Validating a Method to Assess Disease Burden From Insurance Claims

When we weighted health insurance claims with a disease burden score, we were able to generate valid estimates of disability-adjusted life-years.

ABSTRACT

Objectives: To validate a method that estimates disease burden as disability-adjusted life-years (DALYs) from insurance claims and death records for the purpose of identifying the conditions that place the greatest burden of disease on an insured population.

Study Design: Comparison of the DALYs generated from death records and insurance claims with functional status and health status reported by individuals who were insured with one of HealthPartners’ commercial products and completed a health assessment in 2011, 2012, or 2013.

Methods: We calculated values of Spearman’s ρ, the rank-order coefficient of correlation, for the correlation of DALYs with self-reported function and self-reported health. We did the same for the number of medical conditions per member and the cost of claims per member.

Results: The Spearman’s ρ values for the correlation of DALYs with function were —0.241, –0.238, and –0.229 in 2011, 2012, and 2013, respectively (all P <.0001). The respective Spearman’s ρ values for the correlation of DALYs with health were —0.197, –0.189, and –0.192 (all P <.0001). These Spearman’s ρ values were similar in magnitude to those for the correlation of the number of medical conditions per member with function (—0.212, –0.213, and –0.205) and health (–0.199, –0.196, and –0.198) over the 3 years. The Spearman’s ρ values for the correlation of DALYs with function and health were greater than or equal to those for the correlation of cost of claims per member with function (–0.144, –0.193, and –0.186) and greater than those for the cost of claims per member with health (–0.126, –0.150, and –0.151).

Conclusions: Health plans can use DALYs calculated from their own health insurance claims and death records as a valid and inexpensive method to identify the conditions that place the greatest burden of poor function and ill health on their insured populations.

Am J Manag Care. 2019;25(2):e39-e44Takeaway Points

Based on the correlation of disability-adjusted life-years (DALYs) with self-reported function and self-reported health, we conclude that our method of estimating DALYs by combining mortality data with insurance claims weighted by a condition burden score is valid.

  • Important differences are revealed when population burden from a condition is defined by DALYs rather than by the number of individuals with the condition or the cost of the condition to the health plan.
  • The magnitude of these correlations with self-reported function and health is greater than that of cost of claims.
  • Unlike assessing disease burden by counting the number of conditions per individual, calculating the DALYs attributable to conditions identifies opportunities to improve population health by addressing particularly burdensome conditions.

In 2008, Berwick, Nolan, and Whittington proposed the Triple Aim as a measure of healthcare system performance.1 Although valid, feasible, and actionable measures of experience and cost have been developed for health plans, there is not a tool that a health plan can use to assess the health of its own population using its own data. HealthPartners is therefore developing summary measures of health and well-being designed for self-assessment.2 The organization has concluded that the best way to track the health of its member population is to use the information inherent in its death records and claims data. Although there are a number of ways to describe the current health and disease burden of a population, including health-adjusted life expectancy, quality-adjusted life years, and disability-adjusted life-years (DALYs),3-5 calculating DALYs best fits HealthPartners’ needs. Calculating DALYs also allows HealthPartners to compare its members’ experiences with those of Minnesota residents, the entire US population, and other populations in the United States and abroad.6-8

Insurance claims tell the organization which conditions its members are experiencing, but they also present a problem: Insurance claims are agnostic about the burden that the associated conditions place on an individual’s health and function. It is therefore necessary to assign a weight to each claim that reflects the disease burden. We used the Global Burden of Disease (GBD) as our source of condition weights.9

To document whether and the extent to which our method of calculating DALYs is a valid measure of disease burden, as well as the extent to which it provides information that is not available with measures that are more straightforward, we asked the following questions: (1) Are the correlations between DALYs and function and health of the same magnitude as an accepted measure of risk: the number of disease conditions associated with an individual?10,11 (2) Do the correlations between DALYs and function and health provide information that differs from that of 2 simpler indicators: the number of members who have a particular condition and the healthcare costs associated with a condition? (3) Is simply calculating the function losses or the health losses associated with a disease or condition as useful as our method of calculating DALYs?

To the extent that the answer to the first question is yes, we consider our method of calculating DALYs to be valid. To the extent that the answer to the second question is yes, we conclude that our method adds value, and if the answer to the third question is no, we consider that calculating DALYs from insurance claims is useful.

METHODS

The HealthPartners institutional review board (IRB) agreed that this analysis is quality improvement, not research, and is thus not subject to IRB review.

Data Source for Member-Reported Function and Health

Self-report is considered a valid way to measure both the average health and the average function of a population.12,13 It is used as such in both Europe14 and the United States.15 Each year, members insured through HealthPartners’ commercial (non—government-sponsored) programs are offered the opportunity to complete a health assessment. One question on the assessment relates to the respondent’s difficulty in doing daily work because of physical function, 1 relates to the extent to which emotional problems interfere with the respondent’s ability to do their daily work, and 1 asks the respondent to describe their health. For each question, the respondent has 5 choices, ranging from “none” to “I could not do my daily work” for physical function, “not at all” to “extremely” for emotional problems, and “poor” to “excellent” for health. We used these 3 questions in the analyses we report here. We excluded data from any member whose name appeared on one of our health plan, research, or related do-not-contact lists. Overall, this applied to 0.2% of our membership.

Assigning Condition Burden Weights to Insurance Claims

HealthPartners uses the Johns Hopkins ACG [adjusted clinical group] System in its billing process and to organize insurance claims for further analysis. This system groups all claims into 264 expanded diagnostic clusters (EDCs) and further groups the EDCs into 27 more general major expanded diagnostic clusters (MEDCs). To calculate the morbidity component of DALYs, we assigned a GBD Project score9 to each EDC. To carry out this process, 2 physicians (T.E.K. and P.D.P.) independently matched a GBD condition to each of the 264 EDCs. The 2 physicians initially agreed on the weighting for 42% of the EDCs and adjudicated an agreement on the weights for the remainder. The crosswalk is available in the eAppendix (available at ajmc.com). We then used the ACG System Aggregated Diagnosis Groups (ADGs) output to adjust the weights for the severity associated with each EDC. We assigned the upper limit of the respective GBD weight to any EDC that the ADG output classified as “major” (high severity), and we assigned the lower-limit GBD weight to all other EDCs.

Calculating DALYs

DALYs are composed of years of life lost (YLLs) due to premature death and years lived with disability due to morbid conditions. Although the choice of age cut point for YLLs is arbitrary (eg, the US Burden of Disease Collaborators used 86.0 years in a recent publication),16 we chose 75 years because it is the age used by the County Health Rankings.17 YLLs are feasible and inexpensive for us to calculate because we already have access to death records from the state of Minnesota, patient status at discharge (alive, transferred, or deceased) associated with hospital insurance claims, and HealthPartners’ decedent registry. For this analysis, we used the claims of a member only in the year that they completed a health assessment. Because the same condition may have more than 1 EDC attributed to it (eg, a patient treated for coronary heart disease may have claims that track to ischemic heart disease, cardiovascular signs and symptoms, generalized atherosclerosis, and cardiovascular disorders [other], corresponding to EDCs CAR01, CAR03, CAR10, and CAR16), we counted just 1 EDC within each MEDC, the EDC with the largest GBD weight. We used the method described by Kottke et al to calculate the DALYs due to morbidity for members who had claims in more than 1 MEDC.18 We found that calculating DALYs using the prevalence method3 best fit the data we had available.

The costs that we used in this analysis do not include pharmacy costs, because they are not accurate due to carve-outs that are covered by other organizations. If we had included them, the costs associated with chronic inflammatory disease treatment (eg, rheumatoid conditions, psoriasis, and inflammatory bowel disease) would have affected the cost rankings of the rheumatologic, skin, and gastrointestinal MEDCs in particular (19.2% of our total pharmacy costs), but they would not have affected the associations between DALYs and function and health.

Calculating Losses of Function and Health Attributable to Particular MEDCs

As part of the validity testing process, we calculated the loss of function and health attributable to particular MEDCs. To do this, we multiplied the difference between a perfect score for function (ie, 10) and the average score for members with a claim in an MEDC by the number of people in an MEDC. We also took the difference between a perfect score for self-reported health (ie, 5) and the average score for members with a claim in an MEDC and multiplied that number by the number of people in an MEDC.

Statistical Analysis

We present the means, medians, and interquartile ranges (IQRs) for the various measures. We used Spearman’s ρ, the rank test of correlation, as the measure of correlation between DALYs and self-reported health and self-reported function. We also calculated Spearman’s ρ values for the correlation of the number of MEDCs attributed to a patient with self-reported health and self-reported function and the correlation of cost per member with self-reported health and self-reported function. We considered P <.05 to be statistically significant. We used Base SAS 9.4 (SAS Institute; Cary, North Carolina) for statistical testing.

RESULTS

Population Attributes

The number of individuals who were covered by one of HealthPartners’ commercial insurance products and completed a health assessment exceeded 58,000 in each of the years 2011, 2012, and 2013 (Table 1). Among these members, the largest single age group in all 3 years was 50 to 64 years. Nearly two-thirds of the responding members were women, and nearly 90% were non-Hispanic white. The respondents were highly educated—59% in the 2011 cohort and 62% in the other 2 years had completed college or more.

Distribution of the Measures

DALYs per member averaged 0.371 to 0.385 over the 3 years. The medians were somewhat lower than the means; the IQRs ranged from 0.498 to 0.502 (Table 2). The mean function scores ranged from 9.16 to 9.26, and all 3 medians were 10 with IQRs of 1. The mean health scores ranged from 3.73 to 3.78; all medians were 4, and all IQRs were 1. The mean number of MEDCs per member varied from 4.49 to 4.62; all medians were 4, as were the IQRs. The mean cost per member varied from $4519 to $4541, with medians ranging from $1241 to $1329 and IQRs ranging from $3243 to $3299.

Individual-Level Correlations

Calculated at the individual level, the Spearman’s ρ values for the correlation of DALYs with member-reported function ranged from —0.229 to –0.241 for the 3 years (Table 3). The Spearman’s ρ values for the correlation of DALYs with member-reported health were somewhat lower.

The Spearman’s ρ values for the correlation of the number of MEDCs per member with self-reported function were similar to those for the correlation of DALYs with self-reported function. Likewise, the Spearman’s ρ values for the correlation of MEDCs per member with self-reported health were not materially different from the Spearman’s ρ values for the correlation of DALYs with self-reported health.

The Spearman’s ρ values for the correlation of the cost of medical care (not including pharmacy claims) with self-reported function, ranging from —0.144 to –0.193, were not as strong as the ρ values for the correlation of DALYs with self-reported function. The same was true for the correlation between cost of medical care and self-reported health (–0.126 to –0.151).

Association of Losses of Function and Health Attributable to an MEDC With Counts of Members in an MEDC

Based on self-reported function, the correlations between the function or health lost by the members with a claim in a particular MEDC and simply the number of members with a claim in the same MEDC are nearly perfect. Spearman’s ρ values ranged from 0.981 to 0.985 (Table 4). The same is true for the correlations based on self-reported health and the number of members with a claim in a particular MEDC; the Spearman’s ρ values ranged from 0.994 to 0.996 for the 3 years.

Impact of the Ranking Method on MEDC Rank

When the MEDCs are ranked by the DALYs, the number of individuals, and the cost of claims associated with an MEDC, important differences emerge (Table 5). For example, the musculoskeletal MEDC remains ranked first in DALYs and cost and ranked second in number of members with a claim in the MEDC in all 3 years. By contrast, the psychosocial MEDC is ranked 2nd by DALYs but just 8th to 10th by number of individuals with the condition and cost of the condition. The administrative MEDC (which includes the EDCs surgical aftercare, transplant status, administrative concerns and nonspecific laboratory tests, and preventive care), although second in cost, is near the bottom in DALYs because most of the costs attributed to the MEDC are generated by preventive services, to which we do not attribute DALYs. Cardiovascular conditions, ranked 5th by number of individuals with a claim in that MEDC, and malignancies, ranked 5th by cost of claims, ranged from 10th to 14th when ranked by DALYs.

With minor exceptions, the order of the MEDCs was stable over the 3 years when ranked by the number of DALYs attributed to the MEDC. With the exception of the ranks of malignancies and infections, which each varied by 3 levels, and administrative, which moved from the 23rd position in 2011 to 27th in 2013, none of the MEDC ranks differed by more than 2 levels in the 3 years.

DISCUSSION

In the data set of HealthPartners members who were insured with a commercial product and completed a health assessment in 2011, 2012, or 2013, we found the following regarding our method of calculating DALYs from insurance claims.

The coefficients for the correlation of DALYs with self-reported function and self-reported health were of the same magnitude as the coefficients for the correlation of the number of disease conditions an individual is experiencing with self-reported function and self-reported health. The latter is a validated measure of disease burden.10,11 We therefore conclude that our method of calculating DALYs is valid given certain limitations. The advantage of using DALYs rather than counts of conditions per individual is that DALYs can be used to identify the conditions that are creating the greatest burden on the population; simply counting conditions per individual cannot do so.

DALYs provide information that is not apparent from simply counting the number of people with a condition or the healthcare costs associated with a condition. Taking the extra step of applying condition weights to insurance claims adds value.

Because the range of GBD condition weights we used to calculate DALYs is far greater than the range of weights that can be generated by using self-reported function or health as the source of weights, DALYs provide a greater capability to discriminate among conditions.

Using DALYs as our measure of current health allows us to compare the health of our member population with benchmarks that include state and national populations.6 The method also allows us to benchmark against cities like New York.8

Limitations

There are several limitations to our analysis. The magnitude of the correlations between DALYs and self-reported function and health is probably reduced by the fact that not all individuals have the same propensity to seek medical care. This propensity may vary by disease condition. The magnitude of the correlations is also influenced by the factors by which each individual implicitly assesses their functional capability and health. Self-reported function is based on just 2 questions, 1 for physical function and 1 for emotional function. Even more limited, self-reported health is based on a single question. A more extensive scale might be expected to yield larger correlation coefficients. The fact that the data set is limited to employed individuals and their covered dependents, both of whom tend to be highly educated, and includes neither members insured by Medicare nor Medicaid also limits the generalizability. We are currently collecting function and health data self-reports from stratified random samples of adult members who are insured through both commercial and government programs. When we have adequate numbers, we will be able to repeat this analysis for our entire adult membership. An additional limitation is that because only members who completed a health assessment were included in this analysis, the DALYs due to a death were not included if the individual died without completing a health assessment. This fact would particularly affect the DALYs attributable to cardiovascular conditions and malignancies. Finally, minorities were underrepresented, the age range of the respondents was truncated, and the respondents were predominantly from Minnesota. However, we are not seeking to generalize the findings from the HealthPartners membership to a wider population.

CONCLUSIONS

The analysis we report here documents that, with limitations, our method of calculating DALYs from insurance claims is valid. In addition to being valid, the method we have developed is inexpensive and adds value. It therefore holds promise as a feasible and informative method for organizations that submit or process health insurance claims to use to assess the current health status and the greatest losses to health of the populations they serve.Author Affiliations: HealthPartners (TEK, JMG, ML, PDP, SR, JOT, NPP, SMK), Minneapolis, MN; HealthPartners Institute (TEK, JYZ, NPP), Minneapolis, MN.

Source of Funding: Analysis was funded solely and entirely by HealthPartners health plan.

Author Disclosures: The authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Authorship Information: Concept and design (TEK, JMG, SR, JYZ, NPP); acquisition of data (TEK, JMG, ML, PDP, SR, NPP); analysis and interpretation of data (TEK, JMG, PDP, SR, JOT, JYZ, NPP); drafting of the manuscript (TEK, NPP); critical revision of the manuscript for important intellectual content (TEK, SR, JOT, JYZ, NPP, SMK); statistical analysis (ML, SR); obtaining funding (TEK, NPP, SMK); administrative, technical, or logistic support (TEK, ML, SR, JOT, NPP, SMK); and supervision (TEK, JMG, PDP, SMK).

Address Correspondence to: Thomas E. Kottke, MD, MSPH, HealthPartners, 8170 33rd Ave S, MS 21110X, Minneapolis, MN 55425. Email: Thomas.e.kottke@healthpartners.com.REFERENCES

1. Berwick DM, Nolan TW, Whittington J. The triple aim: care, health, and cost. Health Aff (Millwood). 2008;27(3):759-769. doi: 10.1377/hlthaff.27.3.759.

2. Kottke TE, Gallagher JM, Rauri S, Tillema JO, Pronk NP, Knudson SM. New summary measures of population health and well-being for implementation by health plans and accountable care organizations. Prev Chronic Dis. 2016;13:E89. doi: 10.5888/pcd13.160224.

3. Department of Health Statistics and Information Systems, World Health Organization. WHO methods and data sources for global burden of disease estimates 2000-2011. World Health Organization website. who.int/healthinfo/statistics/GlobalDALYmethods_2000_2011.pdf. Published November 2013. Accessed January 18, 2019.

4. Murray CJ, Atkinson C, Bhalla K, et al; US Burden of Disease Collaborators. The state of US health, 1990-2010: burden of diseases, injuries, and risk factors. JAMA. 2013;310(6):591-608. doi: 10.1001/jama.2013.13805.

5. Gold MR, Stevenson D, Fryback DG. HALYS and QALYS and DALYS, oh my: similarities and differences in summary measures of population health. Annu Rev Public Health. 2002;23:115-134. doi: 10.1146/annurev.publhealth.23.100901.140513.

6. US health. Institute for Health Metrics and Evaluation website. healthdata.org/us-health. Accessed September 18, 2018.

7. GBD profile: United States. Institute for Health Metrics and Evaluation website. healthdata.org/sites/default/files/files/country_profiles/GBD/ihme_gbd_country_report_united_states.pdf. Accessed January 18, 2019.

8. Perlman S, Driver D. Disability-adjusted life years (DALYs) in New York City. New York City Department of Health and Mental Hygiene website. www1.nyc.gov/assets/doh/downloads/pdf/epi/databrief11.pdf. Published November 2011. Accessed January 18, 2019.

9. Salomon JA, Vos T, Hogan DR, et al. Common values in assessing health outcomes from disease and injury: disability weights measurement study for the Global Burden of Disease Study 2010 [erratum in Lancet. 2013;381(9867):628]. Lancet. 2012;380(9859):2129-2143. doi: 10.1016/S0140-6736(12)61680-8.

10. Perkins AJ, Kroenke K, Unützer J, et al. Common comorbidity scales were similar in their ability to predict health care costs and mortality. J Clin Epidemiol. 2004;57(10):1040-1048. doi: 10.1016/j.jclinepi.2004.03.002.

11. Charlson ME, Peterson JC, Boutin-Foster C, et al. Changing health behaviors to improve health outcomes after angioplasty: a randomized trial of net present value versus future value risk communication. Health Educ Res. 2008;23(5):826-839. doi: 10.1093/her/cym068.

12. Strawbridge WJ, Wallhagen MI. Self-rated health and mortality over three decades: results from a time-dependent covariate analysis. Res Aging. 1999;21(3):402-416. doi: 10.1177/0164027599213003.

13. Hoeymans N, Feskens EJ, van den Bos GA, Kromhout D. Measuring functional status: cross-sectional and longitudinal associations between performance and self-report (Zutphen Elderly Study 1990-1993). J Clin Epidemiol. 1996;49(10):1103-1110. doi: 10.1016/0895-4356(96)00210-7.

14. Szende A, Williams A, eds. Measuring Self-Reported Population Health: An International Perspective Based on EQ-5D. Rotterdam, the Netherlands: EuroQol Group; 2004. euroqol.org/wp-content/uploads/2016/10/Measuring_Self-Reported_Population_Health_-_An_International_Perspective_based_on_EQ-5D.pdf. Accessed January 18, 2019.

15. 2018 BRFSS questionnaire. CDC website. cdc.gov/brfss/questionnaires/pdf-ques/2018_BRFSS_English_Questionnaire.pdf. Published January 18, 2018. Accessed January 19, 2019.

16. Mokdad AH, Ballestros K, Echko M, et al; US Burden of Disease Collaborators. The state of US health, 1990-2016: burden of diseases, injuries, and risk factors among US states. JAMA. 2018;319(14):1444-1472. doi: 10.1001/jama.2018.0158.

17. Premature death (YPLL). County Health Rankings & Roadmaps website. countyhealthrankings.org/explore-health-rankings/what-and-why-we-rank/health-outcomes/mortality/premature-death/premature-death-ypll. Accessed January 18, 2019.

18. Kottke TE, Faith DA, Jordan CO, Pronk NP, Thomas RJ, Capewell S. The comparative effectiveness of heart disease prevention and treatment strategies. Am J Prev Med. 2009;36(1):82-88.e5. doi: 10.1016/j.amepre.2008.09.010.