Evaluating the Impact of a Direct-to-Patient Clinical Trial Site on Performance, Access, and Patient Representation

Background

A growing body of evidence in the literature demonstrates the value of hybrid models integrating direct-to-patient (DTP) study conduct and remote technologies into clinical trial execution. These approaches—including the use of telemedicine, home visits, DTP clinical supplies, local labs, mobile and wearable devices—are typically integrated with traditional site-based activities.

The body of evidence shows that these decentralized executional models improved patient access, enhanced enrollment performance, and delivered greater operational efficiency. Sponsors and CROs are increasingly adopting these models to balance flexibility with regulatory and quality requirements.

One decentralized approach is the adoption of a DTP site—characterized by home visits conducted by licensed research nurses, DTP clinical supply, DTP Investigational Medicinal Product, central laboratory integrations, and digital data capture—all operating under centralized medical oversight and within established regulatory frameworks. This model is not a replacement for traditional brick-and-mortar sites; rather, it is designed to improve access, reduce patient burden, and enhance enrollment performance while maintaining research quality.

However, little to no evidence exists characterizing the use and impact of DTP site involvement in clinical trial execution. In early 2026, the Tufts Center for the Study of Drug Development, in collaboration with Science 37, conducted a study to address this gap.

Methods

Tufts CSDD and Science 37 collaborated to determine key metrics of interest for this research, including established metrics for which Tufts CSDD has benchmark data as well as novel metrics of performance that may help establish potential benefits or weaknesses of a DTP site, such as Science 37. The site evaluated in this analysis operates under centralized medical oversight provided by multi-state licensed investigator physicians and protocol-trained research nurses conducting in-home visits.

The model has undergone multiple FDA inspections with no Form 483 observations or significant findings, and data generated through this approach has supported regulatory submissions and approvals. Metrics agreed upon included measures of trial scope and complexity such as number of endpoints, number of eligibility criteria, number of procedures (unique and total), and number of sites; measures of enrollment performance including participants screened, enrolled, and completed; milestone dates to calculate cycle times; and participant demographic data.

Additionally, the DTP site sample included background information on the trials such as phase, therapeutic area, whether the trial was for an orphan indication, and the pivotal status of the trial, as well as a new measure of trial reach: average distance participants lived from a major airport (for participants enrolled at the DTP site only).

The DTP site sample included 28 studies intended to be representative of the trials conducted; however, it was a convenience sample selected for this analysis. Tufts CSDD benchmark metrics were pulled from multiple datasets created during recent, unrelated studies.

The primary dataset was a sub-group of 76 trials from a 2022 study that included only phase II and phase III trials that indicated that they had not included any DCT solutions. For metrics not available in the primary dataset (indicated with an * in the results section), a sub-group of 56 trials from a 2025 study that included only phase II and phase III trials that did not indicate the use of at-home options for procedures was used.

Certain site-level metrics were examined as well; for these, data on 857 sites from 15 trials from a 2025 study were used. Benchmark demographic data were pulled from a previously published Tufts CSDD study.¹

Data quality checks were performed on the data from the DTP site sample, and suspicious datapoints (i.e. participant counts resulting in >100% rates, milestone dates leading to negative cycle times) were either confirmed or corrected. From the data, standard metrics and cycle times such as Screen Failure Rate, Dropout Rate, First Patient First Visit to Last Patient First Visit, and First Patient First Visit to Database Lock, were calculated, both planned and actual values.

Raw differences between actual and planned values were calculated, as were those differences as a percentage of the planned values. Descriptive statistics including means, medians, and coefficients of variation were calculated for both the DTP site and benchmark samples.

For enrollment metrics, three outlier trials were removed from the DTP site sample—one for being a numeric outlier and two because the trial designs included uncommon requirements resulting in unrepresentative enrollment numbers. During initial analysis, Tufts CSDD calculated enrollment metrics both with and without these outlier trials, and although mean counts were noticeably affected, medians were much less impacted, and percentages showed little change.

Further, with the outlier trials excluded the means are lower, providing a more conservative estimate of these metrics.

Results

Table 1 shows several characteristics for both the DTP site and primary benchmark sample. Both datasets were primarily or entirely phase II and phase III trials and majority non-orphan indications.

Both include a variety of therapeutic areas, although the DTP site sample has a higher percentage of CNS (21.4%) and gastroenterology trials (21.4%) while the benchmark sample has a higher percentage of dermatology (22.4%) and immunology trials (23.7%).

Table 2 contains four contributors to protocol complexity: total number of eligibility criteria, total number of endpoints, number of unique procedures, and number of total procedures. In all four of these measures, the DTP site sample is lower than the benchmark.

Table 3 contains the same four contributors to protocol complexity, but only for the DTP site sample, grouped by the year the trial started. In this case, the average for all four metrics is higher in trials started between 2022 and 2025 than for trials started before 2022, indicating that the complexity of the trials involving the DTP site are increasing.

Average enrollment numbers (number of participants screened, enrolled, and completed) are all much higher for the DTP site sample compared to benchmark average for individual sites (Table 4). While screen failure rates are similar between the two samples (40.4% and 39.6%, respectively), completion rates are much higher for DTP site sample as well (86.3% and 35.7%, respectively).

Two often valued cycle times are shorter in the DTP site sample than in the benchmark sample: first patient first visit to last patient first visit—381.2 days in the DTP site sample and 454.9 days in the benchmark sample—and first patient first visit to database lock—693.0 days in the DTP site sample and 814.2 days in the benchmark sample (Table 5).

Table 6 provides the raw difference between actual and planned values for enrollment and cycle time metrics, and that difference as a percentage of the planned values. Looking at raw differences between actual and planned, while on average, the DTP site under enrolls by a wider margin compared to the benchmark for trial sites (-20.1 and -1.3 respectively), looking at the difference as percentage of planned enrollment, it is closer to plan than the benchmark (-15.7% and -33.2% respectively).

More participants at the DTP site also completed the trials than planned (7.0), while the average for benchmark sites is somewhat below plan (-0.8).

Trials that included the DTP site tended to enroll more female participants than male (62.4% female), whereas the benchmark trials with no DCT solutions are more closely divided (49.0% female) (Table 7). Trials in the DTP site sample also tend to enroll a lower percentage of white participants (59.3%) and a higher percentage of Black or African American participants (19.9%) compared to traditional trials.

Traditional trials, on average, enroll higher percentages of Asian participants (14.2%) and Hispanic participants (12.6%). Caution is urged when interpreting this table, though, as there are methodological differences in how the data were recorded.

Namely, the DTP site sample includes “not reported” categories not included in the benchmark sample, and the DTP site sample includes Hispanic among the racial demographic groups, whereas it is considered separately in the benchmark sample.

Table 8 contains several metrics around participant enrollment examining the contributions of the DTP site. On average, the DTP site enrolls around 20% to 30% of the participants of a trial, with a mean of about 150. Within these trials, other sites average around 26 participants per site.

Discussion

The analyses here present several compelling indications that the inclusion of DTP sites in a trial can improve both trial performance and participant access and convenience. DTP sites can screen, enroll, and complete far more participants than a single traditional site, with similar screen failure rates and far higher completion rates.

Trials that include them tend to have shorter cycle times compared to a benchmark sample, and several key performance metrics beat planned by a wider metric than the benchmark. Inclusion of DTP sites may help improve patient representation among participant samples, and these analyses give a good first look at the reach DTP sites may be achieving.

These advantages are consistent with the results of other Tufts CSDD studies exploring the advantages the inclusion of DCT solutions can have. One benefit that DTP sites appear to provide is their ability to enroll participants.

In the sample included here, the DTP site typically enrolled between 1/5 and 1/3 of the total participants in a trial, with the average number of participants enrolled 6x the average of the remaining sites in their trials. Compared to the benchmark for sites, they were able to enroll what might otherwise require around 25 sites.

Further, in this sample, the DTP site had an average of 130 participants complete the trial while individual sites in the benchmark sample averaged around 3 participants. The DTP site had roughly as many completed participants as 42 sites in the benchmark sample.

Adding to this, the median number of participants completing the trial per site in the benchmark was 0, indicating that only half the sites in that sample had any participants complete their trial. While cycle times present a less clear picture, there are still strong indications of advantages brought by the inclusion of a DTP site.

Several cycle time durations were shorter for the DTP site sample compared to the benchmark sample—first patient first visit to last patient first visit, last patient first visit to last patient last visit, and first patient first visit to database lock. However, last patient last visit to database lock was much longer in the DTP site sample.

Given that several indicators of protocol complexity show the DTP site sample may include less complex trials than the benchmark sample, it also makes sense to look at the actual to planned comparisons. Specifically, the delta between actual and plan as a percentage of the planned duration.

Here, although the average for the benchmark sample outperforms the DTP site sample in two cycle times—first patient first visit to last patient first visit, and last patient last visit to database lock—the DTP site sample outperforms the benchmark overall, with first patient first visit to database lock being nearly 13% shorter than planned, compared to about 1% longer than planned in the benchmark sample. Medians also favor the DTP site sample, with half the sample beating plan by 12% or more, compared to the benchmark where half the sample beat plan by about 2% or more.

Patient representation is another area that may be impacted by DTP methodologies, as the removal of certain barriers to participation could allow for more recruitment of participant samples that are more representative of the target patient population as a whole. Within the DTP site sample, the average percentage of white participants was much lower, and the percentage of Black participants was much higher compared to the benchmark.

However, in the benchmark, the percentage of Asian participants was much higher, and the percentage of Hispanic participants was moderately higher compared to the DTP site sample. These differences speak to the fact that improving patient representation in clinical trials does not have a single solution and that patient preferences must always be considered in trial design.

Cultural and even personal preferences will always play a significant role in whether patients enroll in a trial. Applying a single solution broadly may help improve the representation of specific demographic group while hindering recruitment in another. Strategies must be deployed with the target populations and patient preferences in mind.

These analyses provide an exciting first look at another topic that is often brought up around DTP sites: reach. Measuring the reach (typically thought of as the distance trial participants travel to reach the trial site) of a traditional site is challenging, and measuring the reach of a DTP site also poses challenges.

However, the sample provided here did include data on the average distance participants recruited by the DTP site were from a major airport. While Tufts CSDD does not have an existing benchmark for the reach of a traditional site, previous research indicates that increased travel distance to clinical sites significantly increases the burden of trial participation, and that travel to and from a trial site is one of the most burdensome aspects of clinical trial participation.²

Given that the majority of clinical trial sites are set in urban locations, it is likely that participants needing to travel 60 or more miles to reach a major airport would be similar distances from a clinical trial site, as well. While this analysis provides some of the first empirical evidence regarding the impact of DTP sites on trial performance, it has a few limitations.

First is the sample size, and the fact that the sample was provided by a single organization. Given these, the results may not be generalizable to all DTP sites. Second, in roughly half of the DTP site sample, the DTP site was only one site among many.

As a result, many of the metrics examined, particularly cycle times, may reflect the performance of the other traditional sites as much or more than the impact of a DTP site. Additional research will need to be conducted looking at a larger sample gathered from multiple organizations, as well as looking at trial performance in trials that relied solely on a DTP methodology.

Despite these limitations, the results presented here provide compelling first evidence of the impact the inclusion of DTP sites may have on trial performance. As mentioned regarding patient representation, the inclusion of DTP sites is not a one-size-fits-all solution to improve trial performance.

Instead, they are one of many possible solutions, albeit an effective one, that can be deployed. Preferences and needs vary from person to person, and although some potential trial participants will continue to prefer a traditional trial site, including a DTP site along with traditional sites will offer the opportunity to participate to those who may otherwise not enroll.

This may be especially beneficial when working with patient populations that are particularly burdened by travel or otherwise unable to access traditional sites. As clinical trials continue to evolve, integrating DTP clinical trial sites alongside traditional infrastructure may be critical to improving both trial efficiency and equitable patient access.

About the Authors

Zachary Smith and Kenneth Getz, Tufts Center for the Study of Drug Development.

Debra Weinstein; Liz Jones; Kelly McKee; Tyler Van Horn, Science 37.

References

1. Smith Z, Getz K. Examining the Association Between DCT Solutions Use and Participant Diversity in Clinical Trials. Ther Innov Regul Sci. 2025 Nov;59(6):1204-1210. doi: 10.1007/s43441-024-00659-w. Epub 2024 May 11. PMID: 38734837.

2. Smith Z, Wilkinson M, Carney C, Grove N, Qutab B, Getz K. Enhancing the Measure of Participation Burden in Protocol Design to Incorporate Logistics, Lifestyle, and Demographic Characteristics. Ther Innov Regul Sci. 2021 Nov;55(6):1239-1249. doi: 10.1007/s43441-021-00336-2. Epub 2021 Aug 30. PMID: 34460095.