code with high HIV incidence; men who have sex with men (MSM); black race; urine tests positive for cocaine or methadone; use of erectile dysfunction medications; and diagnoses of depression, anal warts, and other conditions.
The development cohort included 3,143,963 Kaiser members from 2007 to 2014 with 2 or more years of enrollment; at least one outpatient visit; no prior PrEP use; and no HIV diagnosis.
There were 701 incident HIV cases; the model did a good job at predicting them, with a C-statistic of 0.86 (95% confidence interval, 0.85-0.87). A score of 1.0 would be perfect prediction, and 0.5 no predictive value. Previous efforts at HIV prediction – relying generally on just MSM status and STD history – have C-statistics of around 0.6; prediction models commonly used for cardiovascular and other diseases often have C-statistics of around 0.7, Dr. Marcus explained.
The model was validated in 606,701 Kaiser members during 2015-2017. The validation cohort was slightly younger than the development cohort, with a mean age of 37 versus 45 years, and slightly more diverse, with fewer white patients, 44% versus 52%. Both cohorts had slightly more women than men.
There were 83 new HIV diagnoses in the validation cohort. The C-statistic for HIV prediction was 0.84 (95% CI, 0.8-0.89). The model predicted 32 of 69 (46%) incident HIV cases among men tagged as high risk – at least a 0.2% chance of contracting HIV within 3 years – or very high risk, a 1% chance or higher, which is more than 50 times the risk among the general population. Relying on just MSM and STD history predicted 32% of cases.
Overall, “our model identified nearly half of new HIV cases among males by flagging only 2% of the general population. The results suggest our model would perform well if implemented today. You could replicate our approach in any health system with an EHR. Our specific variables may not translate to every setting, but any health care system can develop this model based on the EHR data they do have,” Dr. Marcus said.
“You could embed this in any EHR system and have it update in real time to flag providers to do a sexual history and talk with patients about PrEP,” she said.
The next step is a pilot project at Kaiser Permanente San Francisco to evaluate the impact on PrEP prescribing and HIV incidence. The model failed to predict 14 incident HIV cases among women in the validation cohort, a problem that also needs to be addressed.
The work was funded by the National Institutes of Health and Kaiser Permanente. Dr. Marcus didn’t have any relevant disclosures.
SOURCE: Marcus JL et al. CROI 2019, Abstract 105.
REPORTING FROM CROI 2019