Assessing the Impact of Body Mass Index Information on the Performance of Risk Adjustment Models in Predicting Health Care Costs and Utilization

Assessing the Added Value of Vital Signs Extracted from Electronic Health Records in Healthcare Risk Adjustment Models

Purpose

Patient vital signs are related to specific health risks and outcomes but are underutilized in the prediction of health-care utilization and cost. To measure the added value of electronic health record (EHR) extracted Body Mass Index (BMI) and blood pressure (BP) values in improving healthcare risk and utilization predictions.

Patients and Methods

A sample of 12,820 adult outpatients from the Johns Hopkins Health System (JHHS) were identified between 2016 and 2017, having high data quality and recorded values for BMI and BP. We evaluated the added value of BMI and BP in predicting health-care utilization and cost through a retrospective cohort design. BMI, mean arterial pressure (MAP), systolic and diastolic BPs were summarized as annual aggregated values. Concurrent annual BMI and MAP changes were quantified as the difference between maximum and minimum recorded values. Model performance estimates consisted of repeated 10-fold cross validation, compared to base model point estimates for demographic and diagnostic, coded events: (1) patient age and sex, (2) age, sex, and the Charlson weighted index, (3) age, sex and the Johns Hopkins ACG system’s DxPM risk score.

Results

Both categorical BMI and BP were progressively indicative of disease comorbidity, but not uniformly related to health-care utilization or cost. Annual change in BMI and MAP improved predictions for most concurrent year outcomes when compared to base models.

Conclusion

When a healthcare system lacks relevant diagnostic or risk assessment information for a patient, vital signs may be useful for a simple estimation of disease risk, cost and utilization.

Improving the Prediction of Persistent High Healthcare Utilizers: Using an Ensemble Methodology

Background

A small proportion of high-need patients persistently use the bulk of health care services and incur disproportionate costs. Population health management (PHM) programs often refer to these patients as persistent high utilizers (PHUs). Accurate PHU prediction enables PHM programs to better align scarce health care resources with high-need PHUs while generally improving outcomes. While prior research in PHU prediction has shown promise, traditional regression methods used in these studies have yielded limited accuracy.

Objective

We are seeking to improve PHU predictions with an ensemble approach in a retrospective observational study design using insurance claim records.

Methods

We defined a PHU as a patient with health care costs in the top 20% of all patients for 4 consecutive 6-month periods. We used 2013 claims data to predict PHU status in next 24 months. Our study population included 165,595 patients in the Johns Hopkins Health Care plan, with 8359 (5.1%) patients identified as PHUs in 2014 and 2015. We assessed the performance of several standalone machine learning methods and then an ensemble approach combining multiple models.

Results

The candidate ensemble with complement naïve Bayes and random forest layers produced increased sensitivity and positive predictive value (PPV; 49.0% and 50.3%, respectively) compared to logistic regression (46.8% and 46.1%, respectively).

Conclusions

Our results suggest that ensemble machine learning can improve prediction of care management needs. Improved PPV implies reduced incorrect referral of low-risk patients. With the improved sensitivity/PPV balance of this approach, resources may be directed more efficiently to patients needing them most.

Estimation and Prediction of Hospitalization and Medical Care Costs Using Regression in Machine Learning

Medical costs are one of the most common recurring expenses in a person’s life. Based on different research studies, BMI, ageing, smoking, and other factors are all related to greater personal medical care costs. The estimates of the expenditures of health care related to obesity are needed to help create cost-effective obesity prevention strategies. Obesity prevention at a young age is a top concern in global health, clinical practice, and public health. To avoid these restrictions, genetic variants are employed as instrumental variables in this research. Using statistics from public huge datasets, the impact of body mass index (BMI) on overall healthcare expenses is predicted. A multiview learning architecture can be used to leverage BMI information in records, including diagnostic texts, diagnostic IDs, and patient traits. A hierarchy perception structure was suggested to choose significant words, health checks, and diagnoses for training phase informative data representations, because various words, diagnoses, and previous health care have varying significance for expense calculation. In this system model, linear regression analysis, naive Bayes classifier, and random forest algorithms were compared using a business analytic method that applied statistical and machine-learning approaches. According to the results of our forecasting method, linear regression has the maximum accuracy of 97.89 percent in forecasting overall healthcare costs. In terms of financial statistics, our methodology provides a predictive method.

Assessing the Added Value of Blood Pressure Information Derived from Electronic Health Records in Predicting Health Care Cost and Utilization

Health care providers are increasingly using clinical measures derived from electronic health records (EHRs) for risk stratification and predictive modeling. EHR-specific data elements such as prescriptions, laboratory results, and vital signs have been shown to improve risk prediction models. In this study, the value of EHR-based blood pressure (BP) values was assessed in predicting health care costs (ie, total, medical, and pharmacy) and key utilization end points (ie, hospitalization, emergency department use, and being among the highest utilizers). The study population included 37,451 patients of a large integrated delivery system in the mid-western United States with complete EHR data files, who were 18-64 years old, had continuous insurance at an affiliated health plan, and had eligible BP records. Both EHRs and insurance claims of the study population were used to extract the predictors (ie, demographics, diagnosis, and BP values) and outcomes (ie, costs and utilizations). Predictors were extracted from 2012 data, whereas concurrent and prospective outcomes were extracted from 2012 to 2013 data. Three base models (BMs) were constructed to predict each of the outcomes. The first BM no. 1 used demographics. The second BM no. 2 added the Charlson comorbidity index to BM no. 1, whereas the third BM no. 3 added the Adjusted Clinical Group Dx-PM case-mix score to BM no. 1. BP was specified as means, ranges, and classes. Adding BP ranges to BM no. 1 and BM no. 2 showed the greatest improvements when predicting costs and utilization. More specifically, adjusted R² and area under the curve of BM no. 2 improved by 32.9% and 14.1% when BP ranges were added to predict concurrent total cost and hospitalization, respectively. The effect of BP measures on improving the risk stratification models was diminished when predicting prospective outcomes after adding the measures to BM no. 3 (ie, the more comprehensive diagnostic model), specifically when represented as BP means. Given the increasing availability of BP information, this research suggests that these data should be integrated into provider-based population health analytic activities. Future research should focus on subpopulations that benefit the most from incorporating vital signs such as BP measures in risk stratification models.

Assessing the Value of Unsupervised Clustering in Predicting Persistent High Health Care Utilizers: Retrospective Analysis of Insurance Claims Data

BACKGROUND

A high proportion of health care services are persistently utilized by a small subpopulation of patients. To improve clinical outcomes while reducing costs and utilization, population health management programs often provide targeted interventions to patients who may become persistent high users/utilizers (PHUs). Enhanced prediction and management of PHUs can improve health care system efficiencies and improve the overall quality of patient care.

OBJECTIVE

The aim of this study was to detect key classes of diseases and medications among the study population and to assess the predictive value of these classes in identifying PHUs.

METHODS

This study was a retrospective analysis of insurance claims data of patients from the Johns Hopkins Health Care system. We defined a PHU as a patient incurring health care costs in the top 20% of all patients' costs for 4 consecutive 6-month periods. We used 2013 claims data to predict PHU status in 2014-2015. We applied latent class analysis (LCA), an unsupervised clustering approach, to identify patient subgroups with similar diagnostic and medication patterns to differentiate variations in health care utilization across PHUs. Logistic regression models were then built to predict PHUs in the full population and in select subpopulations. Predictors included LCA membership probabilities, demographic covariates, and health utilization covariates. Predictive powers of the regression models were assessed and compared using standard metrics.

RESULTS

We identified 164,221 patients with continuous enrollment between 2013 and 2015. The mean study population age was 19.7 years, 55.9% were women, 3.3% had ≥1 hospitalization, and 19.1% had 10+ outpatient visits in 2013. A total of 8359 (5.09%) patients were identified as PHUs in both 2014 and 2015. The LCA performed optimally when assigning patients to four probability disease/medication classes. Given the feedback provided by clinical experts, we further divided the population into four diagnostic groups for sensitivity analysis: acute upper respiratory infection (URI) (n=53,232; 4.6% PHUs), mental health (n=34,456; 12.8% PHUs), otitis media (n=24,992; 4.5% PHUs), and musculoskeletal (n=24,799; 15.5% PHUs). For the regression models predicting PHUs in the full population, the F1-score classification metric was lower using a parsimonious model that included LCA categories (F1=38.62%) compared to that of a complex risk stratification model with a full set of predictors (F1=48.20%). However, the LCA-enabled simple models were comparable to the complex model when predicting PHUs in the mental health and musculoskeletal subpopulations (F1-scores of 48.69% and 48.15%, respectively). F1-scores were lower than that of the complex model when the LCA-enabled models were limited to the otitis media and acute URI subpopulations (45.77% and 43.05%, respectively).

CONCLUSIONS

Our study illustrates the value of LCA in identifying subgroups of patients with similar patterns of diagnoses and medications. Our results show that LCA-derived classes can simplify predictive models of PHUs without compromising predictive accuracy. Future studies should investigate the value of LCA-derived classes for predicting PHUs in other health care settings.

HIV-Prevalence Mapping Using Small Area Estimation in Kenya, Tanzania, and Mozambique at the First Sub-National Level

Background:

Local estimates of HIV-prevalence provide information that can be used to target interventions and consequently increase the efficiency of resources. This enhanced allocation can lead to better health outcomes, including the control of the disease spread, and for more people.

Methods:

In this study, we used the DHS data phase V to estimate HIV prevalence at the first-subnational level in Kenya, Tanzania, and Mozambique. We fitted the data to a spatial random effect intrinsic conditional autoregressive (ICAR) model to smooth the outcome. Further, we used a sampling specification from a multistage cluster design.

Results:

We found that Nyanza (P_i = 13.6%) and Nairobi (P_i = 7.1%) in Kenya, Iringa (P_i = 15.4%) and Mbeya (P_i = 9.3%) in Tanzania, and Gaza (P_i = 15.2%) and Maputo City (P_i = 12.9%) in Mozambique are the regions with the highest prevalence of HIV, within country. Our results are based on publicly available data that through statistically rigorous methods, allowed us to obtain an accurate visual representation of the HIV prevalence at a regional level.

Conclusions:

These results can help in identification and targeting of high-prevalent regions to increase the supply of healthcare services to reduce the spread of the disease and increase the health quality of people living with HIV.

Internet Access, Social Risk Factors, and Web-Based Social Support Seeking Behavior: Assessing Correlates of the "Digital Divide" Across Neighborhoods in The State of Maryland

We aimed to empirically measure the degree to which there is a “digital divide” in terms of access to the internet at the small-area community level within the State of Maryland and the City of Baltimore and to assess the relationship and association of this divide with community-level SDOH risk factors, community-based social service agency location, and web-mediated support service seeking behavior. To assess the socio-economic characteristics of the neighborhoods across the state, we calculated the Area Deprivation Index (ADI) using the U.S. Census, American Community Survey (5-year estimates) of 2017. To assess the digital divide, at the community level, we used the Federal Communications Commission (FCC) data on the number of residential fixed Internet access service connections. We assessed the availability of and web-based access to community-based social service agencies using data provided by the “Aunt Bertha” information platform. We performed community and regional level descriptive and special analyses for ADI social risk factors, connectivity, and both the availability of and web-based searches for community-based social services. To help assess potential neighborhood linked factors associated with the rates of web-based social services searches by individuals in need, we applied logistic regression using generalized estimating equation modeling. Baltimore City contained more disadvantaged neighborhoods compared to other areas in Maryland. In Baltimore City, 20.3% of neighborhoods (defined by census block groups) were disadvantaged with ADI at the 90th percentile while only 6.6% of block groups across Maryland were in this disadvantaged category. Across the State, more than half of all census tracts had 801–1000 households (per 1000 households) with internet subscription. In contrast, in Baltimore City about half of all census tracts had only 401–600 of the households (per 1000 households) with internet subscriptions. Most block groups in Maryland and Baltimore City lacked access to social services facilities (61% of block groups at the 90th percentile of disadvantage in Maryland and 61.3% of block groups at the 90th percentile of disadvantage in Baltimore City). After adjusting for other variables, a 1% increase in the ADI measure of social disadvantage, resulting in a 1.7% increase in the number of individuals seeking social services. While more work is needed, our findings support the premise that the digital divide is closely associated with other SDOH factors. The policymakers must propose policies to address the digital divide on a national level and also in disadvantaged communities experiencing the digital divide in addition to other SDOH challenges.

Measuring the Value of a Practical Text Mining Approach to Identify Patients With Housing Issues in the Free-Text Notes in Electronic Health Record: Findings of a Retrospective Cohort Study

Introduction: Despite the growing efforts to standardize coding for social determinants of health (SDOH), they are infrequently captured in electronic health records (EHRs). Most SDOH variables are still captured in the unstructured fields (i.e., free-text) of EHRs. In this study we attempt to evaluate a practical text mining approach (i.e., advanced pattern matching techniques) in identifying phrases referring to housing issues, an important SDOH domain affecting value-based healthcare providers, using EHR of a large multispecialty medical group in the New England region, United States. To present how this approach would help the health systems to address the SDOH challenges of their patients we assess the demographic and clinical characteristics of patients with and without housing issues and briefly look into the patterns of healthcare utilization among the study population and for those with and without housing challenges.

Methods: We identified five categories of housing issues [i.e., homelessness current (HC), homelessness history (HH), homelessness addressed (HA), housing instability (HI), and building quality (BQ)] and developed several phrases addressing each one through collaboration with SDOH experts, consulting the literature, and reviewing existing coding standards. We developed pattern-matching algorithms (i.e., advanced regular expressions), and then applied them in the selected EHR. We assessed the text mining approach for recall (sensitivity) and precision (positive predictive value) after comparing the identified phrases with manually annotated free-text for different housing issues.

Results: The study dataset included EHR structured data for a total of 20,342 patients and 2,564,344 free-text clinical notes. The mean (SD) age in the study population was 75.96 (7.51). Additionally, 58.78% of the cohort were female. BQ and HI were the most frequent housing issues documented in EHR free-text notes and HH was the least frequent one. The regular expression methodology, when compared to manual annotation, had a high level of precision (positive predictive value) at phrase, note, and patient levels (96.36, 95.00, and 94.44%, respectively) across different categories of housing issues, but the recall (sensitivity) rate was relatively low (30.11, 32.20, and 41.46%, respectively).

Conclusion: Results of this study can be used to advance the research in this domain, to assess the potential value of EHR's free-text in identifying patients with a high risk of housing issues, to improve patient care and outcomes, and to eventually mitigate socioeconomic disparities across individuals and communities.

Comparing the Predictive Effects of Patient Medication Adherence Indices in Electronic Health Record and Claims-Based Risk Stratification Models

Multiple indices are available to measure medication adherence behaviors. Medication adherence measures, however, have rarely been extracted from electronic health records (EHRs) for population-level risk predictions. This study assessed the value of medication adherence indices in improving predictive models of cost and hospitalization. This study included a 2-year retrospective cohort of patients younger than age 65 years with linked EHR and insurance claims data. Three medication adherence measures were calculated: medication regimen complexity index (MRCI), medication possession ratio (MPR), and prescription fill rate (PFR). The authors examined the effects of adding these measures to 3 predictive models of utilization: a demographics model, a conventional model (Charlson index), and an advanced diagnosis-based model. Models were trained using EHR and claims data. The study population had an overall MRCI, MPR, and PFR of 14.6 ± 17.8, .624 ± .310, and .810 ± .270, respectively. Adding MRCI and MPR to the demographic and the morbidity models using claims data improved forecasting of next-year hospitalization substantially (eg, AUC of the demographic model increased from .605 to .656 using MRCI). Nonetheless, such boosting effects were attenuated for the advanced diagnosis-based models. Although EHR models performed inferior to claims models, adding adherence indices improved EHR model performances at a larger scale (eg, adding MRCI increased AUC by 4.4% for the Charlson model using EHR data compared to 3.8% using claims). This study shows that medication adherence measures can modestly improve EHR- and claims-derived predictive models of cost and hospitalization in non-elderly patients; however, the improvements are minimal for advanced diagnosis-based models.

Identifying vulnerable older adult populations by contextualizing geriatric syndrome information in clinical notes of electronic health records

OBJECTIVE

Geriatric syndromes such as functional disability and lack of social support are often not encoded in electronic health records (EHRs), thus obscuring the identification of vulnerable older adults in need of additional medical and social services. In this study, we automatically identify vulnerable older adult patients with geriatric syndrome based on clinical notes extracted from an EHR system, and demonstrate how contextual information can improve the process.

MATERIALS AND METHODS

We propose a novel end-to-end neural architecture to identify sentences that contain geriatric syndromes. Our model learns a representation of the sentence and augments it with contextual information: surrounding sentences, the entire clinical document, and the diagnosis codes associated with the document. We trained our system on annotated notes from 85 patients, tuned the model on another 50 patients, and evaluated its performance on the rest, 50 patients.

RESULTS

Contextual information improved classification, with the most effective context coming from the surrounding sentences. At sentence level, our best performing model achieved a micro-F1 of 0.605, significantly outperforming context-free baselines. At patient level, our best model achieved a micro-F1 of 0.843.

DISCUSSION

Our solution can be used to expand the identification of vulnerable older adults with geriatric syndromes. Since functional and social factors are often not captured by diagnosis codes in EHRs, the automatic identification of the geriatric syndrome can reduce disparities by ensuring consistent care across the older adult population.

CONCLUSION

EHR free-text can be used to identify vulnerable older adults with a range of geriatric syndromes.

Integrating E-Prescribing and Pharmacy Claims Data for Predictive Modeling: Comparing Costs and Utilization of Health Plan Members Who Fill Their Initial Medications with Those Who Do Not

BACKGROUND

Nonfilling of prescribed medications is a worldwide problem of serious concern. Studies of health care costs and utilization associated with medication nonadherence frequently rely on claims data and usually focus on patients with specific conditions. Past studies also have little agreement on whether higher medication costs associated with higher adherence can reduce downstream health care consumption.

OBJECTIVES

To (a) compare the characteristics between people with and without complete medication initiations from a general population and (b) quantify the effect of medication initiation on health care utilization and expenditures with propensity score weighting.

METHODS

We conducted a retrospective cohort study using 2012 and 2013 electronic health records (EHR) and insurance claims data from an integrated health care delivery network. We included 43,097 eligible primary care patients in the study. Annual medication fill rates of initial prescriptions in 2012 were defined as the number of filled prescriptions from claims divided by the number of e-prescriptions from EHRs, while excluding all refills. A claim was considered filled if (a) EHR and claims records were from the same drug class; (b) claims occurred between the date of a current EHR order and that of the next EHR order of the same class; and (c) the maximum fill rate was 100%. The 6 annual outcomes included total costs, medical costs, pharmacy costs, being a high-cost "outlier" (in top 5%), having 1 or more hospitalizations, and having 1 or more emergency department (ED) visits. Individuals were classified as either having completed all medication initiations (100% annual filling rate for initiations) or not. We used propensity score weighting to control for baseline differences between complete and incomplete initial fillers. We adopted linear and logistic regressions to model costs and binary utilization indicators for the same year (concurrently) and next year (prospectively).

RESULTS

Approximately 42% of the study sample had complete medication initiations (100% filling rate), while the remaining 58% had incomplete initiations. Individuals who fully filled initial prescriptions had lower comorbidity burden and consumed fewer health care resources. After applying propensity score weighting and controlling for variables such as the number of prescription orders, patients with complete medication initiations had lower overall and medical costs, concurrently and prospectively (e.g., $751 and $252 less for annual total costs). Complete medication initiation fillers were also less likely to have concurrent health care utilization (OR = 0.78, 95% CI = 0.68-0.90 for hospitalization; OR = 0.77, 95% CI = 0.72-0.82 for ED admissions) but no difference in prospective utilization other than for ED visits (OR = 0.93, 95% CI = 0.87-0.99).

CONCLUSIONS

Identifying the subpopulation of patients with incomplete medication initiations (i.e., filling less than 100% of initial prescriptions) is a pragmatic approach for population health management programs to align resources and potentially contain cost and utilization.

DISCLOSURES

No outside funding supported this study. This study applied the Adjusted Clinical Group (ACG) case-mix/risk adjustment methodology, developed at Johns Hopkins Bloomberg School of Public Health. Although ACGs are an important aspect of this study, the goal of the study was not to directly assess or evaluate the methodology. The Johns Hopkins University receives royalties for nonacademic use of software based on the ACG methodology. Chang, Kharrazi, and Weiner receive a portion of their salary support from this revenue. Chang is also a part-time consultant for Monument Analytics, a health care consultancy whose clients include the life sciences industry, as well as plaintiffs in opioid litigation. Alexander is past Chair of FDA's Peripheral and Central Nervous System Advisory Committee; has served as a paid advisor to IQVIA; is a co-founding Principal and equity holder in Monument Analytics; and is a member of OptumRx's National P&T Committee. These arrangements have been reviewed and approved by Johns Hopkins University in accordance with its conflict of interest policies. The other authors have nothing to disclose.

A method for measuring the effect of certified electronic health record technology on childhood immunization status scores among Medicaid managed care network providers

OBJECTIVE

To provide a methodology for estimating the effect of U.S.-based Certified Electronic Health Records Technology (CEHRT) implemented by primary care physicians (PCPs) on a Healthcare Effectiveness Data and Information Set (HEDIS) measure for childhood immunization delivery.

MATERIALS AND METHODS

This study integrates multiple health care administrative data sources from 2010 through 2014, analyzed through an interrupted time series design and a hierarchical Bayesian model. We compared managed care physicians using CEHRT to propensity-score matched comparisons from network physicians who did not adopt CEHRT. Inclusion criteria for physicians using CEHRT included attesting to the Childhood Immunization Status clinical quality measure in addition to meeting "Meaningful Use" (MU) during calendar year 2013. We used a first-presence patient attribution approach to develop provider-specific immunization scores.

RESULTS

We evaluated 147 providers using CEHRT, with 147 propensity-score matched providers selected from a pool of 1253 PCPs practicing in Maryland. The estimate for change in odds of increasing immunization rates due to CEHRT was 1.2 (95% credible set, 0.88-1.73).

DISCUSSION

We created a method for estimating immunization quality scores using Bayesian modeling. Our approach required linking separate administrative data sets, constructing a propensity-score matched cohort, and using first-presence, claims-based childhood visit information for patient attribution. In the absence of integrated data sets and precise and accurate patient attribution, this is a reusable method for researchers and health system administrators to estimate the impact of health information technology on individual, provider-level, process-based, though outcomes-focused, quality measures.

CONCLUSION

This research has provided evidence for using Bayesian analysis of propensity-score matched provider populations to estimate the impact of CEHRT on outcomes-based quality measures such as childhood immunization delivery.

Impact of Area Deprivation Index on the Performance of Claims-Based Risk-Adjustment Models in Predicting Health Care Costs and Utilization

Traditionally, risk-adjustment models do not address the characteristics of minority populations, such as race or socioeconomic status. This study aimed to evaluate the added value of place-based social determinants on risk-adjustment models in explaining health care costs and utilization. Statewide commercial claims from the Maryland Medical Care Database were used, including 1,150,984 Maryland residents aged 18 to 63 with ≥6 months enrollment in 2013 and 2014. Area Deprivation Index (ADI) was assigned to individuals through zip code. The authors examined the addition of ADI to predictive models of concurrent and prospective costs and utilization; linear regression was adopted for costs and logistic regression for utilization markers. Performance measures included R² for costs (total, pharmacy, and medical costs) and the area under the curve (AUC) for utilization (being top 5% top users, having any hospitalization, having any emergency room [ER] visit, having any avoidable ER visit, and having any readmission). All performance measures were derived from the bootstrapping analysis with 200 iterations. Study subjects were ∼48% male with a mean age of ∼41 years. Adding ADI to the demographics or claims-based models generally did not improve performance except in predicting the probability of having any ER or any avoidable ER visit; for example, AUC of avoidable ER visits increased significantly from .610 to .613 when using ADI rank deciles in claims-based models. Future research should focus on patients with a higher need for social services, assess more granular place-based determinants (eg, Census block group), and evaluate the added value of individual social variables.

Assessing the Population-Level Correlation of Medication Regimen Complexity and Adherence Indices Using Electronic Health Records and Insurance Claims

BACKGROUND

Nonadherence to medication regimens can lead to adverse health care outcomes and increasing costs.

OBJECTIVES

To (a) assess the level of medication complexity at an outpatient setting using population-level electronic health record (EHR) data and (b) evaluate its association with medication adherence measures derived from medication-dispensing claims.

METHODS

We linked EHR data with insurance claims of 70,054 patients who had an encounter with a U.S. midwestern health system between 2012 and 2013. We constructed 3 medication-derived indices: medication regimen complexity index (MRCI) using EHR data; medication possession ratio (MPR) using insurance pharmacy claims; and prescription fill rates (PFR; 7 and 30 days) using both data sources. We estimated the partial correlation between indices using Spearman's coefficient (SC) after adjusting for age and sex.

RESULTS

The mean age (SD) of 70,054 patients was 37.9 (18.0) years, with an average Charlson Comorbidity Index of 0.308 (0.778). The 2012 data showed mean (SD) MRCI, MPR, and 30-day PFR of 14.6 (17.8), 0.624 (0.310), and 81.0 (27.0), respectively. Patients with previous inpatient stays were likely to have high MRCI scores (36.3 [37.9], P < 0.001) and were less adherent to outpatient prescriptions (MPR = 50.3 [27.6%], P < 0.001; 30-day PFR = 75.7 [23.6%], P < 0.001). However, MRCI did not show a negative correlation with MPR (SC = -0.31, P < 0.001) or with 30-day PFR (SC = -0.17, P < 0.001) at significant levels.

CONCLUSIONS

Medication complexity and adherence indices can be calculated on a population level using linked EHR and claims data. Regimen complexity affects patient adherence to outpatient medication, and strength of correlations vary modestly across populations. Future studies should assess the added values of MRCI, MPR, and PFR to population health management efforts.

DISCLOSURES

No outside funding supported this study. The authors have nothing to disclose. The abstract of this work was presented at INFORMS Healthcare Conference, held on July 27-29, 2019, in Cambridge, MA.

How to Classify Super-Utilizers: A Methodological Review of Super-Utilizer Criteria Applied to the Utah Medicaid Population, 2016-2017

A limited number of patients, commonly termed super-utilizers, account for the bulk of health care expenditures. Multiple criteria for identifying super-utilizers exist, but no standard methodology is available for determining which criteria should be used for a specific population. Application is often arbitrary, and poorly aligned super-utilizer criteria might result in misallocation of resources and diminished effects of interventions. This study sought to apply an innovative, data-driven approach to classify super-utilizers among Utah Medicaid beneficiaries. The authors conducted a literature review of research methods to catalogue applied super-utilizer criteria. The most commonly used criteria were applied to Utah Medicaid beneficiaries enrolled during July 1, 2016-June 30, 2017, using their previous 12 months of claims data (N = 309,921). The k-medoids algorithm cluster analysis was used to find groups of beneficiaries with similar characteristic based on criteria from the literature. In all, 180 super-utilizer criteria were identified in the literature, 21 of which met the inclusion criteria. When these criteria were applied to Utah Medicaid data, 5 distinct subpopulation clusters were found: non-super-utilizers (n = 163,118), beneficiaries with multiple chronic or mental health conditions (n = 68,054), beneficiaries with a single chronic health condition (n = 43,939), emergency department super-utilizers with chronic or mental health conditions (n = 7809), and beneficiaries with uncomplicated hospitalizations (n = 27,001). This study demonstrates how cluster analysis can aid in selecting characteristics from the literature that systematically differentiate super-utilizer groups from other beneficiaries. This methodology might be useful to health care systems for identifying super-utilizers within their patient populations.

Assessing the Availability of Data on Social and Behavioral Determinants in Structured and Unstructured Electronic Health Records: A Retrospective Analysis of a Multilevel Health Care System

Background

Most US health care providers have adopted electronic health records (EHRs) that facilitate the uniform collection of clinical information. However, standardized data formats to capture social and behavioral determinants of health (SBDH) in structured EHR fields are still evolving and not adopted widely. Consequently, at the point of care, SBDH data are often documented within unstructured EHR fields that require time-consuming and subjective methods to retrieve. Meanwhile, collecting SBDH data using traditional surveys on a large sample of patients is infeasible for health care providers attempting to rapidly incorporate SBDH data in their population health management efforts. A potential approach to facilitate targeted SBDH data collection is applying information extraction methods to EHR data to prescreen the population for identification of immediate social needs.

Objective

Our aim was to examine the availability and characteristics of SBDH data captured in the EHR of a multilevel academic health care system that provides both inpatient and outpatient care to patients with varying SBDH across Maryland.

Methods

We measured the availability of selected patient-level SBDH in both structured and unstructured EHR data. We assessed various SBDH including demographics, preferred language, alcohol use, smoking status, social connection and/or isolation, housing issues, financial resource strains, and availability of a home address. EHR’s structured data were represented by information collected between January 2003 and June 2018 from 5,401,324 patients. EHR’s unstructured data represented information captured for 1,188,202 patients between July 2016 and May 2018 (a shorter time frame because of limited availability of consistent unstructured data). We used text-mining techniques to extract a subset of SBDH factors from EHR’s unstructured data.

Results

We identified a valid address or zip code for 5.2 million (95.00%) of approximately 5.4 million patients. Ethnicity was captured for 2.7 million (50.00%), whereas race was documented for 4.9 million (90.00%) and a preferred language for 2.7 million (49.00%) patients. Information regarding alcohol use and smoking status was coded for 490,348 (9.08%) and 1,728,749 (32.01%) patients, respectively. Using the International Classification of Diseases–10th Revision diagnoses codes, we identified 35,171 (0.65%) patients with information related to social connection/isolation, 10,433 (0.19%) patients with housing issues, and 3543 (0.07%) patients with income/financial resource strain. Of approximately 1.2 million unique patients with unstructured data, 30,893 (2.60%) had at least one clinical note containing phrases referring to social connection/isolation, 35,646 (3.00%) included housing issues, and 11,882 (1.00%) had mentions of financial resource strain.

Conclusions

Apart from demographics, SBDH data are not regularly collected for patients. Health care providers should assess the availability and characteristics of SBDH data in EHRs. Evaluating the quality of SBDH data can potentially enable health care providers to modify underlying workflows to improve the documentation, collection, and extraction of SBDH data from EHRs.

A predictive risk model for nonfatal opioid overdose in a statewide population of buprenorphine patients

BACKGROUND

Predicting which individuals who are prescribed buprenorphine for opioid use disorder are most likely to experience an overdose can help target interventions to prevent relapse and subsequent consequences.

METHODS

We used Maryland prescription drug monitoring data from 2015 to identify risk factors for nonfatal opioid overdoses that were identified in hospital discharge records in 2016. We developed a predictive risk model for prospective nonfatal opioid overdoses among buprenorphine patients (N = 25,487). We estimated a series of models that included demographics plus opioid, buprenorphine and benzodiazepine prescription variables. We applied logistic regression to generate performance measures.

RESULTS

About 3.24% of the study cohort had ≥1 nonfatal opioid overdoses. In the model with all predictors, odds of nonfatal overdoses among buprenorphine patients were higher among males (OR = 1.39, 95% CI:1.21-1.62) and those with more buprenorphine pharmacies (OR = 1.19, 95% CI:1.11-1.28), 1+ buprenorphine prescription paid by Medicaid (OR = 1.21, 95% CI:1.02-1.48), Medicare (OR = 1.93, 95% CI:1.63-2.43), or a commercial plan (OR = 1.98, 95% CI:1.30-2.89), 1+ opioid prescription paid by Medicare (OR = 1.30, 95% CI:1.03-1.68), and more benzodiazepine prescriptions (OR = 1.04, 95% CI:1.02-1.05). The odds were lower among those with longer days of buprenorphine (OR = 0.64, 95% CI:0.60-0.69) or opioid (OR = 0.79, 95% CI:0.65-0.95) supply. The model had moderate predictive ability (c-statistic = 0.69).

CONCLUSIONS

Several modifiable risk factors, such as length of buprenorphine treatment, may be targets for interventions to improve clinical care and reduce harms. This model could be practically implemented with common prescription-related information and allow payers and clinical systems to better target overdose risk reduction interventions, such as naloxone distribution.

Comparing the Trends of Electronic Health Record Adoption Among Hospitals of the United States and Japan

The goal of this study is to examine the trends of Electronic Health Record (EHR) adoption among hospitals in Japan compared to those in the United States. Japan's nationwide survey of hospitals was utilized to extract the EHR adoption rates among Japanese hospitals. Comparable datasets from the Healthcare Information and Management System Society (HIMSS) and the American Hospital Association (AHA) were utilized to extract EHR adoption rates among U.S. hospitals. The trends of EHR adoption were stratified and analyzed by hospital size and hospital ownership status. As of 2014, the U.S. hospitals had a wider adoption of 'basic with clinical notes' EHRs compared to Japan (45.6% vs. 27.3%), but large hospitals (400+ beds) in Japan have shown a similar adoption rate of EHR systems than those of U.S. (65.6% vs. 68.5%). Governmental hospitals tend to be more advanced in EHR adoption than non-profit hospitals in Japan (53.0% vs. 21.5%). Non-profit hospitals show the highest adoption rate of 'basic' EHR systems in the U.S. as of 2014 (63.3%). Using the 'certified' definition of EHRs, the EHR adoption rate was close to 96% among U.S. hospitals as of 2016; however, updated EHR adoption data from Japanese hospitals has yet to be collected and published. U.S. and Japan have considerably increased EHR adoption among hospitals; however, this analysis indicates different trends of EHR adoption among hospitals by size and ownership status in both countries. Learnings from government programs supporting EHR adoption in the U.S. and Japan can be helpful in planning useful strategies for future hospital-oriented health IT policies in other developed nations.

Use of a medication-based risk adjustment index to predict mortality among veterans dually-enrolled in VA and medicare

BACKGROUND

There is systemic undercoding of medical comorbidities within administrative claims in the Department of Veterans Affairs (VA). This leads to bias when applying claims-based risk adjustment indices to compare outcomes between VA and non-VA settings. Our objective was to compare the accuracy of a medication-based risk adjustment index (RxRisk-VM) to diagnostic claims-based indices for predicting mortality.

METHODS

We modified the RxRisk-V index (RxRisk-VM) by incorporating VA and Medicare pharmacy and durable medical equipment claims in Veterans dually-enrolled in VA and Medicare in 2012. Using the concordance (C) statistic, we compared its accuracy in predicting 1 and 3-year all-cause mortality to the following models: demographics only, demographics plus prescription count, or demographics plus a diagnostic claims-based risk index (e.g., Charlson, Elixhauser, or Gagne). We also compared models containing demographics, RxRisk-VM, and a claims-based index.

RESULTS

In our cohort of 271,184 dually-enrolled Veterans (mean age = 70.5 years, 96.1% male, 81.7% non-Hispanic white), RxRisk-VM (C = 0.773) exhibited greater accuracy in predicting 1-year mortality than demographics only (C = 0.716) or prescription counts (C = 0.744), but was less accurate than the Charlson (C = 0.794), Elixhauser (C = 0.80), or Gagne (C = 0.810) indices (all P < 0.001). Combining RxRisk-VM with claims-based indices enhanced its accuracy over each index alone (all models C ≥ 0.81). Relative model performance was similar for 3-year mortality.

CONCLUSIONS

The RxRisk-VM index exhibited a high level of, but slightly less, accuracy in predicting mortality in comparison to claims-based risk indices.

IMPLICATIONS

Its application may enhance the accuracy of studies examining VA and non-VA care and enable risk adjustment when diagnostic claims are not available or biased.

LEVEL OF EVIDENCE

Level 3.

Exploring the use of machine learning for risk adjustment: A comparison of standard and penalized linear regression models in predicting health care costs in older adults

BACKGROUND

Payers and providers still primarily use ordinary least squares (OLS) to estimate expected economic and clinical outcomes for risk adjustment purposes. Penalized linear regression represents a practical and incremental step forward that provides transparency and interpretability within the familiar regression framework. This study conducted an in-depth comparison of prediction performance of standard and penalized linear regression in predicting future health care costs in older adults.

METHODS AND FINDINGS

This retrospective cohort study included 81,106 Medicare Advantage patients with 5 years of continuous medical and pharmacy insurance from 2009 to 2013. Total health care costs in 2013 were predicted with comorbidity indicators from 2009 to 2012. Using 2012 predictors only, OLS performed poorly (e.g., R2 = 16.3%) compared to penalized linear regression models (R2 ranging from 16.8 to 16.9%); using 2009-2012 predictors, the gap in prediction performance increased (R2:15.0% versus 18.0-18.2%). OLS with a reduced set of predictors selected by lasso showed improved performance (R2 = 16.6% with 2012 predictors, 17.4% with 2009-2012 predictors) relative to OLS without variable selection but still lagged behind the prediction performance of penalized regression. Lasso regression consistently generated prediction ratios closer to 1 across different levels of predicted risk compared to other models.

CONCLUSIONS

This study demonstrated the advantages of using transparent and easy-to-interpret penalized linear regression for predicting future health care costs in older adults relative to standard linear regression. Penalized regression showed better performance than OLS in predicting health care costs. Applying penalized regression to longitudinal data increased prediction accuracy. Lasso regression in particular showed superior prediction ratios across low and high levels of predicted risk. Health care insurers, providers and policy makers may benefit from adopting penalized regression such as lasso regression for cost prediction to improve risk adjustment and population health management and thus better address the underlying needs and risk of the populations they serve.

The internal validation of weight and weight change coding using weight measurement data within the UK primary care Electronic Health Record

PURPOSE

To use recorded weight values to internally validate weight status and weight change coding in the primary care Electronic Health Record (EHR).

PATIENTS AND METHODS

We included adult patients with weight-related Read codes recorded in the UK's Clinical Practice Research Datalink EHR between 2000 and 2017. Weight status codes were compared to weight values recorded on the same day and positive predictive values (PPVs) were calculated for commonly used codes. Weight change codes were validated using three methods: the percentage (%) difference in kilograms at the time of the code and 1) the previous weight measurement, 2) the weight predicted using linear regression, and 3) the historic mean weight. Weight change codes were validated if estimates were consistent across two out of three methods.

RESULTS

A total of 8,108,481 weight codes were recorded in 1,000,002 patients' EHR. Twice as many were recorded in females (n=5,208,593, 64%). The mean body mass index for "overweight" codes ranged from 31.9 kg/m2 to 46.9 kg/m2 and from 17.4 kg/m2 to 19.2 kg/m2 for "underweight" codes. PPVs for the most commonly used weight status codes ranged from 81.3% (80%-82.5%) to 99.3% (99.2%-99.4%). Across the estimation methods, and using only validated weight change codes, mean weight loss ranged from - 5.2% (SD 5.8%) to -7.9% (SD 7.3%) and mean weight gain from 4.2 % (SD 5.5%) to 7.9 % (SD 8.2%). The previous and predicted weight methods were most consistent.

CONCLUSION

We have developed an internationally applicable methodology to internally validate weight-related EHR coding by using available weight measurement data. We demonstrate the UK Read codes that can be confidently used to classify weight status and weight change in the absence of weight values. We provide the first evidence from primary care that a Read code for unexpected weight loss represents a mean loss of ≥ 5 % in a 6-month period, which was broadly consistent across age groups and gender.