Factors Associated with Recruitment and Retention in Randomized Controlled Trials of Behavioral Interventions for Patients with Pediatric Type 1 Diabetes

Abstract

The purpose of this study is to describe recruitment and retention experiences from three behavioral randomized controlled trials conducted among youth with type 1 diabetes. Eligibility, recruitment, and retention data were examined. Study-specific differential study participation and loss-to-follow-up analyses assessed the relations of patient characteristics with treatment completion and 6-month retention. Multivariable logistic regression identified factors independently associated with 6-month retention among all participants. Approximately 70–92 % of randomized participants completed treatment and 58–90 % were retained for follow-up. Older patients and non-Caucasian patients were less likely to enroll. Treatment completion and 6-month retention were less likely among youth who were older, had worse baseline glycemic control, lower household income, and/or unmarried parents. Some subgroups of patients are less likely to participate in research and are more susceptible to loss-to-follow-up. More work is needed to understand the facilitators and barriers to research participation.

Background

The randomized controlled trial (RCT) has long been the gold standard of methodological designs used to assess the efficacy and effectiveness of health care interventions (Kane, Wang, & Garrard, 2007; Meldrum, 2000). By randomizing participants to one of two or more groups, RCTs aim to balance extraneous factors among groups allowing researchers to isolate and ascertain the impact of treatment on identified outcome variables (Kane et al., 2007; Meldrum, 2000). However, challenges with the recruitment and retention of participants throughout the duration of data collection pose threats to internal and external validity. Non-random sampling of participants, bias introduced due to differential attrition between groups, and restricted generalizability of findings in “real world” settings are all concerns (Gul & Ali, 2010; Karlson & Rapoff, 2009; Keith, 2001; Patel, Doku, & Tennakoon, 2003).

RCTs that evaluate behavioral interventions with long-term follow-up often face recruitment and retention challenges. Participant retention may falter due to rescheduled/missed clinic appointments, changes in insurance, transfer of medical care to different physicians or facilities, or unanticipated changes in contact information resulting in loss to follow-up (Karlson & Rapoff, 2009). Above and beyond these logistical concerns, trials involving pediatric populations are subject to unique challenges such as hesitance to discuss sensitive medical issues with researchers, feelings of overwhelming burden due to illness management, parental concerns that researchers will judge illness management capabilities, and difficulty finding transportation and/or child care for study visits (Probstfield & Frye, 2011; Rhee, Ciurzynski, & Yoos, 2008; Stehl et al., 2009; Tercyak, Donze, Prahlad, Mosher, & Shad, 2006). Enrolling families of children with chronic illnesses into RCTs often requires substantial effort because both parents and children are involved in the consent process and data collection. Researchers may need to make multiple contacts in order to speak with parents and adolescents, coordinate with work, school, and clinic schedules, and obtain parental informed consent and child assent before participation (Rhee et al., 2008; Stehl et al., 2009; Tercyak et al., 2006).

Low rates of recruitment and retention in RCTs have been noted among a variety of pediatric populations (Germann, Kirschenbaum, & Rich, 2006; Kamps et al., 2008; Krishna, Balas, Francisco, & Konig, 2006; Stark et al., 2005; Stehl et al., 2009; Tercyak et al., 2006; Wysocki et al., 2006). A comprehensive report of recruitment and retention rates among 40 published behavioral RCTs among pediatric populations revealed that only 63 % of patients who were eligible for enrollment participated, and of those patients, 20 % did not complete initial follow-up assessments and 32 % did not complete extended follow-up assessments (Karlson & Rapoff, 2009). These numbers are lower than the recommendations that at least 70 % of all potentially-eligible patients screened should be enrolled to ensure representative samples, and that at least 80 % of all participants should be retained for follow-up to minimize bias (Harris, 1998; Marcellus, 2004; Patel et al., 2003).

Identification of practical strategies to increase the number and diversity of participants who enroll and complete follow-up assessments has been discussed extensively. There is consensus that tailoring recruitment to the population, establishing positive relationships, demonstrating the benefits of participation to the individual and community, and orienting participants to the expectations of the study are important determinants of participant enrollment (Boys et al., 2003; Germann et al., 2006; Gul & Ali, 2010; Karlson & Rapoff, 2009; Marcellus, 2004). Fostering collaborative relationships with the physicians who care for patients during their participation (Patel et al., 2003; Probstfield & Frye, 2011) and with the patient groups is also recommended.

With respect to retention strategies, the NIH Behavioral Change Consortium presented a set of recommendations in 2005 detailing eight categories of retention strategies (Coday et al., 2005). Although these recommendations provide a preliminary guide to the development and planning of RCTs, there is a dearth of information regarding effective strategies for the recruitment and retention challenges commonly encountered with pediatric populations. Without more comprehensive communication among pediatric researchers about effective/ineffective strategies, it is likely that pediatric RCTs will continue to struggle to recruit and retain adequate samples and dissemination of effective clinical interventions will be delayed.

Our research team has conducted three RCTs among youth with type 1 diabetes (T1D) and their caregivers. The objectives of this paper are twofold: (1) to describe recruitment and retention experiences over the course of three RCTs among youth with T1D spanning young childhood through adolescence and (2) to identify the demographic and clinical variables related to participant recruitment and retention. Our analysis of characteristics associated with key milestones in study participation among pediatric patients with T1D will provide behavioral researchers with valuable information that will allow them to make appropriate data-driven assumptions about the essential recruitment and retention parameters which underlie the design and implementation of all RCTs, especially those targeting pediatric populations with chronic disease.

Methods

The participants included in this report were recruited for three separate NIH-funded RCTs for youth with T1D and their caregivers: the Young Child Project, the DECIDE Project, and the TeamWork Project. Each RCT was IRB-approved by participating institutions. Eligibility criteria across the three studies included a diagnosis of T1D (at least 6 months for the Young Child Project and at least 1 year for the DECIDE and TeamWork Projects) and English fluency. Exclusionary criteria for each study were diagnosis of additional chronic medical conditions and/or developmental disorders; an additional exclusionary criteria for the DECIDE Project was absence of menarche in girls. Brief descriptions regarding the participants and procedures for each RCT are presented below. Detailed study procedures are presented in Table 1.

Table 1.

Randomized controlled trial procedures and retention strategies

Procedure	Young Child Project	DECIDE Project	TeamWork Project
Participant identification	Clinic list review/eligibility screener	Clinic list review/eligibility screener	Clinic list review/eligibility screener
Initial contact	Letter mailed home	Letter mailed home	Letter mailed home
Eligibility screener	Phone screener	Phone screener	Phone screener
Baseline assessment	By phone	By mail/phone	In clinic
Written consent/assent	In clinic	By mail	In clinic
Orientation session	In clinic	N/A	N/A
Randomization groups	Intervention or Diabetes Education	Intervention or Standard Care	Intervention or Diabetes Education
Treatment protocol	5 phone sessions (including 1 group call); Internet forum access	2 family sessions in clinic; 1 group session on a weekend; 1 booster call by phone after each session	4 parent-teen sessions in clinic; 1 booster call by phone after each session
Diabetes education protocol	5 phone sessions	N/A	4 parent-teen sessions in clinic
Follow-up assessments	1, 6, and 12 months post-treatment by phone	1, 6, and 12 months post-treatment by mail/phone	3, 6, 9, 12, and 24 months post-treatment in clinic
Relevant retention strategies	In person orientation session; Day/evening/weekend sessions; Phone/clinic assessments; Multiple modes of communication; Increasing incentives for youth and parents ($50 at baseline, $70 at 6-months follow up); Parking vouchers; Thank you letters, Holiday cards; Group call with other participants	Day/evening/weekend/clinic sessions; Phone/in clinic assessments; Multiple modes of communication; Increasing incentives for youth and parents ($10 at baseline, $20 at 6-months follow up); Parking vouchers; Thank you letters, Holiday cards; In person group session with other participants	In clinic sessions; Phone/in clinic assessments; Multiple modes of communication; Increasing incentives for youth and parents ($25 at baseline, $75 at 6-months follow up); Parking vouchers; Thank you letters, Holiday cards

Young Child Project

The Young Child Project was a multi-site RCT designed to investigate the impact of a telephone-based behavioral intervention on parent coping and children’s glycemic control among primary caregivers of young children between the ages of 1–6 years. Caregivers were recruited from tertiary care endocrinology services at two pediatric medical centers in the Mid-Atlantic and one pediatric medical center in the Midwest. Patients were initially screened for eligibility by reviewing medical charts for child age, date of diagnosis, and medical diagnoses. Parents of children who met initial eligibility criteria were mailed letters with information about the project. Subsequently, research assistants called families to confirm the initial eligibility and assess additional eligibility criteria, such as access to a phone and English proficiency.

In total, 134 caregivers completed the baseline questionnaires and were randomized to the behavioral intervention or diabetes education. Once randomized, intervention participants completed 5 parenting support phone sessions with a study-trained interventionist, including a study team-led group conference call with 1–4 other RCT participants. Diabetes education participants completed 5 phone sessions with a study team member who provided basic diabetes information related to management in young children. Following the phone sessions, caregivers in both groups completed 3 follow-up assessments by telephone over a 1 year period (1, 6, and 12 months post-treatment). Each of the three sites completed eligibility screening and recruitment for patients at their site, met with parents for consent and an in-clinic orientation session, and collected glucometer data at follow-up clinic appointments. All baseline phone questionnaires, treatment phone sessions, and follow-up phone assessments were coordinated and completed by the primary study site. Additional details regarding study development and sample characteristics are published elsewhere (Herbert, Monaghan, Cogen, & Streisand, 2014; Monaghan et al. 2014).

DECIDE Project

The DECIDE Project was a RCT designed to prevent the deterioration of self-care behaviors often evident during adolescence (Streisand & Mednick, 2006). Preadolescents between the ages of 8 and 11 years diagnosed with T1D for at least 1 year and their primary caregivers were recruited from tertiary care endocrinology services at one Mid-Atlantic children’s medical center. Medical charts were screened for child age, date of diagnosis, and medical diagnosis to identify patients who were initially eligible for participation. Parents of eligible children were mailed letters with information about the project and research assistants followed up by phone to verify the initial eligibility screening and to assess for additional eligibility criteria, including absence of menarche in girls and English proficiency.

A total of 86 caregiver-preadolescent dyads completed consent and baseline assessments by mail and were randomized to one of two groups: intervention or standard care, with standard care participants given the option to receive the intervention program at the end of the study. Dyads in the intervention group participated in two in-clinic sessions, attended one weekend/evening session with 2–5 other RCT families, and received a booster call from the study team interventionist approximately 2 weeks after each session. Both intervention and standard care participants completed 3 follow-up assessments over a 1 year period (1, 6, and 12 months post-treatment).

TeamWork Project

The TeamWork Project was a multi-site RCT prevention program for youth with T1D who are in the high risk period of early adolescence, when self-care behavioral deterioration occurs (Holmes, Chen, Mackey, Grey, & Streisand, 2014). Adolescents between the ages of 11–14 years, diagnosed with T1D for at least 1 year, and their caregivers were recruited from tertiary care endocrinology services at two Mid-Atlantic children’s medical centers. Families were initially screened for eligibility by reviewing medical charts and/or upcoming clinic visits. Eligible families were mailed letters with information about the project. Research assistants followed up by phone to verify the initial eligibility screening and to assess for additional eligibility criteria, such as length of diagnosis, comorbid medical diagnoses, and English proficiency.

In total, 257 caregiver-adolescent dyads completed the baseline questionnaires and 226 dyads were randomized to one of two treatment comparison groups: coping skills or education. Following randomization, all participants were asked to complete either 4 in-clinic coping skills or education sessions led by a study team counselor that included both a parent and adolescent. One month after each session, participants in the coping skills group completed a booster session with a study team interventionist by phone. Upon completion of the sessions, up to 5 follow-up assessments were conducted over the course of 2 years (3, 6, 9, 12, and 24 months post-treatment; Holmes et al., 2014).

Data Analysis Plan

To better understand the extent to which study participants were representative of the underlying patient population, we first compared selected demographic and clinical characteristics of patients enrolled into any of these three trials to information abstracted from the electronic health records for T1D patients, between the ages of 2 and 15 years, seen at least once during 2011–2012 in the Child and Adolescent Diabetes Program at the single center from which most patients were recruited. This age range was selected because it is consistent with the age ranges of participants across the three projects. We then evaluated differential loss-to-follow-up in a series of study-specific analyses using independent samples t tests and χ² tests to assess the relationship of demographic and clinical variables (e.g., child age; child and caregiver gender and race; caregiver marital status; caregiver age; household income; illness duration; insulin regimen; baseline glycemic control; and treatment group) with treatment completion, and follow-up retention at the 6 month follow-up time point (the time point at which the primary outcomes were assessed for all three studies). In all χ² analysis results, the first parenthetical number indicates the degrees of freedom and the second parenthetical number indicates the total sample size for the analysis. Lastly, we used multivariable logistic regression to identify factors independently associated with 6-month retention among all patients who agreed to participate in any of these 3 RCTs. Candidate variables for this analysis included all common elements that were uniformly collected during all three studies: child age, gender, and race; primary caregiver age, gender, and marital status; time since diagnosis; and baseline glycemic control. Other variables of interest, such as socioeconomic status and insurance were not collected in a uniform manner and could not be included as candidate variables; however, 25 % of the overall clinic population has public insurance. Given the exploratory nature of this work, no adjustments for multiple comparisons were deployed. All analyses were performed in IBM SPSS Statistics Version 22 (IBM, Armonk, NY) and Stata Version 13.1 (StataCorp, College Station, Texas).

Results

Comparison to Clinic Population

In comparison to the underlying clinic population, patients who enrolled in any of the three T1D behavioral intervention RCTs, and those who completed at least a 6 month follow-up assessment, were younger and were more likely to identify as Caucasian than a comparison group of T1D patients of similar age extracted from clinic medical records. As shown in rows 1 and 3 of Table 2, these comparisons were significant (p < .001 for all comparisons).

Table 2.

Comparisons of study participants who enrolled and who were retained through the 6-month follow-up visit compared to the clinic population of T1D patients ages 2–15 in the referent hospital common to all studies

Characteristics	Clinic population N = 927 prevalence or mean (95 % CI)	Young Child Project enrollees N = 134 prevalence or mean (95 % CI)	DECIDE enrollees N = 86 prevalence or mean (95 % CI)	TeamWork enrollees N = 257 prevalence or mean (95 % CI)	All enrolled participants prevalence or mean (95 % CI)	All retained participants prevalence or mean (95 % CI)a	Difference in prevalence or means: enrolled vs. clinic (95 % CI)	p value for difference	Difference in prevalence or means: retained vs. clinic (95 % CI)	p value for difference
Age (years)	11.4 (11.2, 11.6)	5.3 (5.1, 5.6)	10.8 (10.7, 11.0)	12.8 (12.7, 13.0)	10.4 (10.1, 10.7)	10.1 (9.7, 10.4)	−1.0 (−1.4, −0.6)	<.001	−1.3 (−1.8, −0.9)	<.001
Male (%)	51.5 (48.2, 54.7)	50.7 (42.2, 59.3)	46.5 (35.8, 57.3)	49.4 (43.3, 55.6)	49.3 (44.8, 53.8)	50.3 (45.1, 55.5)	−2.2 (−7.7, 3.3)	.45	−1.2 (−7.3, 4.9)	.72
Caucasian race (%)	51.0 (47.7, 54.3)	78.9 (71.9, 86.0)	73.3 (63.7, 82.8)	69.6 (64.0, 75.3)	72.9 (68.9, 76.9)	75.2 (70.7, 79.7)	21.9 (16.7, 27.1)	<.001	24.2 (18.6, 29.7)	<.001
Time since diagnosis or transfer into clinic (years)	4.0 (3.8, 4.2)	2.0 (1.8, 2.2)	4.2 (3.6, 4.8)	5.1 (4.7, 5.5)	4.1 (3.8, 4.3)	3.9 (3.6, 4.2)	0.1 (−0.2, 0.4)	.54	0.0 (−0.4, 0.3)	.91
6-Month follow-up retention	n/a	87.3 (81.6, 93.0)	58.1 (47.5, 68.8)	75.1 (69.8, 80.4)	75.5 (71.6, 79.3)	n/a	n/a	n/a	n/a	n/a

Characteristics of the underlying patient population were approximated using data obtained from all patients with T1D seen at a single endocrinology department at least once during 2011–2012

^aStudy participants were defined as “retained” if they completed the follow-up visit scheduled for 6 months post-intervention

Recruitment Summary

For the Young Child Project (Fig. 1), 285 letters were mailed, 219 parents were successfully contacted by phone, and 203 patients were eligible. Eighty-two percent of parents (n = 167) who were contacted and eligible consented to participate; of those, 80 % (n = 134) completed baseline assessment and all were randomized. The majority of patients who were randomized were recruited from the primary study site; 27 % were recruited from the second Mid-Atlantic site and 4 % were recruited from the Midwest site. For the DECIDE project (Fig. 2), 260 letters were mailed, 209 parents were successfully contacted by phone, and 155 families were eligible. Seventy percent of parents (n = 108) who were contacted and eligible consented to participate; of those, 79 % (n = 86) completed baseline assessment and all were randomized. Finally, for the TeamWork Project (Fig. 3), 904 recruitment letters were mailed, 627 families were successfully contacted, and 395 were eligible. Due to inconclusive electronically-recorded medical diagnoses, many patients who were initially screened for eligibility were ineligible for the project due to either incorrect or comorbid medical diagnoses upon inspection of medical charts. Seventy-one percent of families (n = 281) who were contacted and eligible consented to participate in the project; of those, 91 % (n = 257) completed baseline assessment. Eighty-eight percent (n = 226) of these 257 participants were randomized. Sixty-seven percent of participants who were randomized were from the primary study site. Thus, in total, 79–91 % of all patients who were contacted and eligible for these RCTs completed baseline assessment and 79–88 % were randomized. Overall, demographic variables were comparable among the three sites; the exception was that 93 % of all Hispanic participants were recruited from the primary study site.

Fig. 1.

Young Child Project CONSORT Table

Fig. 2.

DECIDE Project CONSORT Table

Fig. 3.

TeamWork Project CONSORT Table

Participant Characteristics Related to Treatment Completion

Ninety-two percent of Young Child Project participants (n = 123) who completed baseline assessment completed all treatment phone sessions, 76 % of DECIDE Project participants (n = 32) who completed baseline assessment and were randomized to the intervention group completed both in-clinic sessions and the weekend session, and 70 % of TeamWork participants (n = 180) who completed baseline assessment completed all 4 treatment sessions. Overall, 70–92 % of participants completed treatment sessions.

Parents were less likely to complete all 5 Young Child Project program phone calls if their child was older, t(128) = 2.48, p < .05, d = .44, or not Caucasian, χ²(2, N = 131) = 13.09, p < .01. No demographic or medical variables were related to treatment retention in the DECIDE Project (ps > .05). See Tables 3 and 4.

Table 3.

t test results for treatment completion and 6-month follow-up completion

Treatment completion	Young Child Project (N = 134)					DECIDE Project (N = 42)					TeamWork Project (N = 226)
	Yes	No	t	df	p value	Yes	No	t	df	p value	Yes	No	t	df	p value
Child age (mean years)	5.23	6.15	2.48	128	.01	10.82	11.17	1.29	38	.21	12.75	13.08	1.89	255	.06
Parent age (mean years)	36.63	37.74	0.64	122	.52	Not available					Not available
Household income¥ (mean score)	77.39	77.33	−0.01	128	.99	4.33	3.63	−0.87	36	.39	48.05	42.50	−3.33	254	.001
T1D diagnosis duration (mean years)	1.96	2.07	0.31	127	.78	4.19	3.74	−.43	36	.67	2.86	3.57	0.66	254	.51
Baseline glycemic control (mean HbA1c)	8.07 %	8.51 %	1.81	120	.07	7.65 %	7.80 %	0.35	24	.73	8.70 %	9.14 %	1.94	255	.05
6-Month follow-up completion	Young Child Project (N = 134)					DECIDE Project (N = 86)†					TeamWork Project (N = 226)
	Yes	No	t	df	p value	Yes	No	t	df	p value	Yes	No	t	df	p value
Child age (mean years)	5.40	5.89	1.26	116	.21	10.71	11.00	1.79	81	.08	12.76	13.11	1.96	255	.05
Parent age (mean years)	37.11	37.29	0.10	111	.92	Not available					Not available
Household income¥ (mean score)	81.42	63.08	−2.12	126	.04	4.59	3.35	−2.54	76	.01	48.06	42.01	−3.54	245	.001
T1D diagnosis duration (mean years)	2.04	2.23	0.47	115	.64	4.22	4.15	−0.11	76	.91	5.05	5.31	0.58	254	.56
Baseline glycemic control (mean HbA1c)	8.00 %	8.60 %	2.69	124	.008	7.83 %	8.06 %	0.89	52	.38	8.48 %	9.81 %	6.03	255	.001

^¥Household income was assessed in different ways for each study, but for all studies a lower score indicated lower household income

^†N for the DECIDE Project treatment completion and 6-month follow-up completion are different because only intervention participants completed treatment sessions

Table 4.

Chi square results for treatment completion and 6-month follow-up completion

	Young Child Project (N = 134)					DECIDE Project (N = 42)					TeamWork Project (N = 226)
Treatment completion	Yes	No	χ 2	df	p value	Yes	No	χ 2	df	p value	Yes	No	χ 2	df	p value
Child gender
Female	59	7				15	6				85	45
Male	57	8	0.09	1,131	.79	17	3	1.10	1,41	.45	103	24	8.08	1,257	.01
Child race
Caucasian	84	7				23	9				133	46
Afro-American	17	8				Included in ‘Other’ category					Included in ‘Other’ category
Other	5	0	13.09	2,131	.001	8	1	1.10	1,41	.41	55	23	0.40	1,257	.54
Caregiver marital status
Married	92	9				26	6				142	51
Not married	16	4	2.14	1,121	.23	6	3	0.87	1,41	.38	43	17	.09	1,253	.87
Insulin regimen
Conventional	26	5				17	14				52	38
Intensive	83	6	2.43	1,120	.15	5	4	0.01	1,40	.99	136	30	17.45	1,256	.001
Treatment group
Intervention	76	9				Not applicable					112	41
Control	38	5	0.03	1,128	.99						76	20	1.13	1,249	.36
6-Month follow-up completion	Young Child Project (N = 134)					DECIDE Project (N = 86)†					TeamWork Project (N = 226)
	Yes	No	χ 2	df	p value	Yes	No	χ 2	df	p value	Yes	No	χ 2	df	p value
Child gender
Female	59	6				27	18				100	27
Male	53	11	1.78	1,129	.20	22	17	0.11	1,84	.83	93	37	1.78	1,257	.20
Child race
Caucasian	91	10				38	23				136	43
Afro-American	16	6				Included in ‘Other’ category					Included in ‘Other’ category
Other	5	0	5.73	2,128	.06	11	12	1.44	1,84	.32	57	21	0.24	1,257	.64
Caregiver marital status
Married	93	9				40	22				141	34
Not married	15	5	4.31	1,122	.05	8	13	4.49	1,83	.04	50	28	7.91	1,253	.01
Insulin regimen
Conventional	28	4				21	21				91	51
Intensive	80	9	0.14	1,121	.74	27	14	2.14	1,83	.18	100	12	21.32	1,254	.001
Treatment group
Intervention	68	10				20	20				119	34
Control	38	2	1.77	1,118	.22	29	14	2.61	1,83	.08	74	22	0.02	1,249	.99
Treatment completion
Complete	105	8				20	11				175	13
Incomplete	7	9	29.62	1,129	.001	0	9	11.61	1,40	.001	18	51	121.15	1,257	.001

^†N for the DECIDE Project treatment completion and 6-month follow-up completion are different because only intervention participants completed treatment sessions

Dyads from the TeamWork Project were less likely to complete all 4 program sessions if the youth was female, χ²(1, N = 257) = 8.08, p < .01, had poorer baseline glycemic control, t(255) = 1.94, p < .05, d = .24, was on a conventional insulin regimen, χ²(1, N = 256) = 17.45, p < .001, or had a lower household income, t(245) = −3.33, p < .05, d = −.43. There was also a trend that dyads who were less likely to complete all TeamWork project program sessions had youth who were older, t(255) = 1.89, p = .06, d = .24.

Participant Characteristics Related to 6-Month Retention

Ninety percent of Young Child Project participants (n = 120), 58 % of DECIDE Project participants (n = 49), and 70 % of TeamWork Project participants (n = 182) completed 6-month follow-up assessments. In the Young Child Project, parents were less likely to complete the 6-month follow-up assessment if their child had poorer baseline glycemic control, t(124) = 2.69, p < .01, d = .48, if they were not married, χ²(1, N = 122) = 4.31, p < .05, if they had a lower household income, t(126) = −2.12,p < .05, d = −.38, or if they had not completed all 5 program phone calls, χ²(1, N = 129) = 29.62, p < .001. Parents were somewhat less likely to complete the 6-month follow-up assessment if their child was not Caucasian, χ²(2, N = 128) = 5.73, p = .06. See Tables 3 and 4.

In the DECIDE Project, participants were less likely to complete the 6-month follow-up assessment if the parent was not married, χ²(1, N = 83) = 4.49, p < .05, had a lower household income, t(76) = −2.54, p < .05, d = −.58, or did not complete all program sessions, χ²(1, N = 40) = 11.61, p < .01. Participants were also somewhat less likely to complete the 6-month follow-up assessment if the youth was older, t(81) = 1.79, p = .08, d = .40, or had been randomized to the intervention group, χ²(1, N = 83) = 2.61, p = .08.

Finally, in the TeamWork Project, dyads were less likely to complete the 6-month follow-up assessment if the youth was older, t(255) = 1.96, p < .05, d = .25, had poorer baseline glycemic control, t(255) = 6.03, p < .001, d = .76, was on a conventional insulin regimen, χ²(1, N = 254) = 21.32, p < .001, had an unmarried caregiver, χ²(1, N = 253) = 7.91, p < .01, had a lower household income, t(245) = −3.53, p < .001, d = −.45, or if the dyad did not complete all program sessions, χ²(1, N = 257) = 121.15, p < .001.

Overall retention rates across studies were comparable between the primary and secondary study sites, 74 versus 77 %. The tertiary study site had a 100 % retention rate, but notably, only 5 participants were recruited from this site.

Combined Study Retention Analyses

Data from all three studies were combined and the relations among all common elements (i.e., child age, gender, and race; primary caregiver age, gender, and marital status; time since diagnosis, and baseline glycemic control) and 6-month retention were assessed using multivariable regression, adjusting for site and study. Patients with poorer baseline glycemic control (Adjusted OR 1.38 for each 1-unit increase; 95 % CI 1.16, 1.64; p < .001) were significantly more likely to drop out prior to the 6-month assessment, as were those whose primary caregiver was not married (Adjusted OR 1.96; 95 % CI 1.14, 3.38; p < .05). There was a suggestion that older patients were more likely to become lost-to-follow-up (Adjusted OR 1.22 for each 1-year increase in age; 95 % CI 0.99, 1.51, p = .07), while patients with older parents were less likely to become lost-to-follow-up (Adjusted OR 0.96 for each 1-year increase in age; 95 % CI 0.93, 1.00, p = .08). After accounting for these factors, other characteristics such as race and time since diagnosis (or transfer of care into our clinic) were not independently associated with study retention. See Table 5 for further information.

Table 5.

Multivariate logistic regression results identifying factors independently associated with failure to complete the 6-month assessment across all three studies

Characteristic	Adjusted OR (95 % CI)a	p value
Child age (years)	1.22 (0.99–1.51)	.07
Male child	0.85 (0.53–1.37)	.51
Non-Caucasian race	0.97 (0.56–1.68)	.91
Time since diagnosis or transfer into clinic (years)	1.04 (0.95–1.13)	.39
Caregiver age (years)	0.96 (0.93–1.00)	.08
Male caregiver	0.62 (0.25–1.48)	.28
Unmarried caregiver	1.96 (1.14–3.38)	.02
Baseline HbA1c	1.38 (1.16–1.64)	<.001

^aAll estimates are also adjusted for site and study

Discussion

Despite intensive time commitments for participants, treatment completion and 6 month retention rates for these three studies were high and were comparable to those observed in the majority of the 24 trials of pediatric and adult patients reviewed by Wakim, Rosa, Kothari, and Michel (2011). Regarding recruitment across studies, patients and their parents were more likely to be enrolled if patients were younger and Caucasian. Of families who were contacted and determined to be eligible, consent to enroll ranged from 70 to 82 % across studies. Lower recruitment rates were noted among older patients, which may be attributed to the tendency for older children to have co-morbid conditions and busier schedules as well as the tendency for parents to be the primary participants in studies of younger children. Lower recruitment rates were also noted among non-Caucasian patients when compared to the general clinic population.

Overall, our estimate that greater than two-thirds of reachable and eligible patients will consent to enroll are comparable to other behavioral RCTs conducted among pediatric populations (Flores et al., 2009; Grey, Jaser, Whittemore, Jeon, & Lindemann, 2011; Loding, Wold, Skavhaug, & Graue, 2009; Sullivan-Bolyai et al., 2010; Tercyak et al., 2006; Wysocki et al., 2006). Of the families who were consented and randomized, the percentage that completed baseline assessment was fairly comparable across RCTs, ranging from 76 to 80 %, although treatment completion rates (70–92 %) and 6-month retention rates (58–90 %) were more variable. These results are similar to the pediatric trial attrition rates presented in the review by Karlson and Rapoff (2009). Thus, it would seem reasonable, if not conservative, for researchers designing RCTs for pediatric populations to expect that approximately 30 % of patients who are contacted and eligible will not consent to participate, and of consented participants, approximately 20 % will not be retained for the randomization phase and an additional 20 % will not complete 6-month follow-up assessment.

The factors that were related to treatment completion and retention were largely consistent across RCTs; older child age, worse baseline glycemic control, lower household income, and an unmarried caregiver were all related to lower treatment completion and retention rates. However, there were a few exceptions, including non-Caucasian race, which was only related to treatment retention in the Young Child Project, and conventional insulin regimen, which was only related to treatment completion and retention in the TeamWork Project. These results suggest that specific subgroups of patients may be more susceptible to loss to follow-up. Study teams should consider creative strategies to alleviate the burden of study participation for all families, such as make up sessions, child care for siblings, or covering transportation costs for study assessments.

These studies suggest that there are effective strategies that may maximize recruitment and retention of a diverse sample of participants, many of which have previously been discussed in the literature (Boys et al., 2003; Coday et al., 2005; Germann et al., 2006; Gul & Ali, 2010; Karlson & Rapoff, 2009; Marcellus, 2004; Patel et al., 2003; Probstfield & Frye, 2011). In particular, strategies that permitted patients to participate in these studies at times that were convenient for them and minimized participants’ travel requirements were key methods of participant retention. The phone-based RCT conducted among parents of young children had the highest treatment completion and follow-up retention rates, which may be due to the focus on parents only instead of parent-youth dyads and the convenience of in-home treatment via phone-based assessment and intervention. Indeed the other studies required participants to attend sessions in person; although many times these sessions were paired with endocrinology follow-up appointments, this may have made participation challenging. Additional strategies that appear to have been successful for these studies were flexible, multi-method data collection, in person sessions that increased participant-researcher rapport, the maximization of benefits to the participants, maintenance of multiple phone/email contacts, and use of reminder calls/emails. Furthermore, the research teams established collaborative relationships with the medical care teams of each RCT, which increased support for the projects.

Additional recruitment/retention strategies that were not used but should be considered are mobile health strategies, such as online questionnaires that can be completed from an in-clinic laptop, home computer, or mobile phone, home- or web camera-based data collection and intervention sessions, the use of text messages as appointment reminders, and the integration of patient surveys with the electronic health record. The inclusion of these, and other, new technologies and strategies should be considered thoughtfully and their use should be evaluated systematically in order to determine their impact on participant retention and study outcomes. There is the possibility that participants will misunderstand assessment instruments if they have less contact with the research team. Identifying features of high risk families who are likely to drop-out of the study early may also allow for the use of targeted strategies that promote study completion. That being said, researchers who intend to use tailored intervention and data collection strategies are cautioned to recognize their impact on outcome data (e.g., differential exposure, researchers’ ability to clarify questionnaire items, etc.). Researchers are advised to track these strategies in order to assess their influence. Of course researchers must also carefully balance compensation for participants’ time with undue coercion.

This analysis was limited somewhat by the breadth of characteristics that were available in the clinic database for the assessment of factors associated with study enrollment. For example, underlying socioeconomic factors, such as household income and insurance type, were not available for our overall clinic population and/or were not consistently measured for each of our three studies, so the impact of socioeconomic status could not be assessed. It is possible that findings associated with race in our unadjusted analyses may reflect underlying socioeconomic differences. To whatever extent possible, study teams should record information during the screening process. A second limitation is that the data from these RCTs are from one research team’s work. Although they are multi-site studies with several primary investigators, the recruitment and retention strategies utilized are similar. Despite this, these data represent a broad age range, large samples, and analyses regarding both individual projects and combined projects that reveal robust findings.

Findings from these analyses provide preliminary guidelines regarding recruitment and retention rates among pediatric behavioral RCTs that may assist future researchers as they develop behavioral interventions for youth with chronic illnesses and their families. Researchers are encouraged to continue dissemination of study recruitment and retention data and procedures, follow the CONSORT guidelines, and collaborate with medical staff to facilitate support for RCTs and the inclusion of evidence-based behavioral interventions in routine clinical care. This information will facilitate the identification of the most robust methods to effectively utilize patients’ and clinicians’ time and funding for these projects, which will in turn benefit the families for whom these behavioral interventions are developed.