Measuring Multimorbidity: A Risky Business - …

1 EDITORIAL AND COMMENTM easuring multimorbidity : A Risky BusinessLori A. Bastian, MD, MPH1,2, Cynthia A. Brandt, MD, MPH1,3, and Amy C. Justice, MD, PhD1,21 Pain Research, Informatics, Multimorbidities, and Education (PRIME) Center (West Haven COIN), VA Connecticut Healthcare System, West Haven,CT, USA;2 Division of General Internal Medicine, Department of Medicine, Yale University School of Medicine, New Haven, CT, USA;3 Department ofEmergency Medicine, Yale University School of Medicine, New Haven, CT, Gen Intern Med 32(9):959 60 DOI: Society of General Internal Medicine (outside the USA) 2017 The influence of multiple coexisting conditions( multimorbidity ) on patient outcomes is well established,but the optimal method for Measuring multimorbidity is not. Inthis issue ofJGIM, Radomski and colleagues present findingsfrom a national cohort of Veterans Affairs (VA) and non-VAadministrative files used to estimate and compare health out-comes for veterans over age 65 with diabetes who are usingdual sources of primary outcome was mortality, andthe major finding was that mortality rates differed dependingon which measure of multimorbidity was used.

2 For example,in the model controlling for multimorbidity based on Interna-tional Classification of Disease (ICD) diagnostic codes, thedual-use groups had lower odds of death than VA-predominant users. On the other hand, in the model based onpharmacy records, the dual-use groups hadhigherodds ofdeath than VA-predominant users. The authors point out theimportance of Measuring multimorbidity and outline the needfor reliable risk adjustment methods that are not susceptible tomeasurement differences across health agree that appropriately accounting for multimorbidityis essential and we think that the widening adoption of elec-tronic health records (EHRs) strongly suggests a new ap-proach to measurement. When Dr. Mary Charlson first devel-oped her comorbidity index, it was based on inpatient clinicaldata collected prospectively during the course of ,3 Inboth development and validation studies, the weighted indexperformed well in discriminating risk of ,3 Howev-er, when investigators adapted the index to use ICD diagnosticcodes, its discrimination declined the past20 years, additional indices using ICD codes have been de-veloped in an attempt to improve the performance of thesemeasures.

3 For example, the Elixhauser index (used in theRadomski article) was expanded to 30 unweighted outpatientconditions, including mental , the inaccuracyand measurement bias of ICD codes for clinical conditions iswell documented. Unfortunately, providers within systemsand across systems vary in their coding practices and maycode for only one diagnosis even when patients are seen formultiple growing body of literature has examined the use ofpharmacy records to create a comorbidity measure when pre-scription medications are indicative of the disease example, the RxRisk-V (used in the Radomski article)uses pharmacy records to identify 45 conditions and is weight-ed to predict ,7 Although these pharmacy-basedmeasures do avoid the measurement bias of ICD diagnosticcodes, and pharmacy records are fairly well standardized andportable across US healthcare systems, prescriptions reflectindividual provider decision making and behavior one pro-vider may be more likely than another to prescribe medica-tions for a given , whether based upon ICD codes or pharmacy records,there is significant potential for bias.

4 One must ask, in the eraof paperless EHRs, why are we still talking about billing data(ICD codes or pharmacy records) for risk adjustment? Healthservices researchers would do well to look beyond billing data,to the direct clinical data becoming widely available in work in prognosis has demonstrated the superiority ofclinical data over billing data for predicting like to highlight three sources of better data on burdenof disease: routine laboratory data, modern-day chart reviewvia text processing, and directly collected patient demonstrated by the Veterans Aging Cohort Study(VACS) Index, it is possible to combine demographic datawith routine, clinical Laboratory Improvement Amendments(CLIA)-certified (and standardized) laboratory data such ashemoglobin, platelets, hepatitis C tests, creatinine, and trans-aminases, to develop and validate a prognostic index that ishighly predictive of mortality and hospitalization among agingpopulations with and without HIV (See also ) As in clinical practice, laboratoryvalues are considered accurate until newer values are availablewithin a reasonable time window.

5 The benefit of this approachis substantial. CLIA certification and standardization meansthat a hemoglobin measurement in New Haven, Connecticut,and one in Orlando, Florida, have the same data are not subject to provider or clinic biases indiagnosis, and routine laboratory tests are widely , the discrimination achieved exceeds or rivals thatobtained from more difficult-to-measure online June 26, 2017959 JGIMWith the growth of EHRs and expanded capacity forcomputer processing, it is time to move beyond admin-istrative coded diagnoses and manual chart from clinical encounters including medications,laboratory results, and clinically meaningful diagnosticresults in clinical notes extracted with computational andnatural language processing (NLP) methods could poten-tially create more predictive risk adjustment indexes. Asdemonstrated in a recent manuscript, the extraction ofthe ejection fraction using NLP from text notes in theEHR was critical to determining the risk of heart failurewith preserved versus reduced ejection fraction for HIV-infected there are benefits inusing NLP to extract clinically meaningful pieces ofinformation from the EHR, no one system makes com-plete use of EHR information.

6 Multiple systems or ap-proaches will be , we measure exposures to controlled substances(urine toxicology screen) because patients reports of thisbehavior may not be reliable. Yet, we continue to base tobaccoand alcohol use on patient reporting. Moving forward, weshould also measure these exposures at the point of care andinclude this retrievable information in the EHR. Likewise, wecould retrieve information on physical activity and sleep qual-ity from wearable devices. Multiple assays and applicationshave been validated and are available to measure these expo-sures. Accurate measures of these health behaviors couldsignificantly improve risk , the major limitation of the study by Radomski andcolleagues is the reliance on administrative data. Anothersignificant limitation is that the data are not adjusted fordemographics such as education, and educational level isstrongly correlated with health outcomes. Finally, these dataare from 2008 to 2011, and may no longer be applicable, sincethe VA has recently implemented many changes in measurement of multimorbidity matters to patients,policymakers, researchers, and clinicians, and the use of ad-ministrative coded diagnoses is inherently limited.

7 Next stepsare to improve multimorbidity risk adjustment measures andbase these on provider s notes, laboratory data, and point-of-care Author:Lori A. Bastian, MD, MPH; Pain Research,Informatics, Multimorbidities, and Education (PRIME) Center (WestHaven COIN)VA Connecticut Healthcare System, West Haven, CT,USA (e-mail: with Ethical Standards:Conflict of Interest:The authors declare no conflict of TR, Zhao X, Thorpe CT, Thorpe JM, Naples JG, Mor MK,Good CB, Fine MJ, Gellad Impact of Medication-Based RiskAdjustment on the Association between Veteran Health Outcomes andDual Health System Use. J Gen Intern Med. ME, Pompei P, Ales KL, MacKenzie prognostic comorbidity in longitudinal studies: developmentand validation. J Chron Dis. 1987;40(5):373 ME, Sax FL, MacKenzie CR, Fields SD, Braham RL, illness severity: does clinical judgement work? J ChronDis. 1986;39:439 RA, Cherkin DC, Ciol a clinical comorbidity indexfor use with ICD-9-CM administrative databases. J Clin ;45:613 A, Steiner C, Harris RD, Coffey Measuresfor Use with Administrative Data.)

8 Med Care. 1998;36(1):8 ,SalesAE,LiuCF,FishmanP,NicholP,SuzukiNT ,Sharp and Characteristics of the RxRisk-V: A VA-Adapted Pharmacy Based Case-mix Instrument. Med ;41(6):761 ML, El-Serag HB, Tran TT, Hartman C, Richardson P,Abraham the Rx-Risk-V for Mortality Prediction inOutpatient Populations. Med Care. 2006;44(8):793 LC, Lee SJ, Schonberg MA, Widera EW, Smith indices for older adults: a systematic review. ;307:182 KM, Tate JP, Crothers K, Crystal S, Leaf DA, Womack JA,Brown TT, Justice AC, Oursler adapted frailty-related pheno-type and the VACS index as predictors of hospitalization and mortality inHIV-infected and uninfected individuals. J Acquir Immune Defic ;67:397 Commentary: Observational research in the age of theelectronic health record. Am J Epidemiol. 2014;179:759 MS, Chang C-CH, Skanderson M, et betweenHIV infection and the risk of heart failure with reduced ejection fractionand preserved ejection fraction in the antiretroviral therapy era: resultsform the Veterans Aging Cohort Study.

9 JAMA Cardiol. 2017; 2:536 et al.: Measuring multimorbidity : A Risky BusinessJGIM