Purpose
Noncompliance during prospective studies can bias results and limit conclusions. The current study retrospectively investigated the relationship between study subject characteristics and rates of noncompliance in interventional trials involving common causes of blindness.
Design
Retrospective analysis of 10 randomized clinical trials.
Methods
Subjects were enrolled in investigator-initiated trials studying proliferative diabetic retinopathy, neovascular age-related macular degeneration, diabetic macular edema, and retinal venous occlusive disease. Records were reviewed for hypothesized risk factors of noncompliance and rates of noncompliance, which were defined as at least 1 missed visit or exiting the study early. Demographic information, systemic medical history, and ocular medical history, including visual acuity and central retinal thicknesses, were examined retrospectively using Student t test, Pearson χ 2 test, and logistic regression.
Results
Of 390 subjects included, 212 (54.4%) were compliant with all scheduled study visits and 178 (45.6%) met criteria for noncompliance, with 53 (13.6%) subjects exiting early. Regression models identified 17 variables that were significant in determining subject noncompliance. Among those, distance, comorbidities, diabetic status, concomitant medications, previous clinic visits, length of study, disease under study, and severe adverse events were highly significant risk factors of noncompliance.
Conclusion
The current research identified a substantial proportion of subjects who met the criteria for noncompliance within the trials analyzed. The factors identified in the current work are consistent with published clinical observations and the results of previous clinical trials. These results highlight the importance of considering study design and medical history when designing prospective clinical trials in an attempt to minimize data loss.
Highlights
- •
Analyzed noncompliance across 10 prospective trials involving retinal diseases.
- •
Among 360 subjects, 45.6% met noncompliance criteria and 13.6% exited studies early.
- •
Using regression models, several risk factors for noncompliance were identified.
Well-designed and well-executed randomized clinical trials provide invaluable data for the advancement of medical research. The U.S. Federal Food and Drug Administration (FDA) relies on the results of large-scale clinical trials to determine the safety and efficacy of pharmaceuticals and medical devices. With each pivotal trial submitted to the FDA, regardless of disease area, costing a mean of approximately $19 million, it is a meaningful loss to the pharmaceutical industry and progress in treatment development that only a median of 9.6% of all therapeutics entered into phase I trials between 2006 and 2015 gained approval. ,
There are many reasons why a clinical trial program may fail, including faults in study design. One glaring issue facing clinical trials is insufficiency in subject numbers, which results in reduced power. As the number of subjects who complete the full course of a trial decrease, so does the power and validity of the study. , Shiovitz and associates estimated that a study initially powered at 90% could have an actual power of 50% to 87% depending on the percentage of subjects (40% to 5%, respectively) contributing noninformative data owing to noncompliance to study procedures. Similarly, a study with sufficient subjects may also be inconclusive if the procedures of the study are not strictly followed.
Despite the substantial impact subject noncompliance can have on the results of a clinical trial and relatively high rates of noncompliance in some well-designed studies, research into noncompliance and loss to follow up (LTFU) in prospective studies has been limited, especially in ophthalmology. Noncompliance research in ophthalmology has been more thoroughly studied in clinical settings, with researchers utilizing either questionnaires, demographics, or disease severity to predict LTFU in subjects dealing with diseases such as glaucoma, neovascular age-related macular degeneration (nAMD), and diabetic retinopathy (DR). These studies, however, may not address the factors leading to noncompliance of subjects specifically enrolled in clinical trials. The current study aimed to identify the demographic, systemic medical, and ocular factors that may predict rates of noncompliance among subjects of prospective retina trials through the analysis of 10 investigator-initiated trials studying nAMD, proliferative diabetic retinopathy (PDR), and retinal venous occlusive diseases (RVO).
Methods
Subjects were included in this retrospective review if they were enrolled/randomized in any of 10 investigator-initiated trials ( Table 1 ) involving eyes with nAMD, PDR, RVO, or diabetic macular edema (DME). The current study excluded subjects who did not conduct their study visits at one of the research sites of Retina Consultants of Houston (Houston, Texas, USA). These prospective studies included SAVE, DAVE, WAVE, RAVE, TURF, ENDURANCE, TREX-AMD, TREX-DME, HULK, and RECOVERY. Prospective institutional review board (IRB) approval (Sterling IRB, Atlanta, Georgia, USA; Patient Advocacy Council IRB, Mobile, Alabama, USA; Western IRB, Olympia, Washington, USA; Quorum Review IRB, Seattle, Washington, USA) was obtained for all 10 Health Insurance Portability and Accountability Act–compliant study protocols. Informed consent for both participation in the trial and treatment were collected prior to enrollment in each study. All subjects were scheduled for visits involving regular treatment throughout periods ranging from 6 to 36 months ( Table 1 ). Subjects for each study underwent examination and diagnostic imaging according to their respective protocols at no monetary cost to the subject. Subjects were expected to arrange transportation to and from their study site without financial or personal assistance from the study site.
| Study Name | ClinicalTrials.gov Registration | FDA IND | Setting | Length (Months) | Disease State | Management Cohorts | Patients Analyzed |
|---|---|---|---|---|---|---|---|
| RAVE | NCT00406471 | 12246 | Clinical practice | 36 | Ischemic CRVO | 1. RBZ/0.3 mg q4w 2. RBZ/0.5 mg q4w 3. RBZ/1.0 mg q4w. All cohorts observed after 9 months and PRN q4w after 12 months. | 20 |
| SAVE | NCT01025232 | 106985 | Clinical practice | 24 | nAMD | 1. RBZ/2.0 mg PRN q4w 2. RBZ/2.0 mg PRN q6w. Both cohorts had 3 q4w loading injections and PRN q4w after 12 months. | 89 |
| TURF | NCT01543568 | 12462 | Clinical practice | 6 | nAMD | AFT/2.0 mg PRN q4w after 3 q4w loading injections. | 46 |
| DAVE | NCT01552408 | 113691 | Clinical practice | 36 | DME | 1. RBZ/0.3 mg PRN q4w 2. RBZ/0.3 mg PRN q4w + TRP. Both cohorts had 4 q4w loading injections. | 40 |
| WAVE | NCT01710839 | 12246 | Clinical practice | 12 | Ischemic CRVO/BRVO | RBZ/0.5 mg PRN q4w+ TRP after 6 q4w loading injections. | 30 |
| TREX-AMD | NCT01748292 | 116786 | Multicenter | 36 | nAMD | 1. RBZ/0.5 mg q4w 2. RBZ/0.5 mg TREX after 3 q4w loading injections. | 30 |
| TREX-DME | NCT01934556 | 119146 | Multicenter | 36 | DME | 1. RBZ/0.3 mg q4w 2. RBZ/0.3 mg TREX 3. RBZ/0.3 mg TREX + GILA PRN. All cohorts received 4 q4w loading injections and were PRN q4w after 24 months. | 56 |
| Endurance | NCT02299336 | None | Clinical practice | 24 | DME | AFT/2.0 mg PRN q4w after 5 q4w loading injections. | 26 |
| RECOVERY | NCT02863354 | 131056 | Clinical practice | 12 | PDR | 1. AFT/2.0 mg q4w 2. AFT/2.0 mg q12w | 43 |
| HULK | NCT02949024 | 115683 | Multicenter | 6 | DME | CLS-TA/4.0 mg PRN q4w | 10 |
At the time of their screening visits, subjects self-reported demographic information including age, gender, race, marital status, employment status, insurance, zip code, and smoking status. Clinical data, such as best-corrected visual acuity (VA) and optical coherence tomography (OCT) measurements used in this review, were collected at study visits. All OCT information was retrieved using the Heidelberg Spectralis HRA+OCT (Heidelberg Engineering, Heidelberg, Germany). Further data, such as blood pressure, medical history, ocular history, previous enrollment in trials, and reasons for withdrawing consent, were collected from the study documents. The definition of serious comorbidities was restricted to history of myocardial infarction, stroke, amputation, end-stage liver failure, end-stage renal failure, chronic obstructive pulmonary disease, and/or active cancer diagnosis for which the subject was receiving active treatment. Mean regional household income was calculated using data acquired using the U.S. Census Bureau American Community Survey 2017. The distances of each subject from his or her residence to the study clinic were calculated using the latitude and longitude of the centroid of each zip code, as determined by the U.S. Census Bureau ( Simplemaps.com Version 3.9; Pareto Software, Cincinnati, Ohio, USA). Two sets of coordinates, one of a subject’s residence and the other of the study site, were then converted into distances using the Haversine formula. Thirty-three variables were chosen for analysis based on either hypothesized impact on noncompliance or identification by previous literature as significantly impacting patient compliance in routine clinical care.
Subjects were divided into 2 cohorts: subjects who had no missed visits (compliant) and subjects who had either missed at least 1 study visit or exited their specific study early (noncompliant). All available data were analyzed using 2-sided Student t test and Pearson χ 2 test, conducted in RStudio (Version 1.0.136; RStudio Inc, Boston, Massachusetts, USA). To perform the χ 2 test and regression analysis, each continuous variable was divided into 2 or more categories with comparable numbers of subjects in each group. SPSS (Version 25.0; SPSS Inc, Chicago, Illinois, USA) was used to perform multivariable and univariable logistic regression, in which 0 and 1 were coded as compliant and noncompliant, respectively. The Hosmer-Lemeshow (HL) test was performed to evaluate the fit of each multivariable regression model, which was carried out using the enter method, with an HL P value less than .05 indicating an inaccurate model. Although both multivariable and univariable regression results are reported in the tables, only significance indicated by multivariable logistic regression is referred to in the discussion. Odds ratio (OR), 95% confidence interval (CI) of the OR, and P value were collected for each variable in the regression models. A P value less than .05 was considered statistically significant.
Results
Of 390 subjects analyzed, 212 (54.4%) were fully compliant with all study visits and 178 (45.6%) met the criteria for noncompliance, with 125 (32.1%) subjects completing the study with at least 1 missed visit and 53 (13.6%) subjects exiting the study early. Subjects exited early owing to worsening health (n = 10, 18.9%), death (n = 8, 15.1%), no longer wishing to participate in the study (n = 7, 13.2%), adverse event (AE) occurrence (n = 5, 9.43%), work (n = 3, 5.66%), relocation (n = 3, 5.66%), or transportation problems (n = 1, 1.89%), with 16 (30.2%) subjects being lost to follow-up without documented explanation owing to termination of contact with the study site and inability of the study site to contact the patient despite multiple attempts. Of 33 variables analyzed, 17 were found to be statistically significantly associated with noncompliance; 17 variables and 8 variables were significant predictors of noncompliance in the univariable and multivariable regression models, respectively.
Noncompliance by Demographic Characteristics
Noncompliance rates decreased significantly with increasing age, with a rate of 58.7% for subjects aged <60 years, 45.7% for subjects aged 60 to 74.5 years, and 32.3% for subjects aged ≥75 years ( P = 1.47e-4; Table 2 ). There was a significant difference ( P = 1.46e-5) in mean age between the compliant (68.5 years; CI: 66.7-70.3) and noncompliant (62.6 years; CI: 60.7-64.5) cohorts. Distance from study site were not significantly associated with noncompliance rates in the χ 2 analysis, with rates of 49.0% for subjects living <15 miles from the study site, 39.0% for subjects living between 15 and 25 miles from the study site, and 48.4% for subjects living ≥25.5 miles from the study site ( P = .20). Noncompliance rates were significantly associated with employment status ( P = .026) and insurance status ( P = 7.21e-4). Subjects who were employed had a 52.0% noncompliance rate compared with the 40.7% noncompliance rate of subjects who are not employed (either unemployed, retired, or disabled). Noncompliance rates were lower among subjects with government health insurance (34.4%, Medicare or Medicaid) than with no health insurance (55.6%) or with private health insurance (53.2%). Whereas age, distance from study site, employment status, and insurance were significant in univariable regression, distance was the only significant risk factor in determining noncompliance in multivariable regression ( Table 3 ).
| Compliant (N = 212) | Noncompliant (N = 178) | Total (N = 390) | P Value | |
|---|---|---|---|---|
| Demographic characteristics, n (%) | ||||
| Age | ||||
| <60 years | 52 (24.5) | 74 (41.6) | 126 (32.3) | .000147 |
| 60-74.5 years | 76 (35.8) | 64 (36.0) | 140 (35.9) | |
| ≥75 years | 84 (39.6) | 40 (22.5) | 124 (31.8) | |
| Employment status | ||||
| Not employed | 128 (60.7) | 88 (49.4) | 216 (55.5) | .0264 |
| Employed | 83 (39.3) | 90 (50.6) | 173 (44.5) | |
| Insurance status | ||||
| None | 24 (11.3) | 30 (16.9) | 54 (13.9) | .000721 |
| Government | 107 (50.5) | 56 (31.5) | 163 (41.8) | |
| Private | 81 (38.2) | 92 (51.7) | 173 (44.4) | |
| Systemic medical history, n (%) | ||||
| Serious comorbidities | ||||
| 0 | 192 (90.6) | 148 (83.1) | 340 (87.2) | .0290 |
| >0 | 20 (9.43) | 30 (16.9) | 50 (12.8) | |
| Diabetes | ||||
| Diabetic | 92 (43.4) | 125 (70.2) | 217 (55.6) | 1.09E-7 |
| Not Diabetic | 120 (56.6) | 53 (29.8) | 173 (44.4) | |
| Concomitant medications | ||||
| <8 medications | 73 (37.4) | 40 (23.8) | 113 (31.1) | .000292 |
| 8-14 medications | 71 (36.4) | 51 (30.4) | 122 (33.6) | |
| ≥15 medications | 51 (26.2) | 77 (45.8) | 128 (35.3) | |
| Baseline systolic blood pressure | ||||
| <125 mm Hg | 58 (27.4) | 38 (21.3) | 96 (24.6) | .00194 |
| 125-144 mm Hg | 93 (43.9) | 58 (32.6) | 151 (38.7) | |
| ≥145 mm Hg | 61 (28.8) | 82 (46.1) | 143 (36.7) | |
| Previous clinic visits | ||||
| 1 visit | 56 (26.4) | 56 (31.5) | 112 (28.7) | 4.73E-5 |
| 2-19 visits | 49 (23.1) | 70 (39.3) | 119 (30.5) | |
| ≥20 visits | 107 (50.5) | 52 (29.2) | 159 (40.8) | |
| Ocular medical history, n (%) | ||||
| First recorded VA | ||||
| <65 letters | 62 (29.2) | 78 (43.8) | 140 (35.9) | .00742 |
| 65-79 letters | 109 (51.4) | 78 (43.8) | 187 (47.9) | |
| ≥80 letters | 41 (19.3) | 22 (12.4) | 63 (16.2) | |
| Fellow-eye VA | ||||
| <65 letters | 55 (26.1) | 58 (33.0) | 113 (29.2) | .0180 |
| 65-79 letters | 49 (23.2) | 54 (30.7) | 103 (26.6) | |
| ≥80 letters | 107 (50.7) | 64 (36.4) | 171 (44.2) | |
| First recorded CRT | ||||
| <250 μm | 54 (25.5) | 32 (18.0) | 86 (22.1) | .0282 |
| 250-299 μm | 47 (22.2) | 29 (16.3) | 76 (19.5) | |
| ≥300 μm | 111 (52.4) | 117 (65.7) | 228 (58.5) | |
| Change in CRT | ||||
| <−100 μm | 55 (25.9) | 75 (42.1) | 130 (33.3) | .00144 |
| −100 to −1 μm | 107 (50.5) | 78 (43.8) | 185 (47.4) | |
| ≥0 μm | 50 (23.6) | 25 (14.0) | 75 (19.2) | |
| Length of study | ||||
| 1-12 months | 94 (44.3) | 35 (19.7) | 129 (33.1) | 6.43E-11 |
| 13-24 months | 70 (33.0) | 45 (25.3) | 115 (29.5) | |
| 25-36 months | 48 (22.6) | 98 (55.1) | 146 (37.4) | |
| Disease studied | ||||
| DR | 67 (31.6) | 108 (60.7) | 175 (44.9) | 1.59E-10 |
| AMD | 122 (57.5) | 43 (24.2) | 165 (42.3) | |
| RVO | 23 (10.8) | 27 (15.2) | 50 (12.8) | |
| Total AEs | ||||
| <4 events | 64 (30.3) | 34 (19.4) | 98 (25.4) | .00130 |
| 4-9 events | 92 (43.6) | 66 (37.7) | 158 (40.9) | |
| ≥10 events | 55 (26.1) | 75 (42.9) | 130 (33.7) | |
| Moderate AEs | ||||
| 0 events | 102 (48.3) | 54 (30.9) | 156 (40.4) | .000493 |
| >0 events | 109 (51.7) | 121 (69.1) | 230 (59.6) | |
| Severe AEs | ||||
| 0 events | 179 (84.8) | 101 (57.7) | 280 (72.5) | 2.80E-9 |
| >0 events | 32 (15.2) | 74 (42.3) | 106 (27.5) | |
| Previous ocular studies | ||||
| 0 studies | 147 (69.3) | 139 (78.1) | 286 (73.3) | 3.594E-6 |
| >0 studies | 65 (30.7) | 39 (21.9) | 104 (26.7) |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
