Participation in pediatric epidemiologic research: the SEARCH for Diabetes in Youth Study experience

Abstract

Background:

We evaluated the association of demographic and clinical characteristics with participation in an epidemiologic study of diabetes mellitus among youth.

Methods:

SEARCH for Diabetes in Youth is a multi-center study of physician-diagnosed diabetes in youth under the age of 20 comprising a surveillance and a cohort component. At each center, we enumerated all prevalent cases of diabetes in 2001 (n=6,266) and all incident cases between 2002 and 2004 (n=3,668). After confirmation of eligibility and validation, we invited each case to complete a survey and participate in a study visit. Here we evaluate how age, sex, race, and diabetes type are associated with participation in the survey and study visit.

Results:

Among prevalent cases, participation in the survey was 68% and 41% in the study visit. Among 2002 to 2004 incident cases, participation varied for the survey (76%, 81%, and 82%) and study visit (52%, 60%, and 60%). In multivariate logistic regression analyses among all incident cases, older age was associated with a lower odds of participation in the study visit (15–17 vs. <10 years: OR 0.5, 95% CI 0.4–0.7; 18–19 vs. <10 years: OR 0.3, 95% CI 0.2–0.5), as was having type 2 diabetes vs. type 1 diabetes (OR 0.5, 95% CI 0.4–0.7) and being of African American race vs. non-Hispanic Whites (OR 0.6, 95%CI 0.4–0.8). Results were very similar among prevalent cases.

Conclusions:

Elucidating the relationship between individual characteristics and participation is essential for evaluating nonresponse bias, correcting for it, and for planning and implementing recruitment strategies.

INTRODUCTION

The decision to participate in research is based on numerous factors including participant characteristics, their physical and social environment, and attributes of the research project [1]. Participation rates tend to be lower in minority populations, among persons with lower socio-economic status, more cardiovascular risk factors, or compromised health status [2–7]. Research with minors presents additional challenges because both the parent or guardian and the youth need to agree to participate. Special consideration must be given to families of children with chronic diseases, as the child’s health condition creates an additional burden.

Because low participation rates can in some instances result in biased prevalence estimates and inaccurate measures of association [2, 8], the analysis of participation is an integral part of the evaluation of an epidemiologic study [9]. Information on participants and nonparticipants can furthermore be used to quantify the magnitude and direction of potential bias [3, 10–13] and to correct or compensate for nonparticipation bias [14, 15].

SEARCH for Diabetes in Youth is a large multicenter epidemiologic study of physician-diagnosed diabetes mellitus in children and youth under age 20 years [16]. By design, SEARCH ascertained demographic and clinical information on all eligible cases. Here, we evaluate the association of specific demographic and clinical characteristics with the level of study participation in youth with diabetes.

METHODS

The SEARCH study was designed to estimate the prevalence and incidence of diabetes in children and youth under age 20 years in the United States by age, sex, race/ethnicity, and type of diabetes and to investigate the evolution of diabetes type and its complications over time [16–18]. Details of the SEARCH study design have been published [16]. In short, the SEARCH surveillance component aims to enumerate and identify all eligible cases of diabetes, while the SEARCH cohort component aims to recruit these children and youth to a baseline visit and subsequent follow-up visits. Youth with diabetes were identified in geographically defined populations by the clinical centers located in Ohio, Washington, South Carolina, and Colorado, and in managed care plans in Hawaii and California, and coordinated by the Colorado center from beneficiary rolls in the Indian Health Service (IHS) in several reservation-based American Indian populations.

Each center developed a surveillance system for ascertaining cases based on networks of pediatric and adult endocrinologists, existing pediatric diabetes databases, hospitals, the databases of health plans, and other health care providers. Membership-based centers additionally performed queries of their administrative databases to identify cases, including linkage of computer data on prescriptions, hospitalizations with diabetes as the discharge diagnosis, and laboratory measures of hemoglobin A1c. The institutional review board (IRB) at each center approved the study protocol which complied with the privacy rules of the Health Insurance Portability and Accountability Act (HIPAA).

Eligibility criteria and validation of prevalent and incident cases

SEARCH ascertained prevalent cases in 2001 and incident cases from 2002 onward. This report focuses on the years 2001 through 2004. Eligibility criteria included a) age less than 20 years of age on December 31, 2001 or on that same date in the respective incidence year and b) being a resident or member of the respective population under surveillance in that year. Active-duty military, institutionalized persons, and women with gestational diabetes only were not eligible. Youth with diabetes secondary to other conditions were not eligible for the study visit and excluded from this analysis. In Colorado, Hawaii, and South Carolina, the ascertainment area for the prevalent year constituted a subset of counties under surveillance for the incidence years [16].

Case reports were validated through physician reports, medical record reviews, or in a few instances, self-report of a physician’s diagnosis of diabetes [16]. Case reports were registered anonymously with the Coordinating Center at Wake Forest University in North Carolina using HIPAA compliant procedures.

Participant recruitment, initial survey, and study visit

Study staff recruited cases and/or their parents/legal guardians often with support from health care providers. Cases were invited to complete a survey and to attend a baseline study visit. They were also informed about future 12, 24, and 60 month follow-up visits. After verifying eligibility, the initial patient survey was mailed to potential participants for self-administration. For participants under the age of 18, parents/legal guardians served as proxies for the initial survey. The brief survey (19 questions) queried the case’s age at diagnosis, history of diabetes treatment, race/ethnicity, and information on residence to confirm eligibility. California used both English and Spanish versions, while Washington mailed the Spanish version to patients known to be Spanish speaking. The following items were included: an introductory letter, an invitation to the study visit, an incentive, a pre-paid return envelope, and a postage-paid post card on which cases or their proxies could indicate that they wanted additional information or that they did not want further contact. If the survey was not returned or if potential participants did not indicate that they did not want further contact, study personnel (a professional survey research company in one center) made follow-up telephone calls to complete the survey over the telephone. Occasionally, the survey was completed at the time of the study visit.

The baseline study visit consisted of a physical exam of the case (anthropometry, blood pressure, skin examination for acanthosis nigricans), collections of the case’s blood and urine after an overnight fast and questionnaires. For cases younger than 18, the parent/guardian completed the medical history and family history. Both the parent/guardian and case completed questionnaires on quality of life. In addition, for cases ages 10 and over, the case completed questionnaires on depression and health behaviors. Cases older than 18 completed all questionnaires themselves. Parents of participants under age 18 at the time of the visit provided written informed consent and the participating cases signed assent forms; all participants aged 18 or older signed informed consent documentation. All study visit participation data presented herein refer to the baseline study visit.

Definitions of levels of participation

Henceforth in the interest of simplicity, we will use the term case to include the parents/guardians of minors in the context of defining or describing participation levels, because participation by minors is a family activity. Participation levels were defined as follows: “Nonparticipants” were cases who could not be contacted by mail or phone, those who did not complete a survey, and those who explicitly declined the survey and hence the study visit. “Survey-only participants” included cases who participated in the survey but could not be contacted to schedule the study visit or declined to participate in the study visit. “Study visit participants” were cases for whom a blood draw or anthropometric measurements were completed, irrespective of the setting of the visit. Because the study protocol stipulated that study visit participants had to have completed the initial patient survey, the sum of nonparticipants, survey-only participants, and study visit participants equals the total number of registered cases. An additional category, “survey participants” included all cases who completed the survey regardless of whether they attended a study visit.

Definition of demographic and clinical characteristics

Age groups were defined as age at the end of the respective prevalence (2001) or incidence (2002–2004) year because our aim was to compare characteristics of participants and non-participants consistently. For prevalent cases, duration of diabetes (in years) was defined as the time interval between date of diagnosis and the end of 2001. For incident cases, time to registration was defined as the months between the date of diagnosis and the date of registration by SEARCH. The specific date of diagnosis was missing on 429 of 6,266 (6.8 %) of prevalent cases and on 3 of 3,667 (0.08 %) of incident cases.

Diabetes type, as assigned by the health care provider, was categorized as follows: type 1 (combining 1, 1a, and 1b); type 2; and other type (including hybrid type, maturity onset of diabetes in youth [MODY], type designated as “other”, type unknown by the reporting source, and missing).

Race/ethnicity was assigned by first identifying all cases of Hispanic ethnicity regardless of race. Subsequently, cases who were not Hispanic and belonged to only one of the four individual race groups –African-American, Asian or Pacific Islander, American Indian, and non-Hispanic white – were classified. Finally, for the roughly 3% of our population that was non-Hispanic and multi-racial, the plurality method was applied, which assigns each case to a single race category based on a defined algorithm [19, 20]. A small number of multi-racial cases who could not be classified using the plurality approach (n = 123) and cases who had missing information on race (n = 382, 3.8%) were assigned to a combined other/unknown group.

Basic demographic and clinical information was available for virtually all cases. A hierarchical approach was used to classify participant characteristics when they were available from more than one data source. For the race/ethnicity, sex, age, and date of diagnosis, self-reported information was considered the highest level, followed by data from medical records and then other administrative data sources. We subsequently repeated all analyses using only information based on the medical record for race/ethnicity; our conclusions were unchanged.

Statistical methods

The percent participation was calculated based on the total number of eligible, registered cases in the denominator (after excluding secondary cases) following the definition by Slattery et al. [9] (vs. for instance, contact or cooperation rates). After initial bivariate analyses, multivariate logistic regression analyses were conducted to evaluate the independent contribution of participants’ demographic and clinical characteristics to the levels of participation in these cross-sectional data. Models were constructed separately for prevalent (2001) and incident (2002–2004) cases. The incident years were combined because of very similar year-specific associations and an index year variable (2002, 2003, 2004) was included. Levels of participation were compared with the same nonparticipant group in separate models. We controlled for center given the unique infrastructures and geographic distributions. Interactions between the race/ethnicity of the cases and the study center were also evaluated because a high proportion of Hispanic, Asian or Pacific Islander, and American Indian youth came from a limited number of centers. Models were additionally stratified by 10 year age group, because the overwhelming majority of type 2 diabetes occurs in youth older than 10. Adjusted odds ratios (OR) and 95% confidence intervals (CI) are presented for each of the models. Analyses were conducted using SAS version 8.02.

RESULTS

Between 2001 and 2004, SEARCH ascertained 10,158 cases of youth with diabetes. After excluding 225 cases with secondary diabetes, 6,266 prevalent cases and 3,667 incident cases were included in these analyses (table 1). Participation in the survey and the study visit was lowest for prevalent cases (68%for the survey; 41% for the visit) and improved markedly over time. Of the 2004 incident cases, 82% participated in the survey and 60% in the study visit.

TABLE 1.

Level of participation of youth with diabetes mellitus, by year of study. The SEARCH for Diabetes in Youth Study, 2001 – 2004

	Total	Levels of participation†
	Registered*	Survey		Survey only		Study visit		Nonparticipants‡
Year	n	n	%	n	%	n	%	n	%
Prevalent
2001	6,266	4,309	68	1,775	28	2,534	41	1,957	31
Incident
2002	1,297	989	76	309	24	680	52	308	24
2003	1,213	986	81	253	21	733	60	227	19
2004	1,157	947	82	250	22	697	60	210	18

^*Individuals with secondary diabetes were excluded.

^†Survey = completed survey, irrespective of visit attendance; Survey only = completed survey but unable to contact for visit or declined to participate; Study visit = completed anthropometric measurements or blood draw.

^‡Nonparticipants = unable to contact, did not complete a survey, or explicitly declined survey

The level of participation for the prevalent 2001 and incident 2002–2004 cases is described in table 2 by demographic and clinical characteristics. Both age categories of children (< 10, 10–14 years) were more likely to participate in the survey and the study visit than adolescents (15–17 years) and young adults (18–19 years), and this was true for all years of data collection. Additional age-stratification among children younger than 10 years did not reveal any further differences in participation (data not shown). In 2002–2004, survey participation ranged from a high of 85% among those younger than age 10 years to 56% among those aged 18–19 years. Study visit participation ranged from 65% for the youngest group to 34% for the oldest. Males and females were equally likely to participate. Non-Hispanic White and Hispanic youth demonstrated the highest participation levels and African American youth the lowest. The proportion of cases of youth with type 1 diabetes who completed the survey or study visit was markedly higher than the corresponding proportion among those with type 2 diabetes or those with some other diabetes type. Of the 159 youth with other types of diabetes, 15 had been diagnosed with MODY, 14 as hybrid diabetes, 24 as “other” type or as “not yet specified,” and the vast majority, 106 were identified as “unknown” type or the type information was missing.

TABLE 2.

Participation among youth with diabetes by demographic and clinical characteristics, prevalent (2001) and incident ascertainment (2002 – 2004)

		Total		Levels of participation†
		registered*		Survey		Survey only		Study Visit		Nonparticipants‡
		2001	2001–2004	2001	2001–2004	2001	2001–2004	2001	2001–2004	2001	2001–2004
		n	n	%	%	%	%	%	%	%	%
Age Group	< 10	1,328	1,365	78	85	23	20	55	65	22	15
	10–14	2,090	1,357	75	83	29	23	47	60	25	17
	15–17	1,670	689	67	71	33	25	34	46	33	29
	18–19	1,178	256	50	56	27	22	23	34	50	44
Sex	Male	3,114	1,796	68	81	30	22	39	58	32	19
	Female	3,152	1,871	69	78	27	21	42	57	31	22
Race/	NonHispanicWhite	4,017	2,152	73	85	30	22	43	63	27	15
ethnicity	Hispanic	773	469	77	86	34	29	43	57	23	14
African American		708	576	64	76	28	24	36	52	36	24
	American Indian	162	87	59	63	9	7	50	56	41	37
Asian/Pacific Islander		278	206	74	80	33	25	41	55	26	20
	Other/Unknown	328	177	6	14	3	6	3	8	94	86
Diabetes type	1§	5,383	2839	72	84	29	22	42	62	28	16
	2	760	717	56	68	24	23	32	45	44	33
	Other#	81	78	41	58	25	30	16	28	58	42
Center	1	1,162	465	76	89	41	36	36	54	24	11
	2	1,618	1,014	69	82	23	21	46	61	31	18
	3	207	142	69	81	25	17	44	64	21	19
	4	1,204	513	77	91	32	29	44	61	23	9
	5	390	797	69	80	24	19	45	61	31	20
	6	1,685	736	58	63	23	15	34	48	43	38

^*Individuals with secondary diabetes were excluded.

^†Survey = completed survey, irrespective of visit attendance; Survey only = completed survey but unable to contact for visit or declined to participate; Study visit = completed anthropometric measurements or blood draw.

^‡Nonparticipants = unable to contact, did not complete a survey, or explicitly declined survey.

^§Includes Type 1A, T1, and T1B;

^#Includes MODY, hybrid and unknown type.

The relationships between demographic and clinical characteristics and levels of participation among prevalent cases, as derived from multivariate logistic regression models, are shown in table 3. The results are shown first for the entire population (“all youth”), and also stratified by 10-year age group. Age was strongly and independently associated with the odds of study participation; cases aged 15–17 had about half the odds of participating in the survey and study visit as children younger than 10 years. The 18–19- year-olds had even lower odds for participating. Age stratification revealed that the influence of age on participation was limited to cases older than 10 years. Among prevalent cases, females were slightly more likely to participate in the study visit than males.

TABLE 3.

Association of participation level with demographic and clinical characteristics for prevalent cases (2001) from multivariate models

		Survey vs. nonparticipants						Study visit vs. nonparticipants
		All youth		< 10 years		≥ 10 years		All youth		< 10 years		≥ 10 years
		OR*	95% CI	OR*	95% CI	OR*	95% CI	OR*	95% CI	OR*	95% CI	OR*	95% CI
Age group (years)	< 10	1.0	-	-	-	-	-	1.0	-	-	-	-	-
	10–14	0.9	0.7, 1.1	-	-	-	-	0.8	0.7, 1.0	-	-	-	-
	15–17	0.6	0.5, 0.7	-	-	-	-	0.5	0.4, 0.6	-	-	-	-
	18–19	0.3	0.26, 0.4	-	-	-	-	0.2	0.18, 0.3	-	-	-	-
	Continuous	-	-	1.0	0.95, 1.1	0.86	0.84, 0.89	-	-	1.0	0.96, 1.1	0.83	0.81, 0.86
Sex	Male	1.0	-	1.0	-	1.0	-	1.0	-	1.0	-	1.0	-
	Female	1.1	1.0, 1.3	0.9	0.6, 1.2	1.2	1.0, 1.4	1.2	1.1, 1.4	0.9	0.6, 1.2	1.3	1.1, 1.5
Race/ethnicity†	NHW	1.0	-	1.0	-	1.0	-	1.0	-	1.0	-	1.0	-
	Hisp	1.1	0.8, 1.4	0.9	0.5, 1.6	1.2	0.9, 1.5	1.2	0.9, 1.6	1.0	0.5, 1.8	1.4	1.0, 1.9
	AA	0.7	0.6, 0.9	0.6	0.3, 1.0	0.7	0.5, 0.9	0.7	0.6, 1.0	0.6	0.4, 1.0	0.8	0.6, 1.0
	AI	0.7	0.5, 1.1	0.2	0.1, 0.7	0.8	0.5, 1.2	1.1	0.7, 1.7	0.2	0.04, 0.8	1.2	0.8, 1.9
	ASPI	1.1	0.7, 1.6	1.8	0.6, 5.1	1.1	0.7, 1.7	1.0	0.6, 1.5	2.1	0.7, 6.1	0.8	0.5, 1.4
Diabetes type	1‡	1.0	-	1.0	-	1.0	-	1.0	-	1.0	-	1.0	-
	2	0.6	0.5, 0.7	0.2	0.03, 1.3	0.6	0.5, 0.7	0.5	0.4, 0.7	0.2	0.01, 1.8	0.5	0.4, 0.7
	Other§	0.4	0.3, 0.8	0.5	0.1, 2.2	0.5	0.2, 0.8	0.2	0.1, 0.5	0.7	0.1, 3.0	0.2	0.1, 0.5
Duration of diabetes (years)		0.99	0.97, 1.00	0.85	0.78, 0.92	0.96	0.94, 0.98	0.96	0.94, 0.99	0.82	0.76, 0.89	0.94	0.92, 0.96

^*OR, odds ratio. CI, confidence interval. Adjusted for all characteristics shown and study center.

^†NHW, non-Hispanic White; Hisp, Hispanic; AA, African American; AI, American Indian; ASPI, Asian/Pacific Islander. Effect estimates for other/unknown race/ethnicity category not shown.

^‡Includes Type 1A, T1, and T1B.

^§Includes MODY, hybrid and unknown type.

African American cases consistently had a lower odds of completing the survey than non-Hispanic White cases. We observed similarly decreased odds for study visit participation of African American cases, but these findings did not reach statistical significance except in the combined “all youth” group. American Indian cases under the age of 10 were the only other race/ethnic group among the prevalent cohort to have a lower odds of participation in survey or study visit. We tested for interactions between race/ethnicity and center but found no important effects.

Independent of age, sex and race/ethnicity, having type 2 diabetes or being classified as other or unknown type of diabetes was associated with a lower odds of participation compared to cases of youth with type 1 diabetes. This finding in the total study population was largely driven by the strong inverse association of diabetes type with participation among cases age 10 or older. We furthermore observed that a longer duration of diabetes was strongly and inversely associated with participation.

We observed very similar associations between demographic and clinical characteristics and participation levels among incident cases (table 4) as for prevalent cases. Incident models were additionally adjusted for year of diagnosis to diminish potential cohort effects. Increasing age, type 2 diabetes, and other diabetes type were again independently and strongly associated with decreased odds of participation in survey and visit, this finding being largely driven by cases 10 years or older. There was little difference in participation in the incidence years by sex. Time from diagnosis to study registration was strongly and inversely associated with participation in the survey and study visit. Among incident cases, African American youth had generally lower odds of participation than non-Hispanic white youth, and this finding was strongest in the cases younger than 10 years. Additionally, we observed lower odds of participation in both survey and study visit by Asian or Pacific Islander cases younger than 10 years.

TABLE 4.

Association of participation level with demographic and clinical characteristics for incident cases (2002–2004) from multivariate models

		Survey vs. nonparticipants						Study visit vs. nonparticipants
		All Youth		< 10 years		≥ 10 years		All Youth		< 10 years		≥ 10 years
		OR*	95% CI	OR*	95% CI	OR*	95% CI	OR*	95% CI	OR*	95% CI	OR*	95% CI
Age group (years)	< 10	1.0	-	-	-	-	-	1.0	-	-	-	-	-
	10–14	0.9	0.7, 1.2	-	-	-	-	1.0	0.8, 1.3	-	-	-	-
	15–17	0.6	0.4, 0.7	-	-	-	-	0.5	0.4, 0.7	-	-	-	-
	18–19	0.4	0.3, 0.6	-	-	-	-	0.3	0.2, 0.5	-	-	-	-
	Continuous	-	-	0.99	0.92, 1.06	0.88	0.84, 0.92	-	-	1.01	0.93, 1.09	0.85	0.81, 0.89
Sex	Male	1.0	-	1.0	-	1.0	-	1.0	-	1.0	-	1.0	-
	Female	0.9	0.7, 1.1	0.7	0.5, 1.0	0.9	0.7, 1.2	0.9	0.7, 1.1	0.7	0.5, 1.1	1.0	0.8, 1.3
Race/ethnicity†	NHW	1.0	-	1.0	-	1.0	-	1.0	-	1.0	-	1.0	-
	Hisp	1.0	0.7, 1.4	0.9	0.5, 1.7	1.0	0.7, 1.5	1.0	0.7, 1.4	0.8	0.4, 1.6	1.0	0.7, 1.6
	AA	0.7	0.5, 0.9	0.4	0.2, 0.7	0.8	0.6, 1.2	0.6	0.4, 0.8	0.4	0.2, 0.8	0.7	0.5, 1.1
	AI	0.9	0.5, 1.6	1.0	0.1, 9.2	0.9	0.5, 1.6	1.5	0.5, 2.8	1.0	0.1, 11.4	1.5	0.8, 2.7
	ASPI	0.9	0.5, 1.5	0.4	0.1, 0.9	1.4	0.7, 2.7	0.8	0.4, 1.4	0.3	0.1, 0.9	1.2	0.6, 2.6
Diabetes type	1‡	1.0	-	1.0	-	1.0	-	1.0	-	1.0	-	1.0	-
	2	0.6	0.4, 0.8	1.5	0.4, 6.1	0.6	0.4, 0.7	0.5	0.4, 0.7	0.4	0.1, 2.0	0.5	0.4, 0.7
	Other§	0.5	0.3, 0.8	0.3	0.1, 1.3	0.5	0.2, 0.9	0.3	0.1, 0.6	0.3	0.1, 1.2	0.3	0.1, 0.6
Time to Registration (months)		0.90	0.89, 0.91	0.88	0.86, 0.91	0.91	0.89, 0.92	0.88	0.86, 0.89	0.87	0.84, 0.90	0.89	0.87, 0.90

^*OR, odds ratio. CI, confidence interval. Adjusted for all characteristics shown, study center, and diagnosis year.

^†NHW, non-Hispanic White; Hisp, Hispanic; AA, African American; AI, American Indian; ASPI, Asian/Pacific Islander. Effect estimates for other/unknown race/ethnicity category not shown.

^‡Includes Type 1A, T1, and T1B.

^§Includes MODY, hybrid and unknown type.

DISCUSSION

Evaluations of factors affecting participation in pediatric research are much more limited for epidemiologic studies [5, 7, 21, 22] than for clinical or intervention studies [23–26]. What has emerged to date is that female children, single children, and those with well educated parents are more likely to participate, while non-white children, those living in inner-city areas, and those with less educated parents are less likely to participate [5, 21, 22].

A small body of research has specifically explored the mechanisms by which parents and their children decide to consent to health research and what types of research were acceptable. It has been found that between 60% and 75% parents and their adolescent children agree on the participation decision [27, 28]. The level of risk and the amount of aversion clearly factor into the decision for both parents and adolescents, however, adolescents seem to be more willing to engage in above minimal risk research than parents [28]. While SEARCH was considered a minimal-risk protocol, it is likely that both children and parents felt some degree of aversion to venipuncture and urine sampling. It has furthermore been shown that there are a multitude of parental factors that influence the parents’ decision to participate including a balance of risks and benefits, their knowledge, beliefs and emotional response [29]. In addition, child factors such as the health status of the child and the children’s choice, research related factors, and factors related to the physician play a role [29]. Benefit of participation is a common reason for participation as is altruism [28] [25]. In the SEARCH study, the only research benefit for our participants was that the results of the blood analysis including diabetes autoantibody assays, lipids, c-peptide, and hemoglobin A1C were shared with the treating physician if they so chose.

Our study of youth with diabetes adds to the literature in a number of ways. We found that increasing age of the case was associated with declining rates of participation in both the survey and the study visit, a finding that has – to the best of our knowledge – not been reported before. For instance, participation in the survey was high through age 14 for incident cases, 83–85%. At 71%, response to the survey was still good for 15–17-year-olds, but it declined to 56% among the 18–19-year-old incident cases. While the parents/guardians of very young children clearly make decisions autonomously, with increasing age of a child the decision to participate becomes increasingly shared. A limitation of our study is the fact that we did not have demographic data on the parents/guardians. The National Survey of Children’s Health has recently reported close to 69% completed interviews among households with children under the age of 18 [30], but did not report participation levels by age of the child. Perhaps the extremely busy school and social schedules of today’s adolescents as well as their parent or guardian make the scheduling of early-morning fasting study visits particularly difficult. For young adults, inability to contact was anecdotally the most common reason for their nonparticipation.

Our study furthermore demonstrates that youth with type 2 diabetes or with some other or unknown (non type 1) type of diabetes have significantly lower odds of participating in the survey and study visits than youth with type 1 diabetes. This finding was independent of age, race/ethnicity, and sex. We stratified our analyses by age group to better understand the association between diabetes type and participation and found type 2 diabetes and other diabetes types to still be independently associated with lower odds of participation among those age 10 and older. One potential explanation for this finding may be that disease acceptance and perception of severity differ between persons with type 1 and type 2 diabetes (personal communication, Stephen Daniels). Persons with type 1 diabetes and their families typically have an acute sense of having a severe and life-threatening condition, which often motivates them to participate in research.

Another reason for the type-specific differences in participation may be a differential ability to contact, which may be attributable to the ascertainment source and the quality of contact information. We suspect that a larger proportion of youth with type 2 diabetes (versus type 1) were only seen in emergency rooms or in other outpatient settings and were not admitted to the hospital. The absence of a continuous relationship with a primary care provider or diabetes specialist may have resulted in SEARCH centers acquiring more limited contact information or even no contact information at all for these youth in geographic surveillance areas.

Independent of the aforementioned influences by age and diabetes type, we observed some differences in participation by race/ethnicity of the case. Any interpretation of race/ethnic differences needs to consider that we could not adjust for potential differences in socioeconomic status. The most consistent group to demonstrate lower participation was the African American group compared to non-Hispanic Whites, among both prevalent and incident cases. In age stratified analyses of incident cases, we noticed particularly low odds of survey and visit participation among younger (< 10 years) African American cases compared to White cases. Also restricted to younger children, we found prevalent cases among American Indian children and incident cases among Asian or Pacific Islander children to have lower odds of participation than white children. The small number of individuals with other/unknown race group was also less likely to participate (data not shown), a finding most likely related to the incomplete nature of some of the records. Of note, participation by Hispanic children and youth was at essentially the same level as that of non-Hispanic Whites. Because the six study centers had vastly different racial/ethnic compositions, we explicitly evaluated potential interactions between the center and race/ethnicity but found none.

Our ascertainment and recruitment experience offers several practical lessons for epidemiologic research. Locating and contacting eligible cases or their parents/guardians was a major hurdle. Where permitted by local IRBs or privacy boards, multiple addresses and contact numbers were abstracted. Introductory letters were mailed out with a request for forwarding addresses from the post office so that letters could be resent. In most centers follow-up calls were placed. Disconnected telephone numbers presented a significant challenge. Another problem encountered was that the case’s last name often was not the same as the parent’s or guardian’s last name.

A second major area of concern was eliciting participation after eligible cases or their parents/guardians were contacted. In designing the patient survey, we considered factors such as monetary incentives, length, and appearance, all of which have been shown to affect response rates [1]. For instance a $2.00 bill was included with the mailed survey study-wide. Furthermore, participants received a “thank you” for their time and effort at the study visit, a $40 value. While the value of this incentive was uniform study-wide, there was some variation in the mode of handling, though gift cards were generally the most common procedure. We noted that timely identification after diagnosis seemed to be an independent predictor of participation among incident cases. Each clinical center developed a tailored approach to recruitment, depending in part on the type of relationship it had with potential participants. Increasing the frequency of contact between the center staff and pediatric endocrinologists, who were a key reporting source, would be one strategy to improve timeliness. While our estimate of active refusals remained fairly constant at 5%, one of our limitations is that we did not assess reasons for refusal. Furthermore, we initially did not have an explicit protocol to uniformly define the various types of nonparticipation, which have been employed to develop alternate forms of response rates [9]. We therefore present only the most conservative types of participation estimates, using all eligible cases as the denominator.

The Health Insurance Portability and Accountability Act of 1996 (HIPAA) had a major impact on our ability to contact and recruit participants, especially given the multicenter nature of our study. HIPAA became effective in April 2003. We noted marked differences in the interpretation and implementation of HIPAA across our study centers, as have others [31, 32]. Of the four participating university departments, two were considered covered entities and two were not. Because all of the centers worked directly with health care providers and other covered entities, the HIPAA legislation had implications for all the centers. All six applied for various waivers of consent to obtain and use personal health identifiers, but not all were successful. The variation in nonparticipation across centers – ranging from 9 to 38% among incident cases – was mostly determined by varying levels of inability to contact potential participants, which in turn was influenced by HIPAA. The center with the highest nonparticipation rate was not allowed to directly contact roughly half of the cases within its ascertainment area and had to rely exclusively on individual health care providers for recruitment. The center with the highest response rate could contact all of its registered cases directly. Because of these differences in recruiting opportunities between centers, we additionally conducted our analyses stratified by center (data not shown). Age and type of diabetes were both significantly and strongly associated with participation in the study visit and survey within each of the six centers. The magnitude and direction of these associations were extremely similar across centers, and to the pooled results. African American race was related to participation in three centers (South Carolina, Colorado, California). While we clearly can not rule out the possibility of residual confounding by center, these findings increase our confidence in our results.

An integral part of the evaluation of an epidemiologic study is the description and analysis of participation and associated characteristics. Austin et al. [8] noted that while a low response cannot be equated with bias, studies with lower response rates tend to have more potential for bias. Bias is said to exist when the study estimates obtained from the respondents differ appreciably from estimates obtained from the full sample. A quantitative evaluation of the potential for bias requires data on both participants and nonparticipants. The SEARCH study was designed to collect basic demographic and clinical information on all eligible cases, thus allowing careful characterization of participants and nonparticipants. The data presented in the present paper can be used to evaluate the magnitude of potential bias – and as needed – correct for bias in future SEARCH publications. Statistical tools available include imputation methods [33] and a variety of methods based on weighting adjustment [14], including propensity scores [15]. These methods can be applied to analyses of SEARCH data which utilized survey respondents or clinic participants exclusively. For example, a recent report on the prevalence of elevated albumin excretion [34] found the crude, complete case-based estimate to be extremely similar to a reweighted estimate using a semi-parametric efficient estimator [14]. Finally, it is important to point out that the majority of primary study aims in SEARCH including the estimates of prevalence [17], incidence [18], or trends in incidence of diabetes over time, are not affected by response rates as they are based on all registered cases, for whom basic demographic and clinical information is virtually complete.

In conclusion, in an era of declining response rates [35–37], participation in SEARCH increased markedly over four years. While there is always room for improvement, our response rates are quite high in the current environment; 82% for the survey and 60% for the study visit in the 2004 cohort. In this study of cases of youth with diabetes from diverse racial and ethnic backgrounds, broad geographic areas, and ages ranging from birth to young adulthood, a case’s age proved to explain a large component of participation levels. In addition to older age, type 2 diabetes and other (non-type 1) types of diabetes were associated independently with decreased participation. We also observed lower participation among African American cases, and younger children of American Indian and Asian Pacific Islander race, but not among Hispanic children or youth. Again, it needs to be recognized that for minors, our study essentially assessed participation of the family. Our detailed analysis of response rates allows SEARCH investigators to compensate or correct for potential biases in future analyses limited to SEARCH participants, and illustrates the type of data needed to address biases in other similar studies. Our findings underscore the importance of collecting some data within the guidelines of HIPAA and local IRBs on nonparticipants, careful monitoring of participation rates, and evaluation of factors impacting participation.