Associations of Intensive Lifestyle Intervention in Type 2 Diabetes With Health Care Use, Spending, and Disability

Key Points Question Is an intensive lifestyle intervention for type 2 diabetes associated with long-term health care use and Medicare spending? Findings This ancillary study of a randomized clinical trial linked 2796 participants with type 2 diabetes in a randomized intensive lifestyle intervention with Medicare data. Among linked participants, the intervention was associated with reduced weight, improved diabetes control, and reduced health care costs during the intervention, but there was no reduction in total health care spending after the intervention. Meaning These findings suggest that intensive lifestyle interventions targeted to patients with type 2 diabetes may need to be sustained to reduce long-term health care spending.


Introduction
About 9% of the US population has received a diabetes diagnosis, and treatment costs exceed $327 billion per year. 1 Most of this diabetes is type 2, and primary and secondary prevention are cornerstones of the public health strategy. 2 In particular, excess body weight is linked with worse glycemic control and elevated cardiovascular risk. 3 As a result, there is significant interest in intensive lifestyle management with a focus on diet, physical activity, and weight loss to reduce the incidence and harm of type 2 diabetes. 4 There are reasons to be optimistic. Individualized guidance on diet and exercise are effective at preventing diabetes in the first place and improving diabetes control and reducing cardiovascular risk factors for those with type 2 diabetes. [5][6][7][8] Policy makers, insurers, and employers have taken notice. Versions of intensive lifestyle interventions (ILIs) have been incorporated into Medicare, 9 workplace wellness programs, 10 and patient-centered medical homes. 11 However, these interventions are labor intensive and expensive, often involving lifestyle counselors, exercise specialists, dieticians, and administrative staff. 12,13 Thus, it is important to demonstrate long-term efficacy and value, although some advocates assume it. 14 Many studies find that interventions can reduce short-term health care use, but long-term estimates on health care use and costs often rely on simulations that make assumptions about long-term persistence of these short-term effects. 15,16 In this study, we linked administrative data with a clinical trial to examine whether reductions in health care spending were sustained after an ILI ended. The data were obtained from the Look AHEAD (Action for Health in Diabetes) study, one of the largest ILIs performed to date. Beginning in 2001, 5145 patients aged 45 to 76 years with type 2 diabetes who were overweight or obese were randomized to ILI or to a control group that received diabetes support and education. 8 Although the intervention did not significantly reduce its primary outcome, a composite of death from cardiovascular causes, nonfatal myocardial infarction or stroke, or hospitalization for angina, the intervention successfully reduced participants' weight, increased physical fitness and functional status, lowered hemoglobin A 1c (HbA 1c ), and increased probability of diabetes remission, among other clinical benefits. 8,[17][18][19][20] However, the intervention was also expensive, costing $2865 per patient per year in the first year and gradually decreasing to $1120 per patient per year in years 5 to 9 (in 2012 dollars), whereas the diabetes support and education received by the control group cost less than $202 per patient year in the first year and $103 in years 5 to 9. 13 The intervention also reduced hospitalizations, prescription drug use, and total health care costs during the intervention period by $5280 per patient (in 2012 dollars). 21 An outstanding question is whether these reductions persisted after the intervention, particularly given that the intervention was not cost-saving during the trial.

Study Overview and Design
This ancillary study to the Look AHEAD randomized clinical trial examines the association of an ILI for weight loss targeted to patients with type 2 diabetes with long-term health care use, Medicare spending, and disability insurance enrollment. The Look AHEAD study ended the intervention in September 2012 but continued to follow participants. As a part of the observational studies following the intervention period, the Look AHEAD study obtained written informed consent to perform linkages with administrative records. We linked study participants to Medicare records (for those who consented to such a linkage) to estimate outcomes in the year the intervention ended (2012) and the 3 following years (2013 to 2015). The University of Southern California and Wake Forest institutional review boards approved the data linkage, and the University of Minnesota institutional review board deemed it exempt from review because University of Minnesota researchers did not work with direct identifiers. Our manuscript proposal to the Look AHEAD study, including our proposed outcome measures and analyses, is available in the Manuscript Proposal Supplement 1. The trial protocol for the Look AHEAD study is published elsewhere. 8 This study is reported following Consolidated Standards of Reporting Trials (CONSORT) reporting guideline, as applicable to a secondary analysis of clinical trial data.
The Look AHEAD study randomized participants within 16 study sites to either the ILI group or a control group (diabetes support and education). Randomization occurred between 2001 and 2004. The most intensive portion of the intervention, and the greatest weight loss, occurred during the first year, when participants had weekly sessions with counselors, dieticians, exercise specialists, and behavioral health staff. By the fourth year, individual contacts occurred on a monthly basis, as well as group classes provided throughout the year. 8 The control group received 3 educational sessions per year in the first 4 years, and annually thereafter.

Data Linkage and Study Population
Our analysis compared Medicare outcomes between the ILI and control groups after the intervention. Not all study participants were given the opportunity to consent to data linkages, as the queries were conducted as part of 2 cohort studies following the intervention (the Look AHEAD-

Data Sources
We used the Medicare Master Beneficiary Summary File to determine participants' eligibility, Medicare Advantage enrollment, and spending by service category. One concern about analyses using Medicare data is the absence of claims data for participants in Medicare Advantage. We measured hospital admissions and emergency department (ED) use for enrollees in Medicare Advantage using patient-level data from the 2012 to 2014 Healthcare Effectiveness Data and Information Set. Medicare Advantage plans are required to report these data to the National Committee for Quality Assurance for nearly all Medicare Advantage enrollees; thus, these data capture health care use for most enrollees in our sample. We constructed identical measures of hospitalizations and ED visits for fee-for-service Medicare enrollees from Medicare claims, following prior research comparing fee-for-service Medicare and Medicare Advantage. 22 We linked the Medicare data with Look AHEAD trial data, which includes baseline participant characteristics, intervention arm assignment, and clinical outcomes during the study.

Study Measures
We examined the long-term associations of the ILI with measures of health care use, medical care spending, prescription drug spending, and Medicare eligibility owing to disability or end-stage renal disease. We preregistered our study outcomes on ClinicalTrials.gov. (professional payments, some home health care, medical equipment, laboratory tests, mental health services, and ambulance services). We measured annual prescription drug spending using total drug costs incurred in Medicare Part D. Specifically, this outcome measures total Part D drug costs for a given year, including the ingredient cost, dispensing fees and sales tax, but not manufacturer discounts or rebates. 23 We also investigated annual beneficiary out-of-pocket spending on prescription drugs in Medicare Part D. Total annual Medicare spending was defined as the sum of Medicare Part A and Part B spending and total Part D prescription drug costs. Functional status was measured using the reason for initial Medicare eligibility, reflecting that individuals who are eligible for Social Security Disability Insurance become eligible for Medicare after 2 years. 24 In addition, individuals with end-stage renal disease, which can be a complication of type 2 diabetes, may be eligible for Medicare prior to age 65 years. A small minority of participants were enrolled in Medicare at the beginning of the trial. Thus, to the extent that the ILI improved functional status or reduced diabetes-related kidney disease, the trial could have decreased subsequent Medicare eligibility owing to disability or end-stage renal disease and increased Medicare enrollment owing to age.
A key challenge in our analysis was that not all components of health care spending were available for all study participants, based on enrollment in traditional Medicare vs Medicare Advantage and enrollment in Part D drug coverage. Specifically, annual medical care spending measures (excluding prescription drugs) were only available for fee-for-service Medicare enrollees

Statistical Analysis
First, we examined whether our linked sample (ie, those who consented and for whom we were able to successfully obtain Medicare records) was representative of the broader Look AHEAD study. We compared the characteristics of the linked sample (2796 participants) with originally randomized participants (5145 participants) and then assessed whether the linked sample replicates the effects of ILI on weight and HbA 1c documented for the original sample. 8 Next, we estimated the associations of ILI with hospital and ED use and Medicare spending. We used generalized linear models for health care use and spending outcomes (eAppendix in Supplement 2). 25 We used logit models to estimate Medicare disability status and other binary outcomes. For the person-year-level analysis of annual health care use and spending, we calculated cluster-robust SEs at the person level. We calculated heteroskedasticity-robust SEs for the personlevel disability analysis. All analyses controlled for demographic characteristics (ie, age, sex, and selfreported race/ethnicity), clinical status at initial randomization (ie, obesity status, HbA 1c , hypertension, and cardiovascular disease), socioeconomic status at initial randomization (ie, education and income), and initial study site (using fixed effects). We also controlled for months of Medicare coverage for those who may have enrolled mid-year and year-fixed effects in models investigating health care use and Medicare spending.
We also estimated year-specific associations of the intervention with selected measures of health care use and spending.
Data were analyzed using Stata statistical software version 15.0 (StataCorp). P values were 2-sided, and statistical significance was set at .05. Analysis began in December 2018 and was completed in September 2020.

Results
We found no statistically significant differences in baseline characteristics between linked ILI and control group participants (

Associations of ILI With Weight and HbA 1c During and After the Trial
We found similar associations of ILI with weight and HbA 1c for the linked sample compared with the full Look AHEAD study during and after the trial (Figure 2). 26

Associations of ILI With Health Care Use and Spending
For our primary analysis sample (ie, fee-for-service Medicare enrollees with Part D prescription drug coverage), a similar percentage of person-years in the ILI and control groups had any hospitalization also estimated year-specific ILI vs control group differences in the probability of hospital admission and ED visits, Part D prescription drug costs, and total Medicare spending. We displayed each outcome by year and intervention status, adjusting for baseline characteristics and study site (Figure 3). None of the year-specific adjusted outcomes differed significantly by intervention status.

Intervention Control
Each point represents the mean weight or HbA 1c in each year relative to randomization separately for intervention and control participants, adjusted for baseline participant characteristics and study site.

Discussion
In an era of increasing health care costs associated with chronic illness, it is imperative to identify interventions to reduce long-term spending without harming patient care. To this end, this ancillary study linked data from a large randomized clinical trial to Medicare claims to identify the long-term associations of intensive lifestyle management for patients with type 2 diabetes. Similar to the overall study, we found significant reductions in weight and HbA 1c in the linked intervention group compared with the linked control group, with reductions in weight persisting beyond the intervention period. Despite this, we found no postintervention differences in hospital admissions or ED use, Medicare Part A spending, or eligibility for Medicare owing to enrollment in Social Security Disability Insurance or end-stage renal disease. While we found that the intervention participants had lower gross costs for Part D drugs from 2012 to 2015, they also had higher Part B spending, and there was no difference in total Medicare spending.
Our study is not the first to investigate whether improving diabetes management can reduce health care spending. Observational studies have found that reductions in body mass index and HbA 1c were correlated with reductions in health care spending among patients with type 2 diabetes. 27,28 Other studies have found that interventions successfully managing diabetes were associated with reduced short-term health care spending. 15,16 However, we found that the benefits of successful diabetes interventions may not persist after interventions end. Notably, our ability to track health care use and spending was facilitated by the linkage of trial participants with Medicare data and highlights the utility of administrative data sources for tracking the long-term associations of interventions with patient outcomes.
Our results are also related to recent work using randomized clinical trials to estimate the associations of workplace wellness programs with health care use and spending. 29,30 These programs include components that are similar to the intensive lifestyle intervention we studied, with registered dieticians and other practitioners providing counseling on nutrition and physical activity.
In contrast, the wellness programs were targeted broadly at all employees, and both studies found small and statistically insignificant associations of the programs with health outcomes, health care use, and spending. Our results, paired with earlier estimates from the Look AHEAD trial, show that targeting lifestyle interventions to at-risk populations (ie, patients with type 2 diabetes) may be important for successfully improving chronic disease management and reducing health care use and spending, but spending reductions may not persist beyond the intervention period. 21 While our analysis found that reductions in prescription drug spending persisted beyond the trial period, the differences became statistically insignificant over time. In addition, the benefits in terms of reduced hospitalizations during the intervention did not persist beyond the intervention period, although hospitalizations were measured during the trial using validated self-reports and medical record reviews rather than Medicare data. 21 This result may imply that the benefits of lifestyle intervention diminish over time, particularly as the patient cohort ages and faces other health risks. Extending interventions may be a potential approach to achieve persistent effects of ILI; however, more evidence is needed to determine whether this is an effective strategy to reduce long-term health care use and spending. Alternatively, there is evidence that the Look AHEAD ILI was associated with reduced bone mineral density and increased the risk of frailty fractures. 31   intervention, the considerable clinical and functional benefits of ILI during and after the intervention period may have been worth the costs. Formal cost-effectiveness analysis is necessary to weigh the costs and benefits of the ILI.

Limitations
Our study had several limitations. First, because the questions about data linkage manifested after the study was under way, we were only able to link 54% of the original cohort to Medicare records. To the extent that linked ILI participants differed from linked control participants along unobserved attributes that were correlated with long-term health care use, such imbalance could introduce bias our results. However, we were able to link a nearly identical percentage of ILI and control participants to Medicare data (54% in each group), the linked participants were similar to the original cohort along observed characteristics, the linked intervention and control groups were balanced across observed characteristics, and most importantly, the intervention had similar effects on body weight and HbA 1c for linked participants as the original cohort. Second, we were not able to observe all outcomes for all linked participants. In particular, we only observed total Medicare spending among fee-for-service Medicare enrollees with Part D prescription drug coverage. However, we did not observe differential enrollment by study arm, suggesting that this did not bias our estimates. Third, we were only able to observe health care use recorded in the Medicare program. One of the Look AHEAD trials' clinic sites included a Veterans Health Administration facility. Individuals with both Veterans and Medicare coverage may have received care at Veterans Health Administration facilities that did not appear in Medicare claims. Only approximately 5% of the sample came from this site, and our results were unchanged removing this site from the sample.
Fourth, the Look AHEAD trial recruited volunteers with type 2 diabetes who could complete a fitness test and were motivated to participate in the trial. 8 Thus, the outcomes associated with ILI could differ in a broader population with type 2 diabetes.

Conclusions
This ancillary study found that a randomized clinical trial of an ILI was not associated with reduced total Medicare spending in the years immediately following the intervention. These results suggest that ILIs may need to be sustained for reductions in health care costs to persist. However, some of the benefits of the trial may have yet to be observed, implying the importance of continued evaluation.