Comorbidity Type and Health Care Costs in Type 2 Diabetes: A Retrospective Claims Database Analysis

In a large, privately insured population with newly diagnosed type 2 diabetes, we found that roughly a quarter of patients had only comorbidities unrelated to diabetes (e.g., pulmonary disorders and musculoskeletal diseases), a quarter had severe, coexisting medical conditions that may eclipse diabetes management (e.g., cancer and dementia), and a quarter had both diabetes-concordant and -discordant comorbidities; only 7% had concordant conditions that share care goals with diabetes and no other types of comorbidities. We also found that patients’ health care utilization and costs varied substantially across comorbidity groups. In particular, diabetes patients with clinically dominant conditions accounted for a disproportionately large share of health care costs. Their average spending was almost twice as high as patients with both concordant and discordant comorbidities, and more than four times as high as patients with concordant illnesses only or discordant illnesses only, even after adjusting for demographic characteristics, health insurance type, region, and prior utilization.

Differentiating the nature of comorbidities and their relationships to diabetes is critical to disease management [1, 2, 22]. Previous studies have documented that not only the count but also the type and combination of comorbidities may alter the priorities of diabetes care [1‐12]. Some complex or severe comorbid illnesses may even dictate all other disease management plans, including diabetes [1‐4]. For example, one analysis of the Diabetes Epidemiology Cohort (DEpiC) found that diabetes patients with clinically dominant comorbidities may be less likely to test their HbA1c and LDL-C regularly, visit a medical professional for diabetes-related care, and meet glycemic and lipid management goals, compared with individuals with other types of comorbidities [4]. In an analysis of veterans with diabetes, one in three patients had serious or multiple medical or mental health problems that would increase the risks or decrease the benefits of intensive glycemic control [23]. Our findings further indicate that severely symptomatic conditions (e.g., dementia) or end-stage illnesses (e.g., end-stage renal disease) may drive health care resource use in the diabetes population more than other types of comorbidities, highlighting the substantial health care needs of diabetes patients with complex health status.

Despite the well-recognized comorbidity burden in patients with diabetes, clinical practice guidelines often do not explicitly address “comorbidity interrelatedness” [24], with few exceptions such as the 2012 joint consensus report by the American Diabetes Association and the American Geriatrics Society [25] and the 2018 American Diabetes Association Standards of Medical Care in diabetes for older adults [26], which provide guidance on treating diabetes in the context of geriatric conditions. Although studies have suggested aggressive management of metabolic syndromes and concordant conditions among diabetes patients [27], less attention has been paid to other types of comorbidities. Further, some concordant conditions, such as obesity and renal impairment, have been found to predict high-cost diabetes patients [28]. Although some patients may benefit from disease-specific interventions, for people with multimorbidity, coordinated services across multiple conditions are essential to maximize health care efficiency [29]. In our sample, few diabetes patients (7%) had concordant conditions only, whereas the majority of patients with diabetes were faced with various co-existing health problems. Diabetes care management programs must address the comorbidity mix, treatment burden, and competing care demands, because managing individual health conditions in isolation may be ineffective and inefficient [22, 30‐32]. In some cases, adherence to treatment guidelines for one disease may even adversely exacerbate another [24, 30].

In our sample, more than half of the total costs were concentrated on diabetes patients with clinically dominant conditions. Indeed, dominant comorbidity, such as dementia [33] and cancer [34], can make the management of other conditions, including diabetes, more difficult and expensive. Although dementia [35‐39] and cancer [40‐53] also may be associated with diabetes, these conditions are intentionally classified as dominant, rather than concordant, comorbidities in our analysis to be consistent with the Piette and Kerr framework. These conditions require special consideration because the complexity and seriousness of the disease often eclipse the management of other health problems, including diabetes [1]. In contrast, only 3% of total costs were attributable to diabetes patients with concordant comorbidities only. Better understanding the distribution of health care expenditures in the population is crucial because the information can help payers, providers, and policymakers assess the disease burden in different subgroups, which has implications for patient care [54, 55]. As providers are facing increased pressure to control health care costs, one common strategy may be to target the specific subgroup responsible for a disproportionately large share of expenditures [56, 57]. To that end, the Piette and Kerr framework may serve as a screening tool for identification of high-need, high-cost patients with diabetes and suggest targets for tailored interventions, such as intensive medication treatment [58, 59]. A collaborative, multidisciplinary team approach, such as the Chronic Care Model, may be especially pertinent to managing multiple comorbidities as it has been shown to facilitate patient self-management and improve diabetes care quality [60‐62].

As the economic burden of diabetes continues to rise, payers and providers, including accountable care organizations, need reliable methods to predict health care expenditures and identify patients with high-cost risk. Our findings suggest that explicit consideration of diabetes concordant, discordant, and dominant comorbidities as defined by the Piette and Kerr framework is important not only because this classification is clinically meaningful but because this approach segments patients into “actionable cohorts” [56] with distinctly different expenditure profiles. Appropriate risk adjustment based on patients’ demographic characteristics, comorbidity burden, and other factors is essential for quality reporting and pay-for-performance programs [63‐65]. Additional research comparing performance of the Piette and Kerr classification methods and other diabetes-specific risk adjustment measures in expenditure prediction [66] would be useful in improving diabetes value-based payment models.

Our analysis measured baseline comorbidities prior to the type 2 diabetes diagnosis. Possibly some concordant comorbidities may have not had ample time to develop. Concordant comorbidities that are nonetheless present in our data may be different from those that develop after diabetes diagnosis. It would be helpful for future research to characterize the temporal sequence of comorbidity and the corresponding cost impacts among patients with type 2 diabetes. Further analyses could extend the follow-up period and examine how the relationship between comorbidity type and cost impacts may change over time. A subsequent longitudinal analysis might consider how the Piette and Kerr framework can be used to identify diabetes patients with persistent (in addition to transient) high costs.

Several limitations in our study merit discussion. First, comorbidity burden may have been underestimated as a result of under-diagnosis and under-coding in administrative claims data. Second, the comorbidity measure we analyzed, though widely used and clinically relevant, does not allow us to determine the severity of a given condition, which may affect health care resource use. Although clinical values (such as serum creatinine levels) may reflect disease severity, these data were only available for a subset of patients in Optum Clinformatics^® Data Mart and therefore not used in our analyses. Third, Optum Clinformatics^® Data Mart includes participants in commercial insurance plans and Medicare Advantage plans. Future research is needed to validate our results in other populations, such as fee-for-service Medicare beneficiaries with type 2 diabetes. Fourth, our analysis pertained to patients newly diagnosed with type 2 diabetes. This approach reduces time-dependent biases between patients, such as differences in duration and severity of diabetes as well as survival time [67], which may have important impact on costs. Sensitivity analyses including all patients with prevalent diabetes showed similar results, however. Finally, claims data lack information on patient values and preferences for care, which strongly correlate with one’s health care decisions [31, 68]. Future research should explore how to incorporate weights that reflect individual preferences into comorbidity measures in order to help guide treatment priorities and align treatment plans with patients’ health care goals [1, 31, 60].