Sharing of Individual Patient-Level Data by Trialists of Randomized Clinical Trials of Pharmacological Treatments for COVID-19

Abstract

This study describes access to individual patient-level data from randomized clinical trials during the COVID-19 pandemic to determine whether the intent to share what was reported in the registry, publication, or preprint was consistent with actual data access.

Access to individual patient-level data (IPD) of randomized clinical trials (RCTs) increases the reproducibility of research, facilitates exploration of effect modifiers, and leverages research investments beyond the original research question.¹ This study aimed to describe access to IPD of RCTs of pharmacological treatments for COVID-19 and to determine whether the intent to share IPD as reported in the registry and publication or preprint was consistent with actual access to IPD.

Methods

This study is part of the COVID-NMA initiative, a living systematic review of COVID-19 RCTs.² For this study, RCTs in any language were included if they assessed pharmacological treatments for COVID-19 and were posted as a preprint or published in a peer-reviewed journal from March 1, 2020, through May 31, 2021.³

For each RCT, the trialists’ intent to share IPD was extracted from the trial registry and the preprint or peer-reviewed publication. To conduct an IPD meta-analysis,² we sent email requests to corresponding authors and registry contacts of the RCTs followed by as many as 2 reminders as necessary from September 17, 2020, through September 22, 2021. They were asked to send us their trial IPD directly or through Vivli, an online data repository that offers free help to anonymize and archive COVID-19 data. Authors were offered the opportunity for coauthorship on the IPD-MA manuscript. To capture IPD for RCTs from nonresponding authors, we searched publicly available major data repositories Vivli, Yale Open Data Access, and Clinical Study Data Request. IPD was considered accessible if trialists sent us their database or made their database available through a clinical data repository with a clear data requisition process. All databases sent to us directly were checked for data quality. Characteristics of RCTs with accessible vs inaccessible IPD were compared post hoc with the 2-tailed Mann-Whitney U test or Fisher exact test with significance set at P < .05 using SAS version 9.4 (SAS Institute Inc).

Results

Two hundred twenty-four RCTs were included; 121 trialists (54%) responded. Ninety studies (40%) were preprints (Table). The median study sample size was 95 (IQR, 53-214). Sixty-eight trialists (30%) declared in the registry that they intended to share IPD and 138 (62%) in the publication or preprint (Figure). Overall, IPD was accessible for 50 of 224 RCTs (22%). Of these, 27 trialists provided IPD directly and 23 were available through online data repositories. Ten trialists initiated sharing their data but did not complete the process by December 31, 2021. RCTs with accessible vs inaccessible IPD had a statistically significant larger sample size (median, 118 [IQR, 64-388] vs 86 [IQR, 49-199]; P = .03). Furthermore, funding sources differed between the 2 groups (Table). Six RCTs were subsequently retracted, of which 1 had accessible IPD. The Figure presents the intention to share IPD as declared in the registry, the preprint, or the publication and the actual access to data. Twenty-seven percent (37 of 138) of RCTs with a positive data sharing statement in the preprint or publication had accessible IPD and 21% (14 of 67) with a negative data sharing statement in the registry had accessible IPD.

Table. Description of Sample and Proportion of Studies With Accessible Individual Patient-Level Data^a.

	No. (%) of randomized clinical trials			P valueb
	Total (N = 224)	IPD accessible (n = 50)	IPD not accessible (n = 174)
Study sample size, median (IQR)	95 (53-214)	118 (64-388)	86 (49-199)	.03
Status of report
Preprints	90 (40)	22 (44)	68 (39)	.62
Peer reviewed	134 (60)	28 (56)	106 (61)
Retracted	6 (3)	1 (2)	5 (3)	>.99
Funding source
Public or nonprofit	103 (46)	13 (26)	90 (52)	.003
Private	41 (18)	9 (18)	32 (18)
Mixedc	60 (27)	22 (44)	38 (22)
No funding or not reported	20 (9)	6 (12)	14 (8)
Country income statusd
Low to middle	148 (66)	27 (54)	121 (70)	.05
High	62 (28)	17 (34)	45 (26)
Low to middle and high	14 (6)	6 (12)	8 (5)
Setting
Single center	99 (44)	24 (48)	75 (43)	.63
Multicenter	124 (55)e	26 (52)	98 (56)
Data sharing statement in registry
Yes	68 (30)	14 (28)	54 (31)	.91
No	67 (30)	14 (28)	53 (30)
Undecided	24 (11)	6 (12)	18 (10)
Not reported or unclear	65 (29)	16 (32)	49 (28)
Data sharing statement in preprint or publication
Yes	138 (62)	37 (74)	101 (58)	.24
No	14 (6)	2 (4)	12 (7)
Undecided	1	0	1 (1)
Not reported or unclear	71 (32)	11 (22)	60 (34)

^a Percentages may not add up to 100 due to rounding.

^b Studies with individual patient-level data (IPD) accessible were compared with studies that did not have IPD accessible. When comparing qualitative characteristics within categories, the null hypothesis was that the proportion of each characteristic was constant across the type of study (ie, IPD accessible or IPD not accessible).

^c Mixed funding are studies with both public or nonprofit and private funding sources (eg, publicly funded with a drug donation).

^d Study countries were categorized according to whether the country was classified as a low- to middle-income country, high-income country, or both (eg, multiple study countries, ≥1 that qualified as a low- to middle-income country and ≥1 that qualified as a high-income country) according to the World Bank country income classification.

^e One study did not indicate whether it was a single-center or multicenter study.

Figure. Change in Data Sharing Intentions as Reported in the Registry, in the Preprint or Publication and Actual Individual Patient-Level Data Accessibility.

The x-axis starts with the trial’s data sharing statement as reported in the clinical trial registry, then in the publication or preprint, and lastly, the actual accessibility of the individual patient-level data (IPD). The y-axis represents intention to share (yes, no, not reported, undecided) and actual IPD accessibility (yes = IPD accessed, no = IPD not accessed). Percentages are calculated using the total number of studies (n = 224) as the denominator. The data underlying this Figure are shown in more detail in the Table.

Discussion

In this study of a planned IPD-MA, IPD was only accessible for a small proportion of RCTs despite the majority of trialists declaring an intention to share their data in the preprint or publication. The planned IPD-MA was not feasible considering the low access to data.

Despite an increase in data sharing requirements by funders and journals and the establishment of several secure data repositories, more needs to be done to increase data accessibility.^4,5 Investigators may hesitate to share their data because of concerns for patient consent or confidentiality, risk of data misinterpretation or misleading analyses, or lack of understanding of why and how to share IPD.⁶ Moreover, sharing IPD requires additional time and effort, and institutions do not sufficiently support researchers’ capacity to do so.⁵

Study limitations include that authors may not have received the request. Also, this research was conducted during the COVID-19 pandemic, so results may not be generalizable.

Section Editors: Jody W. Zylke, MD, Deputy Editor; Kristin Walter, MD, Senior Editor.

Supplement 1.

Data Sharing Statement.