Biometrics field trial

As of today, the literature has focused on concerns about data validity, accuracy, provenance, and regulatory issues [ 4 , 7 , 13 ]. To our knowledge, only one study systematically examined the outcomes measured with activity monitors in oncology trials and found a lack of standardization in the types of BMDs used and how their data were collected, analyzed, and interpreted [ 14 ].

Other problems such as how to manage incomplete BMD outcome data e. However, conventional data imputation strategies are not adapted to manage the temporal variations in multivariate time series associated with these types of data [ 3 , 16 ]. Especially, we aimed to answer four questions: 1 Which outcomes are measured with BMDs in trials?

For example, we did not include reports of trials that used outcomes collected by using smartphone apps that prompted patients with questions or patient self-reported data obtained from wearable devices. We excluded reports of trials with outcomes measured by BMDs during surgery or intensive care and trials of children or healthy participants.

We also excluded systematic reviews or meta-analyses, diagnostic studies, methodological publications, editorial-style reviews, abstracts and posters, and secondary analyses of trials. A first set of search equations were developed for each database and were based on Medical Subject Headings MeSH terms and specific free-text words pertaining to wearable devices and continuous measurements Supplementary Appendix 1 a.

This search was conducted on February 4, During peer-review of this article, we complemented our initial search with several additional MeSH terms Supplementary Appendix 1 b. The second search was conducted on June 26, To avoid missing trials because 1 of no validated search strategy to retrieve trials using BMDs as outcomes and 2 the names used to describe BMDs are ever-changing sensors, wearables, trackers, continuous monitoring devices, etc.

For duplicate publications i. One reviewer CG conducted the electronic database searches and another LG conducted the page-by-page hand search. First, they examined titles and abstracts when available and selected full-text articles according to the specified eligibility criteria.

If an abstract was not provided by the database it originated from, and the title appeared to be potentially relevant, we reviewed its full text.

At each stage, we recorded the records retrieved and excluded. For each trial, 2 reviewers CG, VTT independently extracted data from studies by using a standardized form. They recorded the trial characteristics journal name; date of publication; medical area; number of randomized patients; funding source, types of interventions [i.

Then, they extracted all outcomes measured with BMDs and classified them as primary or secondary outcomes. We considered primary outcomes as those that were explicitly reported as such in the published article or in the entry in a public clinical trial registry or, if none was explicitly reported, the outcome s that was stated in the sample size estimation.

All other outcomes were considered secondary outcomes. This framework was chosen because it represents the expected standard required to detect a change between prespecified and published outcomes. Outcomes were defined by their 1 concept of interest e. Results were synthesized by presenting all outcome definitions used to measure the same domain, by concept of interest. For each outcome, two reviewers extracted the precise sensor used and its type inertial measurement units [e.

Then, they evaluated the information reported in the published articles regarding the validity, reliability, and responsiveness of sensors. Information was classified as 1 reported and documented e.

Two reviewers LG, VTT independently compared the outcomes reported in the methods section of published articles with those registered in the latest entry in a public trial registry e. Trial registration numbers were extracted from articles. When no trial registration information was found, the trial was considered unregistered. They distinguished 2 situations:. To assess the management of missing data, the reviewers searched the methods section of articles for classical methods used to handle missing data e.

If no specific method was found, the reviewers evaluated whether the analysis was performed on only complete cases by examining the number of patients analyzed for each outcome. There was some information on the treatment effect for the given patient but it may be limited e. Because of no consensus on methods to account for incomplete outcome data [ 20 ], the reviewers assessed 1 how incomplete BMD outcome data were defined by authors, 2 methods to account for incomplete data, and 3 the amount of incomplete BMD outcome data present in trials.

Data are presented as number percentage for qualitative data and median [interquartile range IQR ] for continuous data. All analyses involved using R v3. The electronic search identified records. We identified records on the basis of the title and abstract and 15 by our page-by-page hand search.

Study flow chart. During peer review, we modified the search equation by adding several new terms. During this second search, conducted on June 26, , we screened and included all eligible trials not included in the initial search. We found substantial heterogeneity in the definition of outcomes used to measure the same concept of interest, with different unique outcome definitions for glycemic control of outcomes , 2 for the assessment of diabetic foot complications of 2 outcomes , 46 for physical activity of 87 outcomes , 13 for blood pressure control of 32 outcomes , 20 for sleep quality of 39 outcomes , 13 for adherence to treatment of 22 outcomes , 2 for pulmonary capacity of 2 outcomes , and 14 for heart rate variability of 14 outcomes Appendix 3.

Heterogeneity in outcome definitions was due to varying combinations of domains, time frames, or algorithm used to process the raw data Fig. For example, in diabetes trials, thresholds used for glucose target range could be 3. Details of the unique definitions of outcomes measured with BMDs are in Appendix 5.

Each node represents a given outcome definition characterized by its domain, measurement method, metric, aggregation method, and time frame. The size of nodes represents the number of times each outcome definition was used in the included trials. Outcome definitions are clustered by outcome domains.

For one trial, the registration number provided in the published article linked to a different study. Examples are provided in Table 4. The remaining trials used classical methods to deal with missing BMD outcome data e. Similar to outcome definitions, the threshold for incomplete outcome data varied in all trials.

To our knowledge, this is the first methodologic systematic review describing how BMDs are used to assess outcomes in a broad sample of recent RCTs, across different conditions. Our findings showed that many trials, including drug development trials, are using BMDs to collect both primary and secondary outcome measures.

Because of no validated search strategy to retrieve trials using BMDs to collect outcomes with ever-changing names used in the literature: sensors, wearables, trackers, continuous monitoring devices, etc. To account for this issue, we complemented our electronic search strategy with a page-by-page hand search in the top 5 general medical journals.

This ensured that the most influential trials, likely to impact practice, would be included, but may have led to omit some trials published in specialized journals in digital health. In this sample of trials, we highlighted several challenges that may affect the validity and transferability of results from trials using BMDs to collect outcomes.

First, our review highlighted an alarming heterogeneity in the definition of outcomes measuring the same concept of interest across the RCTs. With this heterogeneity, comparing and combining RCT results e. One solution to improve the harmonization of outcomes in trials for a given condition is the definition of core outcome sets COSs. COSs are agreed-upon standardized sets of outcomes that should be minimally measured in all trials of a specific clinical area [ 28 ].

To our knowledge, no existing COS promotes the use of outcomes measured by BMDs nor proposes solutions for the standardization of these measures. For example, a recent COS proposed for type 1 diabetes includes only glycated hemoglobin as measure of glycemic control, whereas most recent major trials of this disease have used results from continuous glucose monitoring [ 23 , 29 , 30 ].

Second, BMDs are transforming outcome assessment in clinical trials: instead of single measures at given time points, researchers can now analyze dense longitudinal data and better understand the dynamic effects of interventions over time and outside of usual experimental contexts. Still, at the trial level, this situation contributes to an excessive number of summarized measures of the treatment effect generated from the same raw data, with often no pre-established priority.

Interpreting results was difficult, especially for outcomes with discordant directions of results. This situation represents a risk of selective outcome reporting bias. This is twice what was known for trials not using BMDs [ 32 , 33 , 34 ].

To account for this problem, we call for extensions to the SPIRIT and CONSORT statements specific to the inclusion of BMD outcome data in clinical trial protocols and reports, similar to what has been done for the integration of patient-reported outcome data in these documents [ 35 , 36 , 37 ]. Third, the need to ensure that measurements from BMDs are valid, reliable, and responsive has been underlined in multiple publications and guidelines [ 4 , 10 , 13 , 38 ].

Retrieving the studies evaluating the measurement properties of the devices used in trials is a difficult task. Finally, our review uncovers the problem posed by incomplete outcome data due to poor compliance with the BMD during the trial. Subsequently, major impacts on study results can occur keyword: placebo or nocebo effect , which in turn may lead to non-consideration of the study results by the scientific community or the local authorities.

CCMD would like to support your company in biometrics. We offer you individually coordinated statistical planning in the initial phase as well as statistical support during the whole study, so you can carry out clinical trials without being bothered by any concerns.

You can fully concentrate on the content of the study without having to be afraid the planned study could subsequently be classified as non-significant due to methodological or statistical errors.

Save time and resources through well-designed biometrics planned in time All Services. Clinical biometrics that are a step above Creative study design based on a deep understanding of standards, statistics, and clear communication. Download Case Study. Tell us about your project requirements The deep experience and successful track record of our Biometrics Team are just a few of the many reasons to choose Precision for Medicine.

Clinical trial services. Global Clinical Trial Footprint Sample processing labs, clinical trial sites and offices in five continents provide the clinical reach and scale to manage complex global programs. Clinical Development Strategy Tailored strategies consider the scientific, regulatory, and commercial factors that shape each trial, mitigating risk and advancing the development pathway. Clinical Trial Design Advanced trial-design approaches—including basket, umbrella, and adaptive trials—deliver biomarker driven clinical research.

Biostatistics Seasoned biostatisticians and statistical programmers deliver insight into every trial phase, from study design to regulatory submissions, all backed by meticulous documentation and data monitoring.

Clinical Sample Management Sample inventories from a global network of labs supply real-time processing in 55 countries; consolidated data from central labs, screening labs, and specialty labs with clinical data create actionable reports.