Risk Adjustment Formula Research Articles

Defining a risk-adjustment formula for the introduction of population-based payments for primary care in France

Novel risk-adjusted payment models for financing primary care are currently being experimented in France. In particular, pilot schemes including shared-savings contracts or prospectively allocated capitation payments are implemented for voluntary primary care structures. Such payment mechanisms require defining a risk-adjustment formula to accurately estimate expected expenditure while maintaining appropriate efficiency incentives. We used nationwide data from the French national health data system (SNDS) to compare the performance of different prospective models for total and outpatient expenditure prediction among more than 8 million individuals aged 65 or more and their application at an aggregate level. We focused on the characterization of morbidity status and on the contextual characteristics to include in the formula. We proposed a set of practical routinely available predictors with fair performance for patient-level expenditure prediction (explaining 32% of variance) that could be used to risk-adjust prospective payments in the French setting. Morbidity information was the strongest predictor but could lead to considerable error in predicted expenditures if introduced as independent binary variables in multiplicative models, underlining the importance of summary morbidity measures and of using the appropriate metric to assess model performance. Distribution of aggregate-level allocations was greatly modified according to the method to account for contextual characteristics. Our work informs the introduction of risk-adjusted models in France and underlines efficiency and fairness issues raised.

Development and Assessment of a New Framework for Disease Surveillance, Prediction, and Risk Adjustment

Current disease risk-adjustment formulas in the US rely on diagnostic classification frameworks that predate the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM). To develop an ICD-10-CM-based classification framework for predicting diverse health care payment, quality, and performance outcomes. Physician teams mapped all ICD-10-CM diagnoses into 3 types of diagnostic items (DXIs): main effect DXIs that specify diseases; modifiers, such as laterality, timing, and acuity; and scaled variables, such as body mass index, gestational age, and birth weight. Every diagnosis was mapped to at least 1 DXI. Stepwise and weighted least-squares estimation predicted cost and utilization outcomes, and their performance was compared with models built on (1) the Agency for Healthcare Research and Quality Clinical Classifications Software Refined (CCSR) categories, and (2) the Health and Human Services Hierarchical Condition Categories (HHS-HCC) used in the Affordable Care Act Marketplace. Each model's performance was validated using R 2, mean absolute error, the Cumming prediction measure, and comparisons of actual to predicted outcomes by spending percentiles and by diagnostic frequency. The IBM MarketScan Commercial Claims and Encounters Database, 2016 to 2018, was used, which included privately insured, full- or partial-year eligible enrollees aged 0 to 64 years in plans with medical, drug, and mental health/substance use coverage. Fourteen concurrent outcomes were predicted: overall and plan-paid health care spending (top-coded and not top-coded); enrollee out-of-pocket spending; hospital days and admissions; emergency department visits; and spending for 6 types of services. The primary outcome was annual health care spending top-coded at $250 000. A total of 65 901 460 person-years were split into 90% estimation/10% validation samples (n = 6 604 259). In all, 3223 DXIs were created: 2435 main effects, 772 modifiers, and 16 scaled items. Stepwise regressions predicting annual health care spending (mean [SD], $5821 [$17 653]) selected 76% of the main effect DXIs with no evidence of overfitting. Validated R 2 was 0.589 in the DXI model, 0.539 for CCSR, and 0.428 for HHS-HCC. Use of DXIs reduced underpayment for enrollees with rare (1-in-a-million) diagnoses by 83% relative to HHS-HCCs. In this diagnostic modeling study, the new DXI classification system showed improved predictions over existing diagnostic classification systems for all spending and utilization outcomes considered.

Identifying undercompensated groups defined by multiple attributes in risk adjustment

ObjectiveTo identify undercompensated groups in plan payment risk adjustment that are defined by multiple attributes with a systematic new approach, improving on the arbitrary and inconsistent nature of existing evaluations.MethodsExtending...

Fair regression for health care spending.

The distribution of health care payments to insurance plans has substantial consequences for social policy. Risk adjustment formulas predict spending in health insurance markets in order to provide fair benefits and health care coverage for all enrollees, regardless of their health status. Unfortunately, current risk adjustment formulas are known to underpredict spending for specific groups of enrollees leading to undercompensated payments to health insurers. This incentivizes insurers to design their plans such that individuals in undercompensated groups will be less likely to enroll, impacting access to health care for these groups. To improve risk adjustment formulas for undercompensated groups, we expand on concepts from the statistics, computer science, and health economics literature to develop new fair regression methods for continuous outcomes by building fairness considerations directly into the objective function. We additionally propose a novel measure of fairness while asserting that a suite of metrics is necessary in order to evaluate risk adjustment formulas more fully. Our data application using the IBM MarketScan Research Databases and simulation studies demonstrates that these new fair regression methods may lead to massive improvements in group fairness (eg, 98%) with only small reductions in overall fit (eg, 4%).

Sample Selection for Medicare Risk Adjustment Due to Systematically Missing Data.

To assess the issue of nonrepresentative sampling in Medicare Advantage (MA) risk adjustment. Medicare enrollment and claims data from 2008 to 2011. Risk adjustment predictor variables were created from 2008 to 2010 Part A and B claims and the Medicare Beneficiary Summary File. Spending is based on 2009-2011 Part A and B, Durable Medical Equipment, and Home Health Agency claims files. A propensity-score matched sample of Traditional Medicare (TM) beneficiaries who resembled MA enrollees was created. Risk adjustment formulas were estimated using multiple techniques, and performance was evaluated based on R2 , predictive ratios, and formula coefficients in the matched sample and a random sample of TM beneficiaries. Matching improved balance on observables, but performance metrics were similar when comparing risk adjustment formula results fit on and evaluated in the matched sample versus fit on the random sample and evaluated in the matched sample. Fitting MA risk adjustment formulas on a random sample versus a matched sample yields little difference in MA plan payments. This does not rule out potential improvements via the matching method should reliable MA encounter data and additional variables become available for risk adjustment.

A Hot Health Policy Summer.

Last summer the administration finalized new rules on short-term plans and approved new state waivers. Litigation continues over the individual mandate, risk adjustment, and ACA "sabotage."

Mental Health Risk Adjustment with Clinical Categories and Machine Learning.

To propose nonparametric ensemble machine learning for mental health and substance use disorders (MHSUD) spending risk adjustment formulas, including considering Clinical Classification Software (CCS) categories as diagnostic covariates over the commonly used Hierarchical Condition Category (HCC) system. 2012-2013 Truven MarketScan database. We implement 21 algorithms to predict MHSUD spending, as well as a weighted combination of these algorithms called super learning. The algorithm collection included seven unique algorithms that were supplied with three differing sets of MHSUD-related predictors alongside demographic covariates: HCC, CCS, and HCC+CCS diagnostic variables. Performance was evaluated based on cross-validated R2 and predictive ratios. Results show that super learning had the best performance based on both metrics. The top single algorithm was random forests, which improved on ordinary least squares regression by 10 percent with respect to relative efficiency. CCS categories-based formulas were generally more predictive of MHSUD spending compared to HCC-based formulas. Literature supports the potential benefit of implementing a separate MHSUD spending risk adjustment formula. Our results suggest there is an incentive to explore machine learning for MHSUD-specific risk adjustment, as well as considering CCS categories over HCCs.

Risk Adjustment Revisited Using Machine Learning Techniques

Risk adjustment is vital in health policy design. Risk adjustment defines the annual capitation payments to health insurers and is a key determinant of insolvency risk for health insurers. In this study we compare the current risk adjustment formula used by Colombia’s Ministry of Health and Social Protection against alternative specifications that adjust for additional factors. We show that the current risk adjustment formula, which conditions on demographic factors and their interactions, can only predict 30% of total health expenditures in the upper quintile of the expenditure distribution. We also show the government’s formula can improve significantly by conditioning ex ante on measures indicators of 29 long-term diseases. We contribute to the risk adjustment literature by estimating machine learning based models and showing non-parametric methodologies (e.g., boosted trees models) outperform linear regressions even when fitted in a smaller set of regressors.

A Machine Learning Framework for Plan Payment Risk Adjustment.

To introduce cross-validation and a nonparametric machine learning framework for plan payment risk adjustment and then assess whether they have the potential to improve risk adjustment. 2011-2012 Truven MarketScan database. We compare the performance of multiple statistical approaches within a broad machine learning framework for estimation of risk adjustment formulas. Total annual expenditure was predicted using age, sex, geography, inpatient diagnoses, and hierarchical condition category variables. The methods included regression, penalized regression, decision trees, neural networks, and an ensemble super learner, all in concert with screening algorithms that reduce the set of variables considered. The performance of these methods was compared based on cross-validated R2 . Our results indicate that a simplified risk adjustment formula selected via this nonparametric framework maintains much of the efficiency of a traditional larger formula. The ensemble approach also outperformed classical regression and all other algorithms studied. The implementation of cross-validated machine learning techniques provides novel insight into risk adjustment estimation, possibly allowing for a simplified formula, thereby reducing incentives for increased coding intensity as well as the ability of insurers to "game" the system with aggressive diagnostic upcoding.

Regression Trees Identify Relevant Interactions: Can This Improve the Predictive Performance of Risk Adjustment?

Risk equalization formulas have been refined since their introduction about two decades ago. Because of the complexity and the abundance of possible interactions between the variables used, hardly any interactions are considered. A regression tree is used to systematically search for interactions, a methodologically new approach in risk equalization. Analyses are based on a data set of nearly 2.9 million individuals from a major German social health insurer. A two-step approach is applied: In the first step a regression tree is built on the basis of the learning data set. Terminal nodes characterized by more than one morbidity-group-split represent interaction effects of different morbidity groups. In the second step the 'traditional' weighted least squares regression equation is expanded by adding interaction terms for all interactions detected by the tree, and regression coefficients are recalculated. The resulting risk adjustment formula shows an improvement in the adjusted R2 from 25.43% to 25.81% on the evaluation data set. Predictive ratios are calculated for subgroups affected by the interactions. The R2 improvement detected is only marginal. According to the sample level performance measures used, not involving a considerable number of morbidity interactions forms no relevant loss in accuracy. Copyright © 2015 John Wiley & Sons, Ltd.

Risk Selection Threatens Quality Of Care For Certain Patients: Lessons From Europe's Health Insurance Exchanges.

Experience in European health insurance exchanges indicates that even with the best risk-adjustment formulas, insurers have substantial incentives to engage in risk selection. The potentially most worrisome form of risk selection is skimping on the quality of care for underpriced high-cost patients--that is, patients for whom insurers are compensated at a rate lower than the predicted health care expenses of these patients. In this article we draw lessons for the United States from twenty years of experience with health insurance exchanges in Europe, where risk selection is a serious problem. Mistakes by European legislators and inadequate evaluation criteria for risk selection incentives are discussed, as well as strategies to reduce risk selection and the complex trade-off among selection (through quality skimping), efficiency, and affordability. Recommended improvements to the risk-adjustment process in the United States include considering the adoption of risk adjusters used in Europe, investing in the collection of data, using a permanent form of risk sharing, and replacing the current premium "band" restrictions with more flexible restrictions. Policy makers need to understand the complexities of regulating competitive health insurance markets and to prevent risk selection that threatens the provision of good-quality care for underpriced high-cost patients.

The Performance of Risk Adjustment Models in Colombian Competitive Health Insurance Market

We introduce new risk groups to a standard capitation formula and evaluate risk selection incentives of insurers. The study uses a unique data set of almost 24 million affiliates to Government’s mandatory health insurance system. This data set is very rich in the sense of reporting all claims during year 2010, basic demographic variables, initial diagnostic, health services, pharmaceuticals used, etc. It compromises more than 300 million claims. We construct two diagnostic related groups: an adaptation of the 3M algorithm, and a ad hoc diagnostic related group constructed by the authors. Using standard linear capitations formulas we evaluate incentives for cream skimming using several measures. In general, results show a notable improvement in the explanatory power of health expenditures by introducing the ad hoc diagnostic related groups to the standard Colombian risk adjustment formula. With the new risk groups the R2 of the model is 13.53% as opposed to 1.45% of the current formula. Furthermore, for users in the highest expenditure quintile, expected expenditure is 71% of actual expenditure, as opposed to 27% under the current formula. This suggest there is much space for improving the current Colombian capitation formula using information that is currently available.

On the calculation of the Israeli risk adjustment rates

The Israeli risk adjustment formula, introduced in 1995 and which serves for the allocation of the health budget to the sickness funds, is unique compared to countries with a similar national health insurance system in that it is not calculated on the basis of actual cost data of the sickness funds but on the basis of quantities retrieved from surveys. The current article aims to analyze the implications of the Israeli methodology. The article examines the validity of the Israeli methodology used to set the 2004 risk adjustment rates and compare these rates with the "correct" ones, which are derived from the 2004 internal relative cost scales of the sickness funds. The Israeli methodology ignores services provided by the sickness funds and assumes constant unit cost across the sickness funds, an assumption which is implausible. Comparing the actual and the "correct" rates, it turns out that the actual rates over-compensate all the sickness funds for members in age 0-14, and under-compensate them for insurees aged 55+. In age 0-4, the over-compensation per capita is about NIS 1,500 while the under-compensation in age group 75+ reaches NIS 1,600. The current risk adjustment formula distorts the intended competition on good quality care among the sickness funds, and turns it into a competition on profitable members. After 18 years of using incorrect rates, the Israeli risk adjustment rates should be calculated, as is common in other systems, based on individual cost data from the sickness funds.

The new risk adjustment formula in Germany: Implementation and first experiences

In Germany risk adjustment is a core element of the regulatory framework of competition between sickness funds. It shall create a level playing field between funds with very heterogeneous risk structures. Prior to 2009 risk adjustment was mainly by a demographic model. In 2009 morbidity based risk adjustment was introduced, embedded in a broader reform of the statutory health insurance system. The new formula covers 80 “severe” or “costly and chronic” diseases structured in a system of hierarchical groups.The performance of the new system was evaluated by the Advisory Board of Scientific Experts on Risk Adjustment to the Federal Insurance Office. The evaluation is based on the data of the first year 2009. Individual level as well as group level performance improved considerably comparing to the old model. Also predictive accuracy on the sickness funds level improved. The evaluation analyzed coding and prescription patterns in a joint point analyses. No changes of coding behavior which can be ascribed to the introduction of the new system were found.Some issues for further improvement were identified: A systematic underpayment for higher age groups is observed due to not annualizing the costs of the deceased. Systematic underpayment for multi-morbidity and systematic over-payment for those without any medical conditions can also be observed, although the level of over-/underpayment was clearly reduced compared to the old model.

Performance Measures for Schizophrenia Research: In Reply

Back to table of contents Previous article Next article LettersFull AccessPerformance Measures for Schizophrenia Research: In ReplyDonald Emile Addington, M.B.B.S., Emily McKenzie, M.Sc., and JianLi Wang, Ph.D.Donald Emile AddingtonSearch for more papers by this author, M.B.B.S., Emily McKenzieSearch for more papers by this author, M.Sc., and JianLi WangSearch for more papers by this author, Ph.D.Published Online:1 Jul 2012https://doi.org/10.1176/appi.ps.20120p721aAboutSectionsPDF/EPUB ToolsAdd to favoritesDownload CitationsTrack Citations ShareShare onFacebookTwitterLinked InEmail In Reply: Burns has made three general points in his letter about our article on performance measures for schizophrenia services. First, he has endorsed the use of the Delphi process for engaging multiple stakeholder groups. Second, he has made pertinent observations about the process of using the Delphi based on his considerable experience working with multiple groups, including consumers. More specifically, he has pointed out the potential problem that occurs when groups rate many items as essential, and he has made several practical suggestions about how to encourage the group to be more focused and selective. We recognize the value of his suggestions and, in retrospect, we could have improved our Delphi process.We have a different perspective on his third suggestion, to return to the Delphi group with improved directions in order to obtain a more focused group of performance measures with the goal of changing the practice of individual clinicians with a set of “performance measures that nobody could ignore.” The reason underlying our perspective is that we see the identification of measures as only a small step on the path to implementing performance measures in mental health services. In a prior systematic review, we examined the facilitators and barriers to implementing performance measures (1). The review identified seven broad factors that influenced the implementation of performance measures, including indicator characteristics, promotional strategies, implementation strategies, resources, individual factors, organizational factors, and external factors. In our article on performance measures, which is the subject of Burns' letter, we focused on one factor only, indicator characteristics.We have previous experience with use of the Delphi process as part of the process of identifying performance measures for first-episode psychosis services (2). When we moved to implement these measures, we found that some measures rated as essential were, in practice, hard to specify with precision, and some were difficult to collect. When we studied the use of the available measures to compare services, only one measure, hospitalization, was sufficiently robust and widely available to be considered suitable for setting standards (3). We have since sought to increase the potential value of hospitalization as a performance measure for first-episode psychosis services. First, we developed a risk adjustment formula for hospitalization (4), and second, we tested the validity of hospitalization as a performance measure by demonstrating that it correlated with two other important outcome measures, quality of life and global psychopathology (5).In conclusion, we see the systematic review and rating of evidence as essential first steps in developing performance measures. The Delphi is a useful next step to narrow the number of measures, but it reflects only opinion evidence, one of the weaker levels of evidence in evidence-based medicine. Further practical steps include the rigorous definition of measures and further research to validate the measures. All these steps would still address only one of the seven factors (indicator characteristics) that we identified as facilitators and barriers to implementing performance measures in mental health services.References1 Addington D , Kyle T , Desai S , et al.: Facilitators and barriers to implementing quality measurement in primary mental health care: systematic review. Canadian Family Physician 56:1322–1331, 2010 Medline, Google Scholar2 Addington D , McKenzie E , Addington J , et al.: Performance measures for early psychosis treatment services. Psychiatric Services 56:1570–1582, 2005 Link, Google Scholar3 Addington D , Norman R , Adair C , et al.: A comparison of early psychosis treatment services using consensus and evidence-based performance measures: moving towards setting standards. Early Intervention in Psychiatry 3:274–281, 2009 Crossref, Google Scholar4 Addington DE , Beck C , Wang J , et al.: Predictors of admission in first-episode psychosis: developing a risk adjustment model for service comparisons. Psychiatric Services 61:483–488, 2010 Link, Google Scholar5 Addington DE , McKenzie E , Wang JL : Validity of hospital admission as an outcome measure of services for first-episode psychosis. Psychiatric Services 63:280–282, 2012 Link, Google Scholar FiguresReferencesCited byDetailsCited ByNone Volume 63Issue 7 July 2012Pages 721-722 Metrics PDF download History Published online 1 July 2012 Published in print 1 July 2012

Morbiditätsorientierter Risikostrukturausgleich

Numerous health systems have introduced competition between health plans while banning risk-rated premiums. Risk adjustment for health plans is introduced to reduce incentives for risk selection and to create incentives for health plans to permanently invest in care for the chronically ill. According to the international health economics state of the art, risk adjustment in the German social health insurance system has used information on health status (measured by diagnoses and drug prescriptions) on top of demographic information since 2009. In non-competitive health care systems similar mechanisms are sometimes established, e.g. to achieve an equitable distribution of resources between regions. An evaluation of the first year of health-based risk adjustment demonstrates a superior performance in comparison to the old, demographic risk adjustment. The old risk adjustment formula (without ex post high-cost pooling) showed R(2) of 5.8%, CPM of 10.4% and MAPE of 2,226 €, in contrast to the new health status-based risk adjustment formula (without cash benefit for sick allowance) which reaches R(2) 20.2%, CPM 22.5% and MAPE 1,817 €. However, to make competition between health plans functional for improvement of quality and efficiency of health care, health plans must be granted additional instruments to act as prudent buyers of health care.

Risk-Type Concentration and Efficiency Incentives: A Challenge for the Risk Adjustment Formula

An important goal of risk-adjusted capitation payments (RACPs) to competitive community-rated health plans—that may differ in coverage and/or the organisation of delivering care—is to reduce incentives for risk selection while maintaining incentives for efficiency. In most schemes, RACPs are simply based on the average observed costs in risk groups (in a prior year). We show that under this procedure, incentives for efficiency will not always be maintained: when identical risk types are concentrated in the same health plans—due to selection, specialisation or just coincidence—cost savings can be captured by the RACPs and leak away from these plans.

Improving Risk Adjustment in the Czech Republic

This paper analyses possible options how to improve the risk adjustment of the health insurance system in the Czech Republic. From the possible options it argues for including pharmaceutical cost groups (PCGs) as additional risk factors since it is an improvement that can be implemented almost instantaneously. On real data from an anonymous sickness fund it confirms that predictive performance of PCGs models is consistently better than the performance of the demographic model that is currently used. The study also describes and examines the Czech health insurance market and implications of proposed changes of policy makers. Based on experience from other countries we point to a problem of risk selection if the changes are not accompanied by a tighter regulation, specifically in the form of improved risk adjustment formula.

Inferring Risk-Averse Probability Distributions from Option Prices Using Implied Binomial Trees: Additional Theory and Extensions

Arnold, Crack and Schwartz (ACS) (2010) generalize the Rubinstein (1994) risk-neutral implied binomial tree (R-IBT) model by introducing a risk premium. Their new risk-averse implied binomial tree model (RA-IBT) has both probabilistic and pricing applications. They use the RA-IBT model to estimate the pricing kernel (i.e., marginal rate of substitution) and implied relative risk aversion for a representative agent. They also use the RA-IBT to explore the differences between risk-neutral and risk-averse moments of returns. They also discuss practical applications of the RA-IBT model to Value at Risk and stochastic volatility option pricing models. This paper presents additional theoretical details not contained in ACS. We present a deeper discussion on the assumptions required for the risk-averse trees, we discuss details for extending ACS through the use of general utility functions to generate discount rates in the RA-IBT, and we present further theoretical details on the propagation of risk-averse probabilities through an RA-IBT. We also present an alternate CAPM-driven derivation of the certainty equivalent risk-adjusted discounting formula that is derived using no-arbitrage principles in ACS and an alternate direct estimation routine for the RA-IBT that is similar to Rubinstein’s “one-two-three” technique.

Deriving a risk-adjustment formula for hospital financing: Integrating the impact of socio-economic status on length of stay

The imperfect risk adjustment of prospective payment for hospitals may have dramatic consequences on equity. If the hospital is able to distinguish subgroups of patients with different expected costs within a group for which the risk-adjusted payment per admission is the same, it is likely to select the most profitable cases and deny care to the others. Meanwhile, hospitals refusing to practice patients’ selection may experience solvency problems. In the long term, either those hospitals fail and access to care is at risk, or they decrease the quality of treatments and access to quality is at risk.In Belgium, since 1995, a prospective payment per case has replaced the traditional per diem payments for non-medical expenditures. A fixed number of days are paid to each admission, based on the patient's characteristics, namely diagnosis, age and geriatric profile. In this paper, we examine the imperfect risk adjustment related to the non-inclusion of socio-economic factors in the hospital financing formula.Using data from 61 Belgian hospitals from 1995, we observe that socio-economic status, which is currently not accounted for as risk adjuster, has a significant impact on length of stay (LOS). We estimate that patients in the upper-income categories, patients with a self-employed status and patients with an employee status are beneficial for hospitals’ financial results, due to their shorter stays. On the contrary, the non-active, the low-income patients and patients benefiting from an insurance preferential regime represent, on average, a financial loss for hospitals. Finally, we find that financial results under the current financing scheme are biased due to the non-inclusion of SES risk-adjustors. Hospitals with the most beneficial social case-mix are shown to experience a shift from a positive to a negative financial outcome when SES risk adjustors are included, while the reverse is observed for hospitals with the worst social case-mix.