Diabetes Treatment Intensity and Accelerated Functional Decline in Older Adults: Evidence From the Health and Retirement Study

Introduction

Diabetes mellitus is a leading determinant of functional limitation and disability in older adults, with converging longitudinal evidence linking the condition to loss of independent living across diverse populations. A systematic review and meta-analysis confirmed a consistent association between diabetes and increased risk of physical disability.[1] Prospective data demonstrated that diabetes was associated with incident functional disability in older women,[2] and longitudinal analyses in an elderly urban cohort showed that diabetes independently predicts long-term disability.[3] Subclinical functional impairment attributable to diabetes has also been documented among initially nondisabled older adults, suggesting that deterioration begins before overt disability thresholds are reached.[4] These findings collectively establish diabetes as a major longitudinal predictor of declining physical function in the aging population.

While the diabetes–disability association is well established, substantially less is known about whether escalation of pharmacologic treatment intensity adds prognostic information about the rate of functional decline beyond diabetes status alone. Pharmacologic management of type 2 diabetes typically follows an intensity continuum, from no medication to oral antidiabetic agents to insulin, with transitions often reflecting worsening glycemic control and increasing disease severity. The METABOLIC Study demonstrated the feasibility of linking medication-exposure categories to multidimensional functional assessment in older patients with type 2 diabetes taking oral antidiabetic treatment,[5] and frailty has been found to be associated with pharmacologic therapy patterns among older adults with type 2 diabetes.[6] Evidence of potential overtreatment in older adults with tight glycemic control raises concern that treatment intensification may increase adverse-event risk without commensurate functional benefit.[7] However, no prior longitudinal study has directly tested whether progression through treatment-intensity stages predicts accelerated functional decline trajectories in a nationally representative older-adult cohort.

We addressed this gap using the RAND HRS Longitudinal File, a panel of United States adults aged 50 years and older drawn from a nationally representative survey and followed across up to 15 biennial interview waves (1994–2022).[8] At each interview, respondents who report diabetes are asked whether they take oral medication and whether they use insulin for it, which allows treatment intensity to be measured directly and updated over time. Prior work in the Health and Retirement Study by Bardenheier et al. demonstrated that functional decline predicts subsequent diabetes incidence, the reverse direction from our question, underscoring the bidirectionality of this relationship.[9] The present analysis tests the forward direction: whether time spent on insulin is associated with steeper trajectories of ADL and instrumental ADL (IADL) limitation accumulation.

Methods

This analysis used the RAND HRS Longitudinal File 2022 (Version 1), a user-friendly longitudinal file derived from the Health and Retirement Study (HRS), the premier longitudinal study of aging in the United States. The HRS is sponsored by the National Institute on Aging and conducted by the University of Michigan Survey Research Center. The RAND file is created and maintained by the RAND Center for the Study of Aging. This secondary analysis used deidentified public-use data and did not constitute human participants research requiring institutional review board review. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.[10]

Study Design

The HRS uses a biennial panel design. Its age-eligible sample is adults older than 50 years, and spouses are interviewed regardless of age.[8] We used all interviews from waves 2–16 (1994–2022), including proxy interviews, which the HRS conducts when a respondent is unable to answer; excluding them would remove interviews from respondents whose function had declined most. Each respondent’s baseline was the first interview at which they were aged 50 years or older and diabetes status was known. The analytic cohort comprised respondents with at least 1 ADL measurement, and all counts in the text, tables, and figures refer to this cohort. The 50-year threshold follows the HRS sampling frame and prior HRS analyses of diabetes and function, which used thresholds of 50 or 51 years;[9, 11, 12, 13] its effect was tested directly.

Exposures

Diabetes was defined from the first interview at which the respondent reported a physician diagnosis and carried forward thereafter. At each interview, respondents with diabetes reported whether they currently take oral medication for diabetes and whether they currently use insulin. The primary exposure was time-varying: at each interview, a respondent was classified as having no diabetes, diabetes without insulin (non-insulin), or insulin-treated diabetes, with insulin status carried forward from the first interview at which insulin use was reported. A missing insulin item was treated as no new report, so the respondent kept the state reported at earlier interviews. Because states change only at reported transitions, classification at any interview uses no information from later interviews. In a secondary specification, the non-insulin state was split into diabetes without medication and oral medication without insulin.

Three fixed classifications were analyzed for comparison: the highest treatment reported at any interview during follow-up (the exposure in the earlier version of this analysis), treatment at the baseline interview followed regardless of later change (a landmark design), and the highest stage of diabetes duration since first report (<10 vs ≥10 years). Agreement between the duration stage and reported insulin use was tabulated.

Outcomes

The primary outcome was the count of ADL limitations (0–5 scale; items: bathing, dressing, eating, transferring from bed, and walking across a room). The secondary outcome was the count of IADL limitations (0–5 scale; items: using the telephone, managing money, taking medications, shopping for groceries, and preparing meals). Both outcomes were assessed at each biennial interview; IADL data were available from wave 3 (1996) onward.

Statistical Analysis

Functional trajectories were estimated using linear regression with respondent-clustered robust standard errors. Time was defined as years since each respondent’s baseline interview, scaled in decades (t_it). The primary model took the form:

Y_it = β_0 + β_1 t_it + β_2 S_non,it + β_3 S_ins,it

+ γ_1 T_non,it + γ_2 T_ins,it

+ δ(t_it × a_0i) + X_i' θ + ε_it

where Y_it is the ADL (or IADL) limitation count for respondent i at interview t; S_non,it and S_ins,it indicate the current treatment state; T_non,it and T_ins,it are the decades spent in the non-insulin and insulin states since baseline; a_0i is baseline age, centered at 60 years and scaled per 10 years; and X_i is the vector of baseline covariates. The primary estimands were γ_1 and γ_2, the excess limitation slope per decade spent in each state relative to the trajectory without diabetes. β_2 and β_3 capture the difference in limitation count at entry into each state (or at baseline for respondents already in that state). We tested γ_2 − γ_1 to compare insulin with non-insulin treatment, and estimated a linear trend across the 4 ordered states (no diabetes, diabetes without medication, oral medication, insulin) by constraining the excess slope to increase by a common amount per step. Covariates were baseline age, sex, race, Hispanic ethnicity, education, BMI, hypertension, heart disease, stroke, chronic lung disease, arthritis, current smoking, alcohol use, and CES-D score. Missing binary covariates were coded as a separate category, and missing continuous covariates, including CES-D for respondents whose baseline interview was by proxy, were set to the median with a missingness indicator.

Sensitivity analyses addressed model specification, sample, and measurement. To adjust rates of change as well as baseline levels, all covariates were interacted with time. Generalized estimating equations with an exchangeable working correlation and a linear mixed model with random intercepts and slopes reduced the weight given to respondents with many interviews. Nonlinear time was modeled with a natural cubic spline (4 df) for time since baseline and, separately, with squared terms for decades in each treatment state. A Poisson model with log link estimated excess slopes on a relative scale. Weighted analyses used baseline HRS respondent weights with standard errors clustered on sampling units (stratum × sampling error computation unit). We repeated the primary model restricted to self-respondent interviews; excluding respondents with diabetes who had any missing insulin item; ending follow-up in 2014, before the 2016 revision of the HRS diabetes module; excluding respondents whose reported year of diagnosis implied onset before age 30 years, as a proxy for type 1 diabetes; and restricting to respondents with no ADL limitation at baseline. To test whether treatment adds information beyond diabetes duration, a further model added a time-varying term for decades spent beyond 10 years of diabetes duration. As an absolute measure, we estimated the unadjusted Kaplan-Meier cumulative incidence of a first ADL limitation within 10 years among respondents without ADL limitations at baseline, by treatment at baseline. Deaths during follow-up and loss to follow-up (last interview before 2022 without a recorded death) were tabulated by group.

Because long-acting glucagon-like peptide-1 receptor agonists (GLP-1 RAs) may have delayed or replaced insulin in recent years, changing who is classified as insulin-treated, we repeated the primary model separately for follow-up in 1994–2008 and 2010–2022. Liraglutide, the first once-daily long-acting GLP-1 RA, was approved in the United States in 2010. The HRS does not record GLP-1 RA use, so calendar period served as a marker of availability. Each period was analyzed as its own cohort with baseline reset to the first interview in the period; both periods were fitted in one model with all terms interacted with period, and the interaction for decades on insulin tested whether the insulin excess slope differed. We repeated this with a 2014 cut (approval of once-weekly dulaglutide) and with later follow-up ending in 2014. To assess the 50-year age threshold, we refitted the primary model with minimum baseline ages of 51 (the HRS age-eligibility threshold), 55, 60, and 65 years (the threshold used by American Diabetes Association guidance for older adults),[14] resetting baseline to the first interview at or above each threshold, and tested whether slopes differed between respondents aged 50–64 and ≥65 years at baseline. Statistical significance was defined at α = .05 (two-sided); analyses beyond the primary model were not adjusted for multiple comparisons and should be interpreted as supportive. All models were estimated in Python (statsmodels 0.14) from a single analysis script.

Results

Study Population

The analytic cohort comprised 43,553 respondents with 272,804 person-wave observations (Figure 1); 43,095 respondents (263,355 observations) contributed to the IADL model. During follow-up, 30,100 respondents (69.1%) did not report diabetes, 9,340 (21.4%) had diabetes without insulin, and 4,113 (9.4%) reported insulin use at 1 or more interviews (Table 1). Median follow-up was 10 years without diabetes, 12 years with non-insulin diabetes, and 12 years with insulin; 56.8% of respondents who used insulin died during follow-up, compared with 42.2% without diabetes, and 11.0% and 24.3%, respectively, were lost to follow-up alive. Mean (SD) baseline age was 60.8 (10.1) years, and 55.8% were women. Respondents who used insulin more often had hypertension (65.9% vs 34.5%) and arthritis (45.4% vs 28.0%) at baseline, and fewer had no ADL limitation (74.6% vs 89.0%). Among respondents with diabetes, 22%–30% reported insulin use at each interview year (Figure 2; eTable 1 in Supplement 1).

Primary Outcome

For a respondent aged 60 years at baseline, ADL limitations increased by 0.227 per decade without diabetes. Each decade spent with non-insulin diabetes added 0.048 (95% CI, 0.017 to 0.079; P = .002), and each decade on insulin added 0.150 (95% CI, 0.076 to 0.224; P < .001) (Table 2; Figure 3). The insulin excess was larger than the non-insulin excess by 0.102 (95% CI, 0.023 to 0.181; P = .01). Entry into the insulin state coincided with 0.279 (95% CI, 0.235 to 0.322; P < .001) more ADL limitations, whereas entry into the non-insulin state did not (0.012 (95% CI, −0.007 to 0.031; P = .21)). Across ordered treatment steps, the excess slope was 0.004 (95% CI, −0.047 to 0.055; P = .87) for diabetes without medication, 0.067 (95% CI, 0.029 to 0.104; P < .001) for oral medication, and 0.152 (95% CI, 0.079 to 0.226; P < .001) for insulin (trend per step, 0.040 (95% CI, 0.025 to 0.055; P < .001)). Among respondents without ADL limitations at baseline, the 10-year cumulative incidence of a new ADL limitation was 27.3% (95% CI, 26.7%–27.9%) without diabetes, 41.7% (95% CI, 39.7%–43.7%) with non-insulin treatment, and 55.5% (95% CI, 51.7%–59.3%) with insulin at baseline.

Secondary Outcome

In the IADL model, the excess slope per decade was 0.108 (95% CI, 0.042 to 0.174; P = .001) for insulin and 0.056 (95% CI, 0.024 to 0.087; P < .001) for non-insulin treatment; the difference between them was not significant (0.053 (95% CI, −0.019 to 0.124; P = .15)).

Alternative Exposure Definitions

With respondents assigned to the highest treatment reported during follow-up, the insulin excess ADL slope was 0.152 (95% CI, 0.114 to 0.190; P < .001), similar to the primary estimate (Table 2). With treatment fixed at the baseline interview, the insulin excess was 0.143 (95% CI, 0.046 to 0.240; P = .004) for ADL but not significant for IADL (0.067 (95% CI, −0.015 to 0.149; P = .11)). The duration stage agreed poorly with reported treatment: 41.8% of respondents who reached ≥10 years of diabetes ever reported insulin, and 47.4% of insulin users never reached 10 years during follow-up. When decades beyond 10 years of diabetes duration were added to the primary model, the duration term was 0.169 (95% CI, 0.084 to 0.254; P < .001) and the insulin excess slope attenuated to 0.075 (95% CI, −0.006 to 0.157; P = .07). With fixed groups, insulin (0.110 (95% CI, 0.063 to 0.158; P < .001)) and duration of ≥10 years (0.085 (95% CI, 0.052 to 0.118; P < .001)) were each associated with a steeper slope.

Sensitivity Analyses

The insulin excess ADL slope was 0.134 (95% CI, 0.060 to 0.208; P < .001) with all covariates interacted with time, 0.151 (95% CI, 0.043 to 0.259; P = .006) with survey weights and design, 0.128 (95% CI, 0.060 to 0.196; P < .001) in self-respondent interviews only, 0.163 (95% CI, 0.075 to 0.251; P < .001) among respondents with complete treatment data, 0.149 (95% CI, 0.073 to 0.225; P < .001) after excluding likely type 1 diabetes, and 0.231 (95% CI, 0.152 to 0.311; P < .001) among respondents with no ADL limitation at baseline (Table 3; Figure 4). Models that give less weight to respondents with many interviews gave larger estimates (generalized estimating equations, 0.284 (95% CI, 0.221 to 0.347; P < .001); mixed model, 0.503 (95% CI, 0.461 to 0.545; P < .001)). A spline for time left the estimate unchanged (0.151 (95% CI, 0.078 to 0.225; P < .001)), and squared terms did not improve fit (joint P = .08). On the relative scale of the Poisson model, the rate ratio per decade on insulin was 0.941 (95% CI, 0.876 to 1.010; P = .09): limitations accumulated faster in absolute terms but at a similar proportional rate, because insulin users started from more limitations.

Calendar Period and Age Threshold

The insulin excess ADL slope was 0.112 (95% CI, −0.008 to 0.233; P = .07) for follow-up in 1994–2008 and 0.095 (95% CI, −0.004 to 0.194; P = .06) for follow-up in 2010–2022; neither period estimate was significant on its own, and they did not differ (P = .83; Table 4). A 2014 cut (P = .54) and later follow-up ending in 2014 (0.060 (95% CI, −0.165 to 0.286; P = .60)) gave similar conclusions. The proportion of respondents with diabetes reporting insulin did not fall after 2010, whereas the insulin item was missing for 26%–38% of respondents with diabetes after the 2016 module revision, compared with 14% or less earlier (eTable 1 in Supplement 1).

The insulin excess slope was 0.152 (95% CI, 0.079 to 0.226; P < .001) with a minimum baseline age of 51 years, 0.155 (95% CI, 0.080 to 0.231; P < .001) at 55 years, 0.133 (95% CI, 0.049 to 0.217; P = .002) at 60 years, and 0.116 (95% CI, 0.015 to 0.217; P = .02) at 65 years. In respondents aged ≥65 years at baseline, the estimate was imprecise (0.051 (95% CI, −0.125 to 0.227; P = .57)) compared with 0.187 (95% CI, 0.106 to 0.267; P < .001) at ages 50–64 years (P = .20 for interaction).

Discussion

In this cohort of 43,553 adults aged 50 years and older followed for up to 28 years, time spent on insulin was associated with a steeper absolute accumulation of ADL limitations than time without diabetes, and insulin initiation coincided with a step increase in limitations. The excess rose with each treatment step, and 10-year incidence of a new ADL limitation was roughly twice as high among adults using insulin at baseline as among adults without diabetes. These findings extend evidence linking diabetes to disability,[1, 2] with three qualifications that the sensitivity analyses make clear: the insulin excess was partly shared with longer diabetes duration, it was not present on a relative scale, and it was weaker for IADL than for ADL.

Prior observational studies have documented that diabetes is a consistent predictor of functional limitation. Prospective evidence established diabetes as a risk factor for incident functional disability in older women,[2] and longitudinal analyses in an elderly urban cohort demonstrated that diabetes independently predicts long-term disability.[3] The Health, Aging, and Body Composition study documented subclinical ADL limitation among initially nondisabled older adults with diabetes,[4] and cross-sectional analyses in NHANES linked diabetes, comorbidity burden, and hemoglobin A1c to functional disability.[15] Community-based work further connected diabetes and hyperglycemia to functional disability burden,[16] and a meta-analysis confirmed the consistency of the diabetes–disability association across study designs.[1] In the HRS, diabetes has been associated with fewer disability-free life-years from age 50 years[12] and with geriatric conditions that develop from middle age onward.[13] The present analysis asks a different question: whether the intensity of diabetes treatment is associated with the rate of limitation accumulation.

Several mechanisms may explain why insulin-treated diabetes predicts worse function. Insulin is usually started after other agents fail, so it marks longer disease duration, greater cumulative metabolic burden, and more microvascular and macrovascular complications; the attenuation of the insulin slope after adjustment for duration is consistent with this. Insulin and sulfonylureas carry the highest risk of hypoglycemia,[17] and diabetes has been associated with hospitalized fall injury,[18] while sulfonylureas have been associated with falls and fractures among nursing home residents with type 2 diabetes.[19] Beyond hypoglycemia and frailty,[6] insulin therapy causes weight gain, which was greater with insulin than with other intensive regimens in the UK Prospective Diabetes Study,[20] and obesity has become increasingly associated with disability in older US adults.[21] Insulin also requires dose adjustment, injection technique, and glucose monitoring; lower health literacy is associated with worse glycemic control,[22] and patients with low diabetes-related numeracy commonly misread glucose meters and miscalculate medication doses.[23] Difficulty managing a complex regimen may both reflect and add to early cognitive and functional decline. These pathways could not be tested directly in the HRS. Decision-analytic modeling has demonstrated that expected benefits of intensive glucose control depend on comorbid illness and functional status,[24] and multimorbidity patterning may further accelerate functional decline beyond the contribution of diabetes alone.[11]

The pattern across scales helps interpret the size of the association. On the additive scale, insulin users accumulated more limitations per decade; on the relative scale, they did so at a proportional rate similar to that of adults without diabetes. Because insulin users started from more limitations, both statements hold: the absolute burden grows faster, which is what matters for care needs, but the data do not show that insulin use multiplies the underlying rate of decline. Models that weight each respondent more equally gave larger estimates than the pooled model, so the primary estimate is conservative in magnitude.

Since 2010, long-acting GLP-1 RAs may have allowed some patients with type 2 diabetes to delay or avoid insulin, which could change who is classified as insulin-treated. The HRS does not record GLP-1 RA use, so we examined calendar period instead. Period-specific estimates were imprecise and did not differ, and the proportion of respondents with diabetes reporting insulin did not fall over the study period. These data therefore neither show nor exclude a change in the insulin association after GLP-1 RAs became available. Concerns about overtreatment in older adults with tight glycemic control are relevant,[7] as are descriptions of insulin use and discontinuation among adults aged 75–79 years with type 2 diabetes.[25]

The 50-year threshold matches the HRS sampling frame and prior HRS studies of diabetes and function.[9, 11, 12, 13] Estimates were similar when the cohort was restricted to adults aged 65 years and older, the threshold used in American Diabetes Association guidance for older adults,[14] but imprecise when respondents aged ≥65 years at baseline were analyzed alone. The larger estimate at ages 50–64 years is consistent with HRS evidence that diabetes-related geriatric conditions appear in middle age.[13] The bidirectional relationship between function and diabetes is also germane: functional decline was associated with subsequent diabetes incidence in the HRS,[9] raising the possibility that reverse causation partially explains the observed associations. Psychosocial context, including loneliness and functional limitation, may further shape these relationships in ways not captured by the current covariate adjustment.[26]

For clinical care, the step increase in limitations at insulin initiation and the faster absolute accumulation afterward suggest that starting insulin is a natural point to assess ADL and IADL function, falls risk, and the ability to self-manage injections, and to reassess at follow-up visits. Adults who already need help with daily activities may need simplified regimens, caregiver involvement in insulin administration, or glycemic targets relaxed in line with guidance for older adults.[14] De-intensification strategies for elderly patients[27] and deintensification of insulin and sulfonylurea therapy after severe hypoglycemia[17] are being actively investigated, and the trajectories documented here support patient-centered goals that include maintenance of ADL independence. Next steps include HRS data linked to Medicare Part D claims, which would distinguish GLP-1 RAs, sodium-glucose cotransporter 2 inhibitors, sulfonylureas, and insulin and allow active-comparator designs; cohorts with hemoglobin A1c, hypoglycemia, weight, and health literacy measures to test the mechanisms above; and trials of whether functional assessment at insulin initiation, or de-intensification, changes functional trajectories.

Limitations

This analysis has several limitations. First, treatment was self-reported as current oral medication or insulin use at biennial interviews, without dose, adherence, glycemic, weight, or hypoglycemia measures; the insulin item was missing more often after the 2016 HRS diabetes-module revision, although restriction to respondents with complete treatment data or to follow-up through 2014 gave similar results. Second, the HRS does not distinguish type 1 from type 2 diabetes; excluding respondents with onset before age 30 years did not change the findings, but misclassification remains possible. Third, the HRS does not identify GLP-1 RAs or other non-insulin injectable agents, which respondents may have reported as insulin or omitted, and the calendar-period analysis cannot isolate their effect. Fourth, diabetes duration was counted from the first HRS report, which understates duration for respondents with diabetes at entry, so the duration-adjusted model may not fully separate duration from treatment. Fifth, as an observational cohort study, the analysis is susceptible to residual confounding and to reverse causation, and analyses beyond the primary model were not adjusted for multiple comparisons. Sixth, respondents who used insulin were more likely to die during follow-up; selective survival would tend to attenuate observed differences. Finally, primary estimates are unweighted; weighted estimates were similar.

Conclusions

In this cohort of adults aged 50 years and older from the Health and Retirement Study (1994–2022), insulin-treated diabetes was associated with more ADL limitations at insulin initiation and with faster absolute accumulation of limitations thereafter, compared with no diabetes, with a graded association across treatment steps. The excess was partly shared with longer diabetes duration, was not present on a relative scale, and was weaker for IADL, so it should be read as a marker of advanced disease rather than evidence that insulin causes decline. Insulin initiation may be a practical point at which to assess function and plan support for independence in daily activities.

Article Information

Funding/Support. The Health and Retirement Study is sponsored by the National Institute on Aging (grants U01AG009740 and R01AG053972) and conducted by the University of Michigan. This analysis received no specific funding.

Data Sharing Statement. RAND HRS data are publicly available after registration with the Health and Retirement Study. Analysis code is available from the corresponding author on request.

AI Disclosure: AI tools (ChatGPT, OpenAI) were used for proofreading, language refinement, vocabulary suggestions, and text analysis. All AI-assisted outputs were manually reviewed, verified, and approved by the senior author, who take full responsibility for the final manuscript.

References

1. Evelyn Wong, Kathryn Backholer, Emma Gearon, Jessica L. Harding, Rosanne Freak-Poli, Christopher Stevenson, and Anna Peeters. Diabetes and risk of physical disability in adults: a systematic review and meta-analysis. The Lancet Diabetes & Endocrinology, 1 (2): 106–114, 2013. doi:10.1016/S2213-8587(13)70046-9.

2. Edward W. Gregg, Carol M. Mangione, Jane A. Cauley, Theodore J. Thompson, Ann V. Schwartz, Kristine E. Ensrud, and Michael C. Nevitt. Diabetes and incidence of functional disability in older women. Diabetes Care, 25 (1): 61–67, 2002. doi:10.2337/diacare.25.1.61.

3. Mandip S. Dhamoon, Yeseon Park Moon, Myunghee Cho Paik, Ralph L. Sacco, and Mitchell S. V. Elkind. Diabetes predicts long-term disability in an elderly urban cohort: the Northern Manhattan Study. Annals of Epidemiology, 24 (5): 362–368.e1, 2014. doi:10.1016/j.annepidem.2013.12.013.

4. Nathalie de Rekeneire, Helaine E. Resnick, Ann V. Schwartz, Ronald I. Shorr, Lewis H. Kuller, Eleanor M. Simonsick, Bruno Vellas, and Tamara B. Harris. Diabetes is associated with subclinical functional limitation in nondisabled older individuals: the Health, Aging, and Body Composition Study. Diabetes Care, 26 (12): 3257–3263, 2003. doi:10.2337/diacare.26.12.3257.

5. Stefania Maggi, Marianna Noale, Alberto Pilotto, Antonio Tiengo, Paolo Cavallo Perin, and Gaetano Crepaldi. The METABOLIC Study: multidimensional assessment of health and functional status in older patients with type 2 diabetes taking oral antidiabetic treatment. Diabetes & Metabolism, 39 (3): 236–243, 2013. doi:10.1016/j.diabet.2013.02.003.

6. Akiko Nishimura, Chie Masuda, Chiyo Murauchi, Miho Ishii, Yuko Murata, Terumi Kawasaki, Mayumi Azuma, and Shin-ichi Harashima. Relationship between frailty and diabetic pharmacologic therapy in older adults with type 2 diabetes: a cross-sectional study. Drugs & Aging, 41 (6): 531–542, 2024. doi:10.1007/s40266-024-01119-8.

7. Kasia J. Lipska, Joseph S. Ross, Yinghui Miao, Nilay D. Shah, Sei J. Lee, and Michael A. Steinman. Potential overtreatment of diabetes mellitus in older adults with tight glycemic control. JAMA Internal Medicine, 175 (3): 356–362, 2015. doi:10.1001/jamainternmed.2014.7345.

8. Amanda Sonnega, Jessica D. Faul, Mary Beth Ofstedal, Kenneth M. Langa, John W. R. Phillips, and David R. Weir. Cohort profile: the Health and Retirement Study (HRS). International Journal of Epidemiology, 43 (2): 576–585, 2014. doi:10.1093/ije/dyu067.

9. Barbara H. Bardenheier, Edward W. Gregg, Xiaohui Zhuo, Yiling J. Cheng, and Linda S. Geiss. Association of functional decline with subsequent diabetes incidence in U.S. adults aged 51 years and older: the Health and Retirement Study 1998–2010. Diabetes Care, 37 (4): 1032–1038, 2014. doi:10.2337/dc13-2216.

10. Erik von Elm, Douglas G. Altman, Matthias Egger, Stuart J. Pocock, Peter C. Gøtzsche, and Jan P. Vandenbroucke. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. The Lancet, 370 (9596): 1453–1457, 2007. doi:10.1016/S0140-6736(07)61602-X.

11. Ana R. Quiñones, Sheila Markwardt, and Anda Botoseneanu. Diabetes-multimorbidity combinations and disability among middle-aged and older adults. Journal of General Internal Medicine, 34 (6): 944–951, 2019. doi:10.1007/s11606-019-04896-w.

12. Barbara H. Bardenheier, Ji Lin, Xiaohui Zhuo, Mohammed K. Ali, Theodore J. Thompson, Yiling J. Cheng, and Edward W. Gregg. Disability-free life-years lost among adults aged ≥50 years with and without diabetes. Diabetes Care, 39 (7): 1222–1229, 2016. doi:10.2337/dc15-1095.

13. Christine T. Cigolle, Pearl G. Lee, Kenneth M. Langa, Yuo-Yu Lee, Zhiyi Tian, and Caroline S. Blaum. Geriatric conditions develop in middle-aged adults with diabetes. Journal of General Internal Medicine, 26 (3): 272–279, 2011. doi:10.1007/s11606-010-1510-y.

14. American Diabetes Association Professional Practice Committee for Diabetes. 13. Older adults: Standards of Care in Diabetes—2026. Diabetes Care, 49 (Suppl 1): S277–S296, 2026. doi:10.2337/dc26-S013.

15. Rita Rastogi Kalyani, Christopher D. Saudek, Frederick L. Brancati, and Elizabeth Selvin. Association of diabetes, comorbidities, and A1C with functional disability in older adults: results from the National Health and Nutrition Examination Survey (NHANES), 1999–2006. Diabetes Care, 33 (5): 1055–1060, 2010. doi:10.2337/dc09-1597.

16. Job G. Godino, Lawrence J. Appel, Alden L. Gross, Jennifer A. Schrack, Christina M. Parrinello, Rita R. Kalyani, B. Gwen Windham, James S. Pankow, Stephen B. Kritchevsky, Karen Bandeen-Roche, and Elizabeth Selvin. Diabetes, hyperglycemia, and the burden of functional disability among older adults in a community-based study. Journal of Diabetes, 9 (1): 76–84, 2017. doi:10.1111/1753-0407.12386.

17. Anastasia-Stefania Alexopoulos, Anna R. Kahkoska, Virginia Pate, Marie C. Bradley, Joshua D. Niznik, Carolyn T. Thorpe, Til Stürmer, and John B. Buse. Deintensification of treatment with sulfonylurea and insulin after severe hypoglycemia among older adults with diabetes. JAMA Network Open, 4 (11): e2132215, 2021. doi:10.1001/jamanetworkopen.2021.32215.

18. Rebecca K. Yau, Elsa S. Strotmeyer, Helaine E. Resnick, Deborah E. Sellmeyer, Kenneth R. Feingold, Jane A. Cauley, Eric Vittinghoff, Nathalie de Rekeneire, Tamara B. Harris, Michael C. Nevitt, Steven R. Cummings, Ronald I. Shorr, and Ann V. Schwartz. Diabetes and risk of hospitalized fall injury among older adults. Diabetes Care, 36 (12): 3985–3991, 2013. doi:10.2337/dc13-0429.

19. Kate L. Lapane, Bill M. Jesdale, Catherine E. Dubé, Camilla B. Pimentel, and Swapnil N. Rajpathak. Sulfonylureas and risk of falls and fractures among nursing home residents with type 2 diabetes mellitus. Diabetes Research and Clinical Practice, 109 (2): 411–419, 2015. doi:10.1016/j.diabres.2015.05.009.

20. UK Prospective Diabetes Study (UKPDS) Group. Intensive blood-glucose control with sulphonylureas or insulin compared with conventional treatment and risk of complications in patients with type 2 diabetes (UKPDS 33). The Lancet, 352 (9131): 837–853, 1998. doi:10.1016/S0140-6736(98)07019-6.

21. Dawn E. Alley and Virginia W. Chang. The changing relationship of obesity and disability, 1988–2004. JAMA, 298 (17): 2020–2027, 2007. doi:10.1001/jama.298.17.2020.

22. Dean Schillinger, Kevin Grumbach, John Piette, Frances Wang, Dennis Osmond, Carolyn Daher, Jorge Palacios, Gabriela Diaz Sullivan, and Andrew B. Bindman. Association of health literacy with diabetes outcomes. JAMA, 288 (4): 475–482, 2002. doi:10.1001/jama.288.4.475.

23. Kerri Cavanaugh, Mary Margaret Huizinga, Kenneth A. Wallston, Tebeb Gebretsadik, Ayumi Shintani, Dianne Davis, Rebecca Pratt Gregory, Lynn Fuchs, Robb Malone, Andrea Cherrington, Michael Pignone, Darren A. DeWalt, Tom A. Elasy, and Russell L. Rothman. Association of numeracy and diabetes control. Annals of Internal Medicine, 148 (10): 737–746, 2008. doi:10.7326/0003-4819-148-10-200805200-00006.

24. Elbert S. Huang, Qi Zhang, Niren Gandra, Marshall H. Chin, and David O. Meltzer. The effect of comorbid illness and functional status on the expected benefits of intensive glucose control in older patients with type 2 diabetes: a decision analysis. Annals of Internal Medicine, 149 (1): 11–19, 2008. doi:10.7326/0003-4819-149-1-200807010-00005.

25. Jonathan Z. Weiner, Anjali Gopalan, Pranita Mishra, Kasia J. Lipska, Elbert S. Huang, Neda Laiteerapong, Andrew J. Karter, and Richard W. Grant. Use and discontinuation of insulin treatment among adults aged 75 to 79 years with type 2 diabetes. JAMA Internal Medicine, 179 (12): 1633–1641, 2019. doi:10.1001/jamainternmed.2019.3759.

26. Daniela Corno and Rachel J. Burns. Loneliness and functional limitations among older adults with diabetes: comparing directional models. Journal of Psychosomatic Research, 154: 110740, 2022. doi:10.1016/j.jpsychores.2022.110740.

27. Daniela Pírela and Rajesh Garg. De-intensification of diabetes treatment in elderly patients with type 2 diabetes mellitus. Endocrine Practice, 25 (12): 1317–1322, 2019. doi:10.4158/EP-2019-0303.