The use of electronic medical records for recruitment in clinical trials: findings from the Lifestyle Intervention for Treatment of Diabetes trial

The LIFT Diabetes randomized trial was designed to investigate the effects of two contrasting interventions (a 12-month community-based intensive lifestyle intervention versus a clinic-based enhanced diabetes self-management program) on cardiovascular disease (CVD) risk in overweight and obese adults with type 2 diabetes [11, 12].

The trial protocol was approved by the Institutional Review Board (IRB) of the Wake Forest School of Medicine, and a limited temporary waiver of the Health Insurance Portability and Accountability Act (HIPAA) [13] authorization was issued by the IRB, which permitted study investigators to access protected health information to confirm eligibility and facilitate initial contact for recruitment as allowed by government regulations. All participants in the study gave a verbal consent before the telephone screen and a written informed consent at their initial screening visit.

Participants eligible for recruitment into the LIFT Diabetes trial were aged 21 or older, had a confirmed diagnosis of type 2 diabetes (HbA1c ≥6.5 %, a physician diagnosis of diabetes on the participant’s problem list, or the use of diabetes medication), a body mass index (BMI) of 25 or greater (27 or greater if on insulin), and a regular source of medical care. Participants were excluded because of the following: history of prior cardiovascular disease or history of cancer with expected survival less than 2 years; glycosylated hemoglobin (HbA1c) >11 %, blood pressure >160/100 mmHg, triglycerides >600 mg/dl; history of prior weight loss surgery; unstable psychiatric disease; inability to walk two blocks without stopping; drug or alcohol abuse; use of drugs known to affect body weight (e.g., corticosteroids); pregnancy or breastfeeding; and advanced renal disease (estimated glomerular filtration rate, eGFR <45 ml/min/1.73 m²).

The LIFT Diabetes trial aimed to recruit a sample of 55–60 % African American, 10–15 % Hispanic, and 25–35 % non-Hispanic White (NHW) participants. The trial recruited 260 participants in Winston-Salem, North Carolina, USA and surrounding areas.

The primary recruitment method employed was the use of the EMR system and targeted mailing. Individuals with a diagnosis of diabetes mellitus on the problem list in their medical records were identified via an EMR system containing records for Wake Forest Baptist Health (WFBH) patients. These individuals were then initially screened electronically (e-screening) for major exclusions (history of CVD, cancer, prior weight surgery) cited on the problem list. During this initial e-screening phase, a less strict cut-off value was used for clinical and biological parameters such as BMI, blood pressure, HbA1c, triglycerides, and eGFR, since these measurements can change significantly over time. For example, for blood pressure, a systolic pressure of 170 mmHg and/or a diastolic pressure of 110 was used for e-screening, in place of the predefined cut-off of >160/100 mmHg. Individuals remaining potentially eligible were mailed a brochure with an opt-out postcard to return if not interested. If a postcard was not received after 2 weeks, a study staff member reached out to potential participants to determine interest and conduct a telephone screen. To assess the efficacy of the EMR as a screening tool, sensitivity and specificity analyses were done on a random subsample of participants with diabetes on the problem list.

Another recruitment method used was direct referrals from physicians or other healthcare providers, study participants, and study team members. To foster referrals from within the WFBH system, several study investigators made presentations at Internal Medicine and Family Medicine faculty meetings. For healthcare provider referrals, LIFT Diabetes contact cards were made available for interested individuals to provide their contact information.

Study team members attended six health screening events in the community (health fairs and church screenings), where interested participants were either screened on site or contacted at a later date by telephone. The study team also distributed flyers in targeted local pharmacies in the community and different WFBH medical practices. In order to increase the representation of African Americans in the trial, especially African American men, television advertisements were aired during specific programs and specific times of the day, and a study investigator completed a radio interview. LIFT Diabetes study advertisements were also printed and published in a local newspaper and magazine and in church bulletins, and posted online (on the Wake Forest Baptist Health website for clinical studies and an online press). All study recruitment materials were approved by the IRB of the Wake Forest School of Medicine and focused on three main criteria: adults aged 21 and older, a diagnosis of type 2 diabetes, and being overweight or obese.

Participants who expressed interest in the study were screened either via telephone or face to face using a scripted screening instrument. Five telephone screening calls were attempted for each participant who did not opt out, and a voicemail message requesting a return call was recorded whenever possible. To determine potential eligibility during the phone screen, major exclusions were assessed. A typical telephone screen lasted on average 15 min. For most participants, eligibility to progress to the next screening phase was established during the telephone interview. For a few, a second review by the study physician was required.

All participants deemed potentially eligible after the telephone screen were invited to attend a clinic where additional screening was performed to assess final eligibility. During the clinic visit, the following eligibility parameters were assessed: confirmed diagnosis of diabetes, blood pressure ≤160/100 mmHg, BMI ≥25 kg/m², urine (dipstick analysis) proteinuria <4+, HbA1c ≤11 %, eGFR ≥45 ml/min, triglycerides ≤600 mg/dl, absence of a history of CVD, ability to exercise (using the Physical Activity Readiness Questionnaire), and absence of severe depression (using the Patient Health Questionnaire - 9 instrument). Participants who remained eligible after the first clinic screening visit were invited to a baseline visit, during which they were randomized to one of the two arms of the trial. The duration of recruitment and randomization was dependent on the number of personnel, the duration of clinic visits, and the willingness of eligible individuals to participate. Of note, we aimed to recruit participants in waves, as we were limited to a maximum of six concurrent groups due to staffing and space constraints.

Personnel effort was estimated using full-time equivalents (FTEs) for all persons involved with recruitment. An FTE was estimated based on the percentage effort allocated to the activity of recruitment. For example, an FTE of 1.0 was equivalent to a 100 % effort, which translated to 40 h of work per week by the personnel. The number of persons directly involved with recruitment was not constant throughout the entire period, and as such personnel effort was estimated on a monthly basis. The actual amount of time (in hours) spent by personnel during recruitment was estimated. Estimates were done for the telephone screening encounter as well as both clinic visits. For the telephone screening encounters, the estimated time included the time to obtain a verbal consent, complete the screening questionnaire, and enter the data into the online database system. For both clinic visits (eligibility screen and baseline), the estimated time included the time to obtain a written consent, complete all relevant forms, and enter the data into the online database system.

Results are presented as mean and standard deviation or number (percentage) where appropriate. Differences in demographic and clinical variables between enrolled and non-enrolled participants were assessed using the Kruskal-Wallis test for continuous variables and the chi-square or Fisher’s exact test (where cell counts are low) for categorical variables. The estimate for the overall recruitment yield was calculated as the ratio of participants enrolled in the study to the total number of participants reached by telephone. The recruitment yield was also estimated per recruitment method and by race/ethnic group. (The cumulative number of participants randomized against the study goal is plotted in Fig. 2.) To test if exclusion criteria differentially affected enrollment of African Americans, logistic regression models were fitted to estimate odds ratios for African Americans compared to other race/ethnic groups (as the reference) for each of the major criteria. For this analysis, 41 participants were excluded because they had missing data on race. A P value of less than 0.05 was considered statistically significant. All analyses were performed using SAS version 9.3 (SAS Institute Inc., Cary, NC).

Recruitment efforts began in May 2013 and were concluded by the first week in March 2015. A total of 260 participants were randomized over a 21-month period (June 2013 to March 2015). The recruitment process was completed in four phases. In the first phase, a total of 5122 unique medical records (after duplicate records had been eliminated) were abstracted from the EMR database and assessed for major exclusions. After this initial assessment, 2767 EMR records were retained for a telephone screen. An additional 465 telephone screen-eligible participants came from direct referrals, media advertisements, community screening events, and other sources (Fig. 1). In the second phase, 1102 (64.8 % from EMR) participants contacted completed a telephone screen (a total of 6899 phone calls were made, including missed calls), of which 593 (53.8 %) qualified for a clinic screening visit (third phase of recruitment). Of the 593 potential participants, 289 were declared eligible for a baseline visit and eventual randomization. In the last phase of recruitment (randomization phase), 260 (61.5 % from EMR) participants were randomized (130 in each arm of the study). Figure 2 displays the cumulative number of participants randomized versus the study goal.

Among all individuals reached by telephone for an initial screen, 842 (76.4 %) were excluded from the study. At the level of the telephone screen, compared to enrolled participants, a higher percentage of non-Hispanic blacks (NHB) were excluded (57.3 versus 48.1 %, P = 0.01) (see Table 1). More non-Hispanic whites (NHW) were enrolled than excluded at this stage of screening (45.4 versus 34.1 %). At the level of the clinic screening visit, compared to those who enrolled, excluded individuals were significantly younger, and a lower percentage were on medications for diabetes, hypertension, and dyslipidemia (Table 1). There were no differences in sex, BMI, HbA1c, duration of diabetes, blood pressure, or smoking status between enrolled participants and those excluded.

Table 2 illustrates the reasons for exclusion of individuals during the LIFT Diabetes screening process. The major health reason for exclusion was a history of CVD or cardiovascular procedure (16.5 % of participants). A BMI <25 kg/m² (<27 kg/m² for insulin users) and an HbA1c > 11 % accounted for more than 8.9 and 5.2 % of exclusions, respectively. About 5.4 % of those screened reported being unable to exercise. In multivariable logistic regression analysis, compared to other race/ethnic groups, NHB were not more likely to be excluded from the study because of predefined criteria (Additional file 1: Table S1).

Overall, the recruitment yield was 23.6 %. The yield was higher for NHW (30.2 %), compared to NHB (21.4 %) and Hispanics (20 %). When recruitment methods were compared, the yield was highest for referrals (27.5 %). The yield for participants recruited through the EMR was 22.4 %, higher than that for community screening events (13.6 %) and media advertisements (20.5 %) (Fig. 3a).

Overall, 61.5 % of randomized participants were recruited through the EMR system, compared to 25.4 % recruited through direct referrals. About two-thirds of all participants recruited via referrals were NHB. A smaller percentage of randomized participants were recruited through media advertisements (5.8 %), health and church screening events (1.2 %), and other methods (6.1 %). Figure 3b displays the percent of randomized participants by recruitment method and by race/ethnicity. There was a significance difference in proportions of NHW, NHB, and other race/ethnic groups by the recruitment methods (P < 0.0001). About three-quarters (72.9 %) of randomized NHW were recruited through the EMR, compared to 49.6 % of randomized NHB (Fig. 3b). In comparison, 32.8 and 10.4 % of randomized NHB, versus 18.6 and 1.7 % of randomized NHW, were recruited through direct referrals and media advertisements, respectively.

Average monthly personnel effort throughout the recruitment period ranged from 0.613 to 0.825 FTE per week (24.5–33.0 h per week) per personnel. The average time spent to complete a telephone screen was 0.5 h (30 min). A total of 1102 telephone screens were completed, which amounts to an average of 551 h. Clinic visits lasted on average 2 h each. There were a total of 478 completed eligibility screening visits and 281 completed baseline visits. The total time spent for both clinic visits averaged 1518 h. Per recruited participant, the average time spent was 4.5 h.

For the 2130 uncompleted telephone screens (either due to a nonresponse to the telephone call or because the participant expressed no interest in participating), each encounter attempt lasted on average 5 min or less. These time estimates do not take into account the time spent for unanswered telephone call attempts, telephone calls to schedule clinic appointments, and time spent designing brochures/flyers and media advertisements.