Sociodemographic, Clinical, and Psychosocial Predictors of Short- and Long-term Study Retention in the Diabetes Prevention Program (DPP) Outcomes Study (DPPOS)

Abstract

OBJECTIVE

Success of longitudinal studies depends on retention of participants. We examined characteristics as predictors of retention among participants with prediabetes and type 2 diabetes (T2D) in the Diabetes Prevention Program (DPP) and the follow-up DPP Outcomes Study.

RESEARCH DESIGN AND METHODS

A total of 3,234 adults at high risk of T2D joined the DPP (1996–1999, mean age 51 ± 10 years). They were randomized to lifestyle, metformin, or placebo intervention, and then followed through 2020. Logistic regression models estimated the association between baseline sociodemographic, clinical and psychosocial characteristics (life events, family functioning, social support), and short-term retention (∼3 years). Cox proportional hazards models, censoring at death, estimated the association between baseline and time-varying characteristics and time to dropout over the entire 20 years of follow-up.

RESULTS

Among surviving participants (n = 3,218), 93% were retained after 3 years, and 75% of those surviving remained engaged over 20 years. Younger age was associated with dropout during DPP and over 20 years of follow-up. Female sex, non-White race and ethnicity, employment, and lack of baseline depressive symptoms were associated with better long-term retention. Over time, better health state (SF-36) (hazard ratio [HR]: 0.89 per 0.1 point; 95% CI: 0.83–0.95) was associated with retention. Greater BMI (HR: 1.06 per 5 kg/m²; 95% CI: 1.00–1.12), more recent life events (HR: 1.08; 95% CI: 1.02–1.14), and depressive symptoms (HR: 1.11 per 5 points; 95% CI: 1.05–1.18) were associated with reduced retention. Among adults 45–59 years of age at baseline, development of T2D was associated with better retention (HR: 0.75; 95% CI: 0.58–0.97).

CONCLUSIONS

Twenty-year retention of a racially and geographically diverse cohort with prediabetes is possible. Retention was associated with age, psychosocial factors, T2D development, and BMI.

Graphical Abstract

Introduction

Participant retention is essential in longitudinal clinical studies. Attrition threatens the internal validity, weakens statistical power to detect differences between groups (1), reduces the generalizability of findings, and may have financial and ethical implications. Cochrane review guidelines for conducting systematic reviews and meta-analyses include retention levels as a specific criterion for evaluating trial quality (2). In studies with event-driven analysis approaches, retaining the maximum number of participants at risk is critically important because even a small amount of dropout has the potential to impact validity negatively (3). For these reasons, identifying characteristics of participants at risk for dropout and applying targeted retention strategies to keep those individuals actively engaged has emerged as an important strategy for study success. These strategies are especially important to consider among participants that have higher attrition risk, such as older adults and people experiencing chronic health conditions.

Overall retention rates of longitudinal cohorts (4–6) and retention strategies have been previously described (7–10), as have participant-reported motivators and barriers to participation (11). However, detailed participant-level factors associated with an increased risk of dropout are rarely reported (12,13), providing little insight into predictors of retention. Little is known about the effect of social determinants of health (SDOH) on retention in longitudinal studies. Health inequities influenced by SDOH are associated with a higher risk of chronic diseases and their complications, exposing a potential threat to the validity of the results if there is variation in attrition by sociodemographic and psychosocial characteristics.

The Diabetes Prevention Program (DPP, 1996–2001) enrolled a racially and geographically diverse cohort of adults at high risk for developing type 2 diabetes. Together, DPP and its Outcomes Study (DPPOS, 2002–2020) provide over 20 years of longitudinal follow-up. In other prevention projects, participant characteristics associated with short-term retention varied but generally included older age, female sex, non-Hispanic White race and ethnicity, and higher household income (13–15). Conversely, in long-term studies of elderly participants, the oldest participants were less likely to attend in-person visits (12). In DPP, older age was associated with better medication treatment adherence (16), and older age, lower BMI, and non-Hispanic White race and ethnicity were associated with better adherence to the lifestyle intervention (17).

Additional evidence on participant-level factors, including SDOH, related to retention in longitudinal studies is needed to aid researchers in identifying participants most at risk for loss to follow-up. We aimed to identify participant characteristics associated with short- and long-term study retention. These findings may inform the design and management of future longitudinal clinical studies of aging adults to facilitate support strategies for participants at risk for loss to follow-up.

Research Design and Methods

DPP and DPPOS Study Designs

The study protocols for DPP and DPPOS are publicly available at https://dppos.bsc.gwu.edu/web/dppos/about, and the design and methods for both DPP and DPPOS are detailed elsewhere (18,19). Briefly, DPP was a multicenter, randomized controlled clinical trial that recruited 3,234 participants (45% from ethnic and racial minoritized groups disproportionately affected by type 2 diabetes) from 27 clinical centers across the U.S. (1996–1999). Eligible participants were ≥25 years old, had a BMI ≥24 kg/m² (≥22 kg/m² for Asian/Pacific Islanders), and had prediabetes (fasting plasma glucose 95–125 mg/dL, or ≤125 mg/dL in the American Indian centers; 2-h glucose 140–199 mg/dL). DPP participants were randomly assigned to receive an intensive lifestyle intervention (ILS group), metformin (MET group), or a placebo pill. ILS participants were offered an individualized 16-lesson curriculum over 24 weeks, followed by monthly sessions throughout DPP (20). MET participants were assigned to take blinded 850 mg metformin twice daily; placebo participants were assigned a matching placebo pill twice daily (18). Participants were followed for an average of 3.2 years.

Given the efficacy of the interventions, masking of medication was terminated, and participants’ groups were disclosed in August 2001. All participants were then offered the ILS curriculum (20) in group format during a 6-month bridge period and invited to participate in the long-term follow-up study (DPPOS, 2002 to present). All DPPOS participants were offered quarterly group sessions to reinforce weight and activity goals. ILS participants were offered additional booster lifestyle sessions, twice annually. Metformin was continued open label for the MET group. DPPOS was divided into three 5- to 7-year duration phases, each with different scientific goals. The protocols were approved by the local institutional review boards of the participating sites, and all participants provided informed consent.

Study Participants

Participants who were alive at the end of DPP (31 July 2001) were included in the current analysis (n = 3,218), representing over 99% of the initial cohort.

Outcome Measures

Short-term retention (dropout during DPP, over ∼3 years) was defined dichotomously as having attended a scheduled visit within the previous 5 months at the close of DPP, corresponding to the definition used previously (18). During DPP, in-person visits were held quarterly.

Long-term retention (dropout during DPPOS, over ∼20 years) was defined dichotomously as enrollment in the third phase of DPPOS (beginning 2015, approximately 16–19 years after randomization, median 17 years). During DPPOS, participants attended semiannual visits in person or by phone. In time-to-event analyses, the date of last study visit attended prior to loss to follow-up was used through 23 February 2020 (DPPOS-3 data lock). To account for death, participants were censored at the date of death if they had attended a visit within the year prior to the date of death. Participants who remained engaged were administratively censored 20 years after randomization.

Demographic and Clinical Covariates

Self-reported age (years), sex (male/female), five-category race and ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, American Indian, Asian/Pacific Islander), education (high school or lower vs. some college), marital status (never married, living together, married, separated, divorced, widowed), employment status (full- or part-time, retired, homemaker, not employed, other), income group (<$35,000/year, $35,000/year to less than $75,000/year, ≥$75,000/year, refused) and household size (1, 2, 3–4, ≥5) were collected at baseline.

Glycated hemoglobin (HbA_1c) levels were measured annually. Incident diabetes was identified annually (oral glucose tolerance test) and semiannually (fasting glucose), with values confirmed (18). To calculate BMI (kg/m²), weight was measured annually by trained personnel, and height was measured at baseline. Depressive symptoms were measured using the Beck Depression Inventory (BDI; range: 0–63) (21). The six-dimensional health state short-form (SF-6D) health utility index derived from the 36-item Short Form Health Survey (SF-36) was included as a health-related quality-of-life measure (22). The six dimensions are physical functioning, role limitations, social functioning, pain, mental health, and vitality. The SF-6D is a continuous measure, scored on a 0.29–1.00 scale, with 1.00 indicating optimal health. Serious adverse events (SAEs) were collected at each visit and between visits.

Psychosocial Characteristics

During DPP, self-administered questionnaires about recent life events (Life Events Index), perceived social support (Social Provisions Scale [SPS]) (23), and family functioning (Family Household Assessment) (24) were collected at the end of the study screening, 6 months after baseline, and annually throughout the DPP clinical trial period (approximately three to four times per participant). These questionnaires were not administered during DPPOS.

The Life Events Index, based on the Social Readjustment Rating Questionnaire (SRRQ) (25), was collected as a measure of exposure to stress via a 20-item life event battery (Supplementary Material). Participants were asked about life events, including both events likely to be desirable (e.g., marriage, baby) and events most likely to be undesirable (e.g., death of someone close, serious arguments, divorce), as well as some that could be perceived as either (e.g., big changes at work, changes in surroundings). Participants were asked to rate the perceived impact of the event on their lives (e.g., happened, bad effect; happened, but no effect; happened, good effect; did not happen). These results were further summarized dichotomously as occurred or not. To create a summary value, the number of events reported were summed, and, separately, the number of events with a “good” and “bad” effect were summed.

SPS was collected as a measure of perceived social support (23). SPS is a self-administered questionnaire including 24 items rated by the respondent on a Likert-type scale with responses of strongly disagree, disagree, agree, and strongly agree. The items fall into six subscales (4–16 points each): Attachment, Social Integration, Reassurance of Worth, Reliable Alliance, Guidance, and Opportunity for Nurturance. A total social support score (24–96 points) was computed by adding all six of the individual provision scores. A higher score suggests greater perceived social support.

The McMaster Family Assessment Device (FAD) was collected as a measure of family functioning. The FAD is a 60-item self-administered questionnaire that measures an individual’s perceptions of their family (24,26,27). Each item is rated by the respondent on a Likert-type scale to describe the extent to which the statement describes the participant’s family. DPP used a subset of the questions comprising the Problem Solving and General Functioning subscales, and the two scales were summed to obtain a total score (higher score indicating less healthy family functioning). The questionnaire was completed by participants currently living in a household and thus was included in only a subset of analyses.

Statistical Analyses

Descriptive statistics were used to examine the baseline characteristics of the study population among those surviving at the end of DPP (∼3 years) and at the third phase of DPPOS (∼17 years).

Logistic regression models estimated the association of baseline sociodemographic, clinical and psychosocial characteristics (life events, social support, family functioning), and retention during DPP (∼3 years). Covariates were added to the models in three sets (sociodemographic, clinic, psychosocial). Treatment group was included in all models. Model 1 (base model) assessed the relationship between sociodemographic characteristics (age, sex, race and ethnicity, education, income group, marital history, household size, and employment) and retention during DPP. Model 2 added clinical characteristics (BMI, HbA_1c, depressive symptoms, health state). Model 3 added psychosocial characteristics (social support, life events). Lastly, Model 4 added family functioning but is limited to participants reporting living in a household at baseline.

Cox proportional hazard regression models estimated the association of baseline and time-varying participant characteristics (sociodemographic, clinical, psychosocial) with time to study dropout over 20 years of follow-up (long-term retention). The event date for the analysis was the date of the last participant visit prior to loss to follow-up. Participant characteristics were modeled as baseline only (assessing exposure before enrollment), and select characteristics were modeled as time-varying covariates to model changes during DPP (life events, perceived social support, family functioning) or throughout 20 years of follow-up during DPP and DPPOS (BMI, diabetes, HbA_1c). Covariates were added in a fashion similar to the logistic models: model 1 includes sociodemographic characteristics, model 2 added clinical characteristics, model 3 added psychosocial characteristics, and model 4 added family functioning and is limited to participants living in a household.

The time-varying models include treatment group and the baseline sociodemographic characteristics in all models that separately test the effect of time-varying clinical characteristics (BMI, HbA_1c, diabetes status, depressive symptoms, SF-6D, cumulative SAEs) and psychosocial characteristics (social support, life events). For all models, treatment group, age, and sex were tested as potential effect modifiers, and models were stratified if a significant interaction term was present. The assumption of proportional hazards was evaluated for each model by using Schoenfeld residuals. R version 4.2.1 was used for all analyses; the “survival” package was used to fit the Cox models (28,29). Statistical tests were two-sided, with a significance level of 0.05 without adjustment for multiple testing.

Ethics Statement

The study was conducted in accordance with the guidelines of the Declaration of Helsinki and all procedures involving research study participants. Prior to initiating the study protocol, each participant provided written informed consent, and each study center obtained approval from its respective institutional review board. The trials were registered at ClinicalTrials.gov (Diabetes Prevention Program: NCT00004992; Diabetes Prevention Program Outcomes Study: NCT00038727; Diabetes Prevention Program Outcomes Study AD/ADRD: NCT05704309).

Data and Resource Availability

DPP/DPPOS has provided the protocols and lifestyle and medication intervention manuals to the public through its public website (https://www.dppos.org). DPPOS abides by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) data-sharing policy and implementation guidance as required by the National Institutes of Health (NIH)/NIDDK (https://repository.niddk.nih.gov/studies/dppos/).

Results

At the end of DPP, after an average of 3.2 years, 16 participants had died; 93% (n = 2,976) of surviving participants (n = 3,218) were retained and had attended a scheduled visit within the previous 5 months. During DPP, 91.5% of participants attended at least 80% of the expected annual and semiannual visits. At the beginning of the long-term follow-up study (DPPOS-1), 86.3% (n = 2,766) of the surviving participants (n = 3,206) enrolled. Subsequently, 80.2% (n = 2,493 of 3,110) enrolled in DPPOS-2 ∼10 years after randomization, and 76.2% (n = 2,259 of 2,966) enrolled in DPPOS-3, 17 years after randomization. Overall, 2,779 of the originally randomized 3,234 (85.9%) participants enrolled in at least one phase of the DPPOS follow-up study (Supplementary Table 4).

Participant Characteristics

The distribution of select baseline sociodemographic, clinical, and psychosocial characteristics among participants surviving at the end of DPP (∼3 years after randomization) is shown in Table 1. At baseline, participants were an average of 50 years of age, 68% female, and 45% identifying as non-White race and ethnicity. Participants reported an average of three recent (within past 12 months) life events (mean ± SD: 3.3 ± 2.5) at baseline, and the number reported annually decreased throughout DPP to 2.6 ± 2.4 by the last report (P < 0.001). Of those events, 1.1 ± 1.6 had a perceived “bad effect” and 0.9 ± 1.3 had a perceived “good effect.” At baseline, the total social support score was 82.1 ± 9.7, and this decreased slightly by the end of DPP (∼3 years) to 81.4 ± 9.8 (P < 0.001). Among participants living in a household at baseline, the average family problem-solving and general functioning scores were 1.8 ± 0.5 and 1.7 ± 0.5, respectively, and both increased slightly (P = 0.008, P = 0.049) throughout DPP, with ∼15% and ∼24% in the unhealthy family functioning range for problem solving (>2.2) and general functioning (>2.0), respectively.

Table 1.

Baseline participant characteristics among cohort surviving at end of DPP

	Baseline	End of DPP*
n		3,218
Treatment, n (%)
Lifestyle	1,076 (33.4)	---
Metformin	1,065 (33.1)	---
Placebo	1,077 (33.5)	---
Age (years)	50.5 ± 10.6	---
Age-group, n (%)
25–44 years	998 (31.0)	---
45–59 years	1,583 (49.2)	---
≥60 years	637 (19.8)	---
Sex, n (%)
Male	1,031 (32.0)	---
Female	2,187 (68.0)	---
Race and ethnicity, n (%)
NH White	1,754 (54.5)	---
NH Black	643 (20.0)	---
Hispanic	508 (15.8)	---
Am Indian	171 (5.3)	---
Asian/Pac Isl	142 (4.4)	---
Education, n (%)
High school or less	831 (25.8)	---
Some college or more	2,387 (74.2)	---
Annual income, n (%)
<$35,000	1,003 (31.2)	---
$35,000 to <$75,000	1,282 (39.9)	---
≥$75,000	678 (21.1)	---
Refused	254 (7.9)	---
Marital history, n (%)
Never married	418 (13.0)	---
Living together	125 (3.9)	---
Married	1,987 (61.7)	---
Separated	91 (2.8)	---
Divorced	449 (14.0)	---
Widowed	148 (4.6)	---
Household size, n (%)
1	540 (16.8)	---
2	1,060 (32.9)	---
3–4	1,158 (36.0)	---
≥5	460 (14.3)	---
Employment, n (%)
Full or part time	2,393 (74.4)	---
Retired	413 (12.8)	---
Homemaker	204 (6.3)	---
Not employed	121 (3.8)	---
Other	87 (2.7)	---
BMI, kg/m2	34.0 ± 6.7	33.2 ± 7.1
HbA1c	5.9 ± 0.5	6.0 ± 0.8
Fasting glucose, mg/dL	106.5 ± 8.3	109.1 ± 19.5
Weight, kg	94.2 ± 20.3	92.1 ± 21.2
Waist circumference, cm	105.0 ± 14.5	103.8 ± 14.9
Beck Depression Inventory	4.6 ± 4.6	3.9 ± 5.2
Beck Anxiety	4.0 ± 5.0	3.8 ± 5.3
SF-6D	0.8 ± 0.1	0.7 ± 0.1
Life events
Life events reported, n	3.3 ± 2.5	2.6 ± 2.4
Life events, good effect, n	0.9 ± 1.3	0.6 ± 1.1
Life events, bad effect, n	1.1 ± 1.6	1.0 ± 1.5
Social support
Guidance subscale	14.0 ± 2.1	13.9 ± 2.0
Reassurance of worth subscale	13.5 ± 1.9	13.4 ± 1.9
Social integration subscale	13.5 ± 1.9	13.4 ± 1.9
Attachment subscale	13.8 ± 2.1	13.7 ± 2.1
Nurturance subscale	13.1 ± 2.3	12.9 ± 2.3
Reliable alliance subscale	14.1 ± 1.9	14.0 ± 1.9
Total social support	82.1 ± 9.7	81.4 ± 9.8
Family functioning
Living in family household, n (%)
No	558 (17.3)	559 (17.4)
Yes	2,660 (82.7)	2,656 (82.6)
Problem-solving score	1.8 ± 0.5	1.9 ± 0.5
General functioning score	1.7 ± 0.5	1.8 ± 0.5

Am, American; NH, non-Hispanic; Pac Isl, Pacific Islander. *Includes data from each participant’s final visit within DPP, on average ∼3 years after randomization.

Association of Baseline Characteristics and Dropout During DPP (3 Years)

Supplementary Table 1 summarizes the association between baseline participant characteristics (life events, social support, family functioning) and dropout during DPP. Across all four models, younger age was the only factor significantly associated with short-term dropout (Supplementary Table 1). In the fully adjusted model (model 3), every 10-year increase in age was associated with a 35% decrease in the odds of dropout (odds ratio: 0.64; 95% CI: 0.52, 0.79). Number of recent life events, social support, and family functioning were not associated with dropout during DPP.

Association of Baseline Characteristics and Dropout Over 20 Years (DPPOS)

Table 2 summarizes the association between baseline participant characteristics and dropout over 20 years defined by time to last visit attended (i.e., dropout) over 20 years of follow-up. Figure 1 shows the probability of loss to follow-up by age-group, with the youngest age-group (25–44 years old at baseline) having the highest probability of dropout (P < 0.001). Supplementary Figure 1 shows the probability of loss to follow-up by age quartiles and quintiles. Consistent with the DPP dropout results, younger age was associated with dropout over 20 years. In the fully adjusted model (model 3), every 10-year difference in baseline age was associated with a 33% reduction in the risk of long-term dropout (HR: 0.77 [0.70, 0.86]). In addition, women had a 22% lower risk of long-term dropout compared with men (HR: 0.78 [0.65, 0.93]). Compared with non-Hispanic White race and ethnicity, participants reporting non-Hispanic Black and American Indian race and ethnicity had a 19% and 58% lower risk of long-term dropout (P < 0.001), respectively. Conversely, participants who were retired, homemakers, not employed, or with other employment had an increased risk of long-term dropout compared with participants who reported full- or part-time employment at baseline (P = 0.016). Furthermore, every 5 kg/m² of BMI was associated with a 7% increase in the risk of long-term dropout (HR: 1.07 [1.01, 1.13]), and higher scores on BDI were also associated with an increased risk of long-term dropout (P = 0.022). None of the psychosocial characteristics at baseline were associated with dropout over 20 years. HRs were similar when the family functioning score was added and the sample size was limited to participants who reported living in a household at baseline (model 4).

Table 2.

Association of baseline participant characteristics for dropout during DPPOS (time to last visit over 20 years), among cohort surviving at end of DPP

Characteristic	Model 1			Model 2			Model 3			Model 4*
	HR	95% CI	P value	HR	95% CI	P value	HR	95% CI	P value	HR	95% CI	P value
Treatment group			0.6			0.7			0.7			0.7
Lifestyle	—	—		—	—		—	—		—	—
Metformin	1.06	0.90, 1.25		1.06	0.90, 1.25		1.07	0.89, 1.28		1.07	0.89, 1.28
Placebo	0.98	0.83, 1.16		1.00	0.85, 1.19		0.99	0.82, 1.20		0.99	0.82, 1.20
Age (per 10 years)	0.73	0.67, 0.80	<0.001	0.76	0.69, 0.83	<0.001	0.77	0.70, 0.86	<0.001	0.78	0.70, 0.86	<0.001
Female sex (ref: male)	0.88	0.75, 1.02	0.094	0.83	0.71, 0.97	0.024	0.78	0.65, 0.93	0.007	0.79	0.66, 0.94	0.009
Race and ethnicity			<0.001			<0.001			<0.001			<0.001
NH White	—	—		—	—		—	—		—	—
NH Black	0.79	0.66, 0.95		0.79	0.64, 0.96		0.81	0.65, 1.03		0.81	0.64, 1.02
Hispanic	0.94	0.77, 1.15		0.96	0.79, 1.18		1.01	0.81, 1.25		1.00	0.80, 1.24
Am Indian	0.39	0.26, 0.59		0.38	0.25, 0.58		0.42	0.27, 0.64		0.41	0.27, 0.64
Asian/Pac Isl	0.78	0.54, 1.11		0.83	0.58, 1.21		0.87	0.59, 1.28		0.85	0.57, 1.25
College (ref: high school)	0.91	0.77, 1.08	0.3	0.91	0.77, 1.08	0.3	0.91	0.76, 1.09	0.3	0.92	0.76, 1.10	0.4
Income group			0.2			0.13			0.2			0.2
<$35,000	—	—		—	—		—	—		—	—
$35,000 to <$75,000	0.92	0.77, 1.10		0.94	0.79, 1.13		0.90	0.73, 1.11		0.90	0.73, 1.11
≥$75,000	0.93	0.74, 1.17		0.95	0.75, 1.20		0.93	0.72, 1.20		0.94	0.73, 1.22
Refused	1.20	0.93, 1.55		1.28	0.99, 1.66		1.20	0.90, 1.61		1.21	0.91, 1.63
Marital history			0.058			0.11			0.4			0.4
Never married	—	—		—	—		—	—		—	—
Living together	1.27	0.86, 1.88		1.37	0.92, 2.04		1.30	0.84, 2.01		1.32	0.85, 2.04
Married	1.10	0.85, 1.42		1.15	0.89, 1.48		1.07	0.78, 1.47		1.07	0.78, 1.47
Separated	1.74	1.18, 2.58		1.71	1.15, 2.55		1.63	1.00, 2.64		1.60	0.99, 2.60
Divorced	1.26	0.97, 1.64		1.26	0.97, 1.65		1.18	0.80, 1.72		1.18	0.81, 1.73
Widowed	1.43	0.98, 2.10		1.35	0.91, 2.00		1.09	0.59, 2.00		1.07	0.58, 1.97
Household size			0.4			0.2			0.8			0.8
1	—	—		—	—		—	—		—	—
2	0.86	0.68, 1.09		0.84	0.66, 1.07		0.99	0.54, 1.80		0.95	0.52, 1.73
3–4	0.80	0.63, 1.03		0.77	0.60, 0.99		0.91	0.50, 1.67		0.88	0.48, 1.62
≥5	0.87	0.65, 1.18		0.86	0.63, 1.16		1.00	0.54, 1.86		0.97	0.52, 1.81
Employment			0.003			0.004			0.016			0.018
Full or part time	—	—		—	—		—	—		—	—
Retired	1.45	1.12, 1.86		1.43	1.10, 1.84		1.40	1.04, 1.87		1.40	1.04, 1.88
Homemaker	1.29	0.98, 1.70		1.31	1.00, 1.73		1.36	1.02, 1.80		1.34	1.01, 1.78
Not employed	1.29	0.93, 1.79		1.19	0.85, 1.67		1.16	0.79, 1.70		1.16	0.79, 1.69
Other	1.64	1.14, 2.35		1.68	1.17, 2.42		1.62	1.09, 2.39		1.61	1.09, 2.38
BMI (per 5 kg/m2)				1.09	1.03, 1.14	0.003	1.07	1.01, 1.13	0.034	1.07	1.00, 1.13	0.038
HbA1c, %				0.98	0.85, 1.14	0.8	0.97	0.82, 1.14	0.7	0.96	0.82, 1.13	0.6
Depressive symptoms (BDI)				1.02	1.00, 1.04	0.015	1.02	1.00, 1.04	0.022	1.03	1.01, 1.05	0.014
SF-6D				1.11	0.44, 2.75	0.8	0.92	0.34, 2.55	0.9	0.92	0.34, 2.54	0.9
Social support total							1.00	0.99, 1.01	0.6	1.00	0.99, 1.01	0.6
Life events, bad effect, n							1.02	0.96, 1.08	0.5	1.02	0.96, 1.08	0.5
Life events, good effect, n							1.00	0.95, 1.06	>0.9	1.01	0.96, 1.06	0.8
Family functioning										0.91	0.82, 1.01	0.074

n = 3,218. Shown are Cox proportional HR and 95% CI. Bold text indicates P < 0.05. Outcome is time to last visit attended (i.e., dropout). Censoring at death date if participant attended a visit within previous 1 year of date of death. Administrative censoring of all remaining participants at 20 years since randomization. HR >1 signifies greater likelihood of dropout, that is, association with worse retention. Conversely, HR <1 signifies lower likelihood of dropout, that is, association with better retention. Am, American; NH, non-Hispanic; Pac Isl, Pacific Islander. *Among participants reporting living in a household at baseline (n = 2,648).

Figure 1.

Probability of loss to follow-up by age-group over 20 years.

Association of Characteristics Over Time and Long-term Dropout

Figure 2 summarizes Cox models evaluating the association of select time-varying clinical and psychosocial characteristics (i.e., repeated measures) among cohort survivors at the end of DPP with the risk of long-term dropout, adjusted for treatment group and baseline sociodemographic factors. As with baseline BMI, every 5 kg/m² increase of BMI over time was associated with a 6% increase in the risk of long-term dropout (HR: 1.06 [1.00, 1.12]), and every 5-unit increase in BDI score was associated with an 11% increase in the risk of long-term dropout (HR: 1.11 [1.05, 1.18]) (Supplementary Tables 2 and 3). Time-varying HbA_1c levels were not associated with the risk of dropout. Among the psychosocial characteristics, both number of recent life events with a “bad effect” and those with a “good effect” were associated with an increased risk of dropout (HR: 1.08 [1.02, 1.14] and HR: 1.05 [1.00, 1.09], respectively); however, perceived social support was not. Only health state (SF-6D) was associated with a decreased risk of dropout (HR: 0.89 [0.83, 0.95]); a 0.1-unit (∼1 SD) increase in SF-6D was associated with an 11% reduction in the risk of dropout. Cumulative number of SAEs was associated with an 11% increase in the risk of dropout per SAE reported (HR: 1.11 [1.04, 1.19]).

Figure 2.

Association of time-varying characteristics for long-term dropout (time to last visit over 20 years), among cohort surviving at end of DPP, n = 3,218. Shown are Cox proportional HR and 95% CI. Outcome is time to last visit attended (i.e., dropout). Censoring at death date if participant attended a visit within previous 1 year of date of death. Administrative censoring of all remaining participants at 20 years since randomization. HR >1 signifies greater likelihood of dropout, that is, association with worse retention. Conversely, HR <1, signifies lower likelihood of dropout, that is, association with better retention. A: Each time-varying characteristic modeled separately, adjusted for select baseline characteristics treatment group, age (per 10 years), sex, race and ethnicity, education, income group, marital history, household size, and employment. *Entered into model simultaneously to adjust for one another. B: A significant interaction was observed between time-varying diabetes status and age (P = 0.03); however, the interaction with time-varying BMI did not reach statistical significance (P = 0.11). Diabetes status and BMI were modeled simultaneously to adjust for one another. Model also adjusted for sex, race and ethnicity, education, income group, marital history, household size, and employment.

For each of the select time-varying clinical characteristics (BMI, HbA_1c, and diabetes status), interactions with age (years), sex, and treatment group were tested. A significant interaction was observed between diabetes status and age (P = 0.03); however, the interaction with BMI did not reach statistical significance (P = 0.11). Thus, the association of diabetes status over time stratified by age-group (25–44 years, 45–59 years, and ≥60 years) is presented in Fig. 2. When stratified by age-group, there was a significant protective relationship between diabetes and study dropout over 20 years of follow-up among participants aged 45–59 years at baseline. Development of diabetes was associated with a 25% reduction in the risk of dropout over 20 years (HR: 0.75 [0.58, 0.97]) in this age-group; this relationship was not observed in the other two age-groups. When stratified by age-group, we also observed differences for other predictors’ relationships with dropout over 20 years, with some predictors significant in one age-group but not another (Supplementary Table 3).

Conclusions

Retention of a racially and ethnically diverse diabetes-related cohort over nearly two decades of longitudinal follow-up was robust, as evidenced by the high enrollment rates during the first three phases of DPPOS at 5 (DPPOS-1, 86%), 10 (DPPOS-2, 80%), and 17 (DPPOS-3, 76%) years postrandomization. Several characteristics were found to be positively associated with both short-term (∼3 years) and long-term (∼20 years) study retention. We observed that younger age was associated with dropout over 20 years of follow-up, while female sex, American Indian and non-Hispanic Black race and ethnicity, and full- or part-time employment were associated with long-term retention. Our finding of non-White race and ethnicity associated with higher long-term retention is consistent with other studies demonstrating successful retention of minoritized populations (30–32) and contrasts with the prevailing belief that they are more challenging to enroll and retain in research studies. Lower baseline and follow-up BMI, fewer depressive symptoms, and higher health utility (SF-6D) scores over time were associated with long-term retention. Importantly, we did not observe a relationship between treatment group and retention.

We did not observe a relationship between perceived social support or family functioning and retention. Both positive and negative recent life events reported during DPP/DPPOS were associated with worse retention over time. Perhaps the stress associated with life events, whether having a positive or negative effect, distracted participants from engagement. SAEs were associated with an increased risk of dropout. Finally, in analyses stratified by age-group, among those aged 45–59 years at baseline, the failure to prevent diabetes, the study’s major goal, was associated with increased long-term retention. These findings may suggest those who developed diabetes in middle age may have had a heightened awareness of the future risks of diabetes complications, motivating their long-term commitment to study participation. However, further research is needed to understand differences in retention predictors by age-group.

Previous analyses of engagement in DPP have focused on treatment adherence (16,33,34). The characteristics associated with medication adherence were similar to those related to retention in our analysis. During the DPP masked clinical trial, older age and higher income were both associated with medication (metformin and placebo) adherence (16). During the open-label long-term follow-up study, older age, lower depression scores, and higher income were similarly associated with better medication adherence (33). For participants randomized to the lifestyle arm, those who were younger and having greater obesity and were non-White or in a racial and ethnic subgroup lost less weight (17). These earlier findings, paired with the current analysis, suggest shared characteristics among the most engaged and adherent participants.

Similar to our findings, studies of short-term retention in behavioral interventions among adults at high risk of type 2 diabetes or with diabetes have shown that worse retention is consistently associated with younger age (14,35–39). The Look AHEAD Research Group found that younger age, symptoms of depression, and nonmarried status at baseline as well as randomization to the minimal intervention were associated with higher likelihood of missing consecutive study visits during the first 4 years of follow-up (39). In an analysis of three independent 2-year clinical trials of behavioral weight loss, fewer years of education, younger age, and higher BMI were independently associated with attrition; no significant associations with psychosocial characteristics (depressive symptoms, SF-36) were observed (37).

The Atherosclerosis Risk in Communities (ARIC) study investigated factors related to differences in retention among Black and White participants in their prospective cohort between 1987 and 2013 (5). Those who remained engaged in the study had better cardiovascular disease risk factor profiles and a more favorable socioeconomic status at baseline, both of which are also more common in White versus Black participants. Conversely, in DPPOS, we observed differences in long-term retention by race and ethnicity, with non-Hispanic Black and American Indian participants having the lowest likelihood of dropout compared with the non-Hispanic White participants. Composition of race and ethnicity varied by clinic, and some of these findings may be due to variation in the levels of geographic and local clinic community engagement rather than the self-reported race and ethnicity itself.

In contrast, an analysis of long-term retention of older adults in the Cardiovascular Health Study (CHS) found that participants with a recent study visit were more likely to be younger, male, and in good health compared with participants with other visit types (12). Conflicting findings with the association of age may be due to the fact that CHS participants were significantly older (all participants >65 years old at enrollment) than DPPOS participants at baseline. Strotmeyer et al. (12) reported that the differences observed in retention according to age were more pronounced as the study progressed, consistent with the healthy participant bias in the early years of the study. These findings are consistent with the Kaplan-Meier curves illustrated in Fig. 1 and Supplementary Fig. 1, which suggest that participants in the oldest age-group (≥60 years old at baseline) may have a steeper decline in retention around 12–15 years from randomization.

In the clinical setting, in the National DPP (NDPP) of the Centers for Disease Control and Prevention lifestyle change program, retention is low, with half of participants retained at 6 months (15). In an analysis of retention in the year-long NDPP program, Cannon et al. (15) found that lower retention was associated with younger age, consistent with our findings, and minoritized race and ethnicity, in contrast to our findings.

The strengths of the present analysis include the prospective design and longitudinal assessment of both clinical (diabetes, HbA_1c, BMI) and psychosocial (recent life events, social support, and family functioning) characteristics, allowing the inclusion of time-varying covariates. Analyses were adjusted for potential confounding variables. Participants in DPP/DPPOS are racially diverse and have been extensively phenotyped for over 20 years, making findings of the present analyses relatively generalizable in the context of randomized, controlled, and longitudinal studies. Study limitations include the possibility of recall bias, measurement error, and unmeasured confounders. Additionally, DPP only collected limited self-reported individual SDOH, including income, education, and household size. The collection of most psychosocial characteristics was terminated at the end of DPP, and we did not have information about these exposures throughout DPPOS. We also did not differentiate between the relatively small number of those who were lost to follow-up and those who voluntarily withdrew from the study.

Conclusion

At the conclusion of DPP, after an average of 3 years, over 90% of surviving participants were retained. Nearly two decades after randomization, just over 75% remained engaged in the longitudinal follow-up phase, demonstrating that long-term retention of a racially diverse diabetes-related cohort is possible. Long-term retention was greater in older-age participants and in non-Hispanic-Black and American Indian individuals. A variety of psychosocial factors had an adverse effect on retention. Retention of younger participants in research studies requires further investigation to provide strategies for short- and long-term success.

This article contains supplementary material online at https://doi.org/10.2337/figshare.29120558.

Funding Statement

Research reported in this publication was supported by NIDDK of NIH under awards U01 DK048489, U01 DK048339, U01 DK048377, U01 DK048349, U01 DK048381, U01 DK048468, U01 DK048434, U01 DK048485, U01 DK048375, U01 DK048514, U01 DK048437, U01 DK048413, U01 DK048411, U01 DK048406, U01 DK048380, U01 DK048397, U01 DK048412, U01 DK048404, U01 DK048387, U01 DK048407, U01 DK048443, and U01 DK048400, by providing funding during DPP and DPPOS to the clinical centers and the coordinating center for the design and conduct of the study, and collection, management, analysis, and interpretation of the data. Funding was also provided by the National Institute of Child Health and Human Development, the National Institute on Aging, the National Eye Institute, the National Heart Lung and Blood Institute, the National Cancer Institute, the Office of Research on Women’s Health, the National Institute on Minority Health and Health Disparities, the Centers for Disease Control and Prevention, and the American Diabetes Association. The Southwestern American Indian Centers were supported directly by NIDDK, including its Intramural Research Program, and the Indian Health Service. The General Clinical Research Center Program, National Center for Research Resources, and Department of Veterans Affairs supported data collection at many of the clinical centers. Merck KGaA provided medication for DPPOS. DPP/DPPOS also received donated materials, equipment, or medicines for concomitant conditions from Bristol-Myers Squibb; Parke-Davis; LifeScan Inc.; Health-O-Meter; Hoechst Marion Roussel, Inc.; Merck-Medco Managed Care, Inc.; Merck and Co.; Nike Sports Marketing; SlimFast; and Quaker Oats Co. McKesson BioServices, Matthews Media Group, Inc., and the Henry M. Jackson Foundation provided support services under subcontract with the coordinating center. The sponsor of this study was represented on the steering committee and played a part in study design, how the study was done, and publication. All authors in the writing group had access to all data.

The content is solely the responsibility of the authors and does not necessarily represent the official views of NIH.