Association between decentralised clinical trial adoption and trial duration: a retrospective cross-sectional study of metabolic disease trials

Abstract

Objectives

To examine the association between decentralised clinical trial (DCT) adoption and trial duration in metabolic disease trials.

Design

Retrospective cross-sectional analysis using analyst-curated metadata from the GlobalData Clinical Trials Database, matched with ClinicalTrials.gov records via unique identifiers.

Setting

Industry-initiated phases 1–3 trials for metabolic diseases involving the USA (first patient enrolment 2015–2023).

Participants

444 trials (phase 1: n=140; phase 2: n=155; phase 3: n=149).

Main outcome measures

The primary outcome was clinical trial duration (CTD), defined as the interval from first patient in (FPI) to last patient last visit. The secondary outcome was the primary completion period (PCD–FPI), used for sensitivity analysis.

Results

Among 444 trials, 124 (27.9%) were identified as DCTs. Adoption differed significantly across clinical phases (phase 1: 11.4%; phase 2: 29.7%; phase 3: 41.6%; $p\S amp;lt;\mathrm{}0.001$). Two-way analysis of variance showed that clinical phase was significantly associated with CTD ($F=\mathrm{}27.4,\mathrm{}\mathrm{}\mathrm{}p\S amp;lt;\mathrm{}0.001$), whereas the main effect of DCT adoption was not significant ($F=\mathrm{}0.202,\mathrm{}\mathrm{}\mathrm{}p=\mathrm{}0.653$). Phase 2 and 3 DCTs exhibited numerically shorter mean durations (17.3 vs 18.3 months; 23.4 vs 25.0 months), but these differences did not reach statistical significance ($p\S amp;gt;\mathrm{}0.05$). In phase-stratified regression analyses, DCT status remained non-significant across all phases. Older adult inclusion was associated with shorter CTD in phase 3 $(\mathrm{\beta }=\mathrm{}-0.39,\mathrm{}\mathrm{}\mathrm{}p\S amp;lt;\mathrm{}0.001$). Sensitivity analysis using PCD-FPI yielded consistent findings.

Conclusions

DCT adoption was not significantly associated with shorter trial duration in metabolic disease trials after adjustment for clinical phase and clinical characteristics. These findings may reflect the current stage of DCT implementation, in which operational complexities may coexist with theoretical expectations of efficiency. Further evaluation in more mature implementation settings may clarify whether decentralised approaches are associated with improved efficiency.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This study integrated metadata from ClinicalTrials.gov and the GlobalData Clinical Trials Database, enabling a large-scale real-world assessment of decentralised clinical trial adoption across clinical phases.

• The application of a systematic operational mapping framework enabled the standardised classification of decentralised elements.

• The study population was restricted to industry-initiated trials for metabolic diseases to enhance design homogeneity across therapeutic areas.

• Sensitivity analysis using the primary completion date minus first patient in primary completion period was performed to verify the robustness of duration metrics.

• As a retrospective cross-sectional study based on secondary registry data, temporal ordering could not be established and causal inferences are not possible; residual confounding from unmeasured factors or protocol-level characteristics may remain.

Introduction

Decentralised clinical trials (DCTs) are transforming the clinical trial landscape, offering potential improvements in efficiency and accessibility. Since their rapid expansion during the COVID-19 pandemic, DCTs have gained global prominence as a sustainable research model. By leveraging digital health technologies to shift trial activities to participants’ homes, DCTs aim to reduce participant burden and improve the diversity and representativeness of clinical evidence.^{1 2} As of October 2023, eight countries, along with the European Medicines Agency (EMA, December 2022)³ and the US Food and Drug Administration (FDA, May 2023),⁴ have established frameworks for conducting DCTs by issuing relevant guidelines. Following the EMA and FDA, countries such as Australia,⁵ Singapore,⁶ Sweden,⁷ Switzerland⁸ and Canada⁹ have issued similar guidelines. Moreover, Korea is in the process of laying the groundwork for introducing DCTs by revising the Clinical Trial Electronic Consent Guidelines (December 2023) and operating a public–private expert advisory committee.^{10 11}

Recent analyses indicate a significant rise in DCT adoption, with the USA emerging as a primary hub for decentralised research, accounting for over half of identified DCT activities globally.¹² Although the USA has seen a rapid integration of decentralised elements, a notable gap remains between theoretical expectations and real-world implementation maturity.² This shift is particularly pronounced in therapeutic areas such as metabolic disorders, which rank among the leading sectors for DCT utilisation due to their operational complexity and inherent suitability for remote physiological monitoring.^{12 13} According to the FDA, DCTs are defined as those that involve conducting some or all clinical trial processes at locations other than traditional clinical trial sites and are not a new concept or technology. These trials can use either a fully decentralised approach or a partially decentralised hybrid approach depending on the extent of activities conducted outside the clinical site.⁴ They are expected to enhance the efficiency of clinical trials by saving time and costs while increasing the representativeness and applicability of trial outcomes. Additionally, they may improve participant autonomy, potentially reduce inconvenience and distress, facilitate trial participation and retention, and help alleviate the burden on trial staff through more efficient participant selection.^{12 14},¹⁶

The clinical development period of a drug is estimated to span approximately 10.5 years, which is a critical factor for companies in terms of access to new drugs and research and development productivity. This period constitutes a substantial portion of the timeline and cost of overall drug development.^{17 18} Clinical trials are typically categorised into phases 1, 2 and 3, with both the number of subjects and the duration of the clinical development period increasing in the later phases.¹⁹ Individual clinical trials encompass setup, recruitment and enrolment, intervention and follow-up, and close-out and reporting. The dates of the first patient enrolment and last patient visit are recorded on ClinicalTrials.gov and are used to define the clinical trial duration (CTD). Trial duration varies not only by phase but also by factors, including therapeutic areas, trial size, inclusion of older adults,^{20 21} enrolment timing, multinational status and site size.¹³²²,²⁵

DCTs have been proposed as a strategy that may be associated with shorter CTD^{25 26}; however, existing studies remain inconclusive.²⁷ Specifically, some empirical evaluations have reported that digital elements are associated with faster patient recruitment and improved retention, thereby shortening timelines.^{28 29} For instance, in an analysis of clinical trial data from various stakeholders, DiMasi et al suggested that DCT elements were linked to approximately 10% shorter phase II and III development cycle times, or roughly 3 months per phase. Beyond time savings, their analysis demonstrated that DCTs, compared with non-DCTs, were associated with reduced screen failure rates and a lower frequency of substantial protocol amendments, suggesting broader operational efficiencies. However, that study did not fully account for confounding factors—such as therapeutic areas and trial size—thereby limiting its applicability.³⁰ In contrast, other analyses have indicated that the operational complexity and data management challenges inherent in decentralised designs have been associated with operational delays in some settings and with no statistically significant differences in trial duration compared with traditional site-based trials.³¹,³³ These discrepancies are often attributed to methodological heterogeneity and uncontrolled confounding factors.²⁷ Therefore, given the inconsistent findings regarding the association between DCT adoption and trial duration, further investigation is warranted.

Furthermore, concerns have been raised regarding the lack of sufficient evidence for optimal DCT adoption, the need for adequate training and oversight of sponsors and investigators, and potential delays in the integration of new technologies.³⁴,³⁶ Although existing evidence regarding the association between DCT adoption and trial duration is currently limited, this may correspond to the evolving maturity of their utilisation. Their growing adoption underscores the importance of comprehensive evaluation, particularly analyses that account for confounding factors such as therapeutic areas, trial size and study design.

In light of these considerations, the present study aimed to examine the association between DCT adoption and CTD in completed metabolic disease trials, where DCTs are frequently employed.¹² By focusing on metabolic diseases—an area characterised by relatively standardised clinical trial designs¹³—this study sought to account for potential confounding factors and to examine differences in trial timelines across decentralised and site-based trials.

Methods

Study Sample

This study analysed industry-initiated clinical trials registered on ClinicalTrials.gov with first patient enrolment between 1 January 2015 and 31 December 2023. To minimise confounding effects from disparate therapeutic areas, the scope was restricted to representative metabolic diseases (diabetes, obesity and dyslipidaemia) and drug intervention studies including the USA as a participating country. Investigator-initiated trials were excluded due to their potential variability in financial and administrative resources. Furthermore, trials focused on non-metabolic conditions (eg, diabetic foot, carcinoma, rare diseases such as Fabry disease), phase 4 studies and trials with an ‘incomplete’ status were excluded to ensure a homogeneous sample and to confirm trial duration. Additionally, trials that could not be cross-referenced with the GlobalData database due to the absence of a unique National Clinical Trial (NCT) ID were excluded to ensure data integrity and the accuracy of decentralisation status classification. While ClinicalTrials.gov primarily categorises studies by the condition or disease rather than therapeutic area, we applied the inclusion criteria systematically to identify relevant metabolic trials. Following the systematic screening of 964 initially identified records, a final sample of 444 industry-sponsored phases 1–3 clinical trials was included in the analysis. Detailed study identification and the selection process, including specific exclusion counts for each stage, are illustrated in the flow diagram (figure 1).

Figure 1. Flow chart of study selection with defined search criteria. This figure illustrates the step-by-step screening process for industry-sponsored phases 1–3 metabolic disease trials (diabetes, obesity and dyslipidaemia) conducted between 2015 and 2023. Data were synthesised from ClinicalTrials.gov and GlobalData. N indicates the number of clinical trials remaining after each exclusion step. DCT, decentralised clinical trial.

Data Sources

Two primary data sources were used for this study. Initially, primary trial metadata were extracted from ClinicalTrials.gov on 4 February 2024. To identify decentralisation status and specific DCT elements, this information was cross-referenced with the GlobalData Clinical Trials Database (Pharma Intelligence Platform,³⁷ GlobalData, London, UK) using unique NCT IDs. GlobalData aggregates clinical trial information from over 250 registries and corporate sources across 189 countries, providing a comprehensive dataset that supplements the limitations of single-registry data. Of the 964 trials identified in the primary triage, data for 956 studies were successfully integrated.

GlobalData employs a systematic methodology to classify trials as ‘decentralised’. Their analysts identify DCT components—such as telemedicine, remote monitoring and electronic Clinical Outcome Assessments (eCOA)—through a combination of automated data mining and manual review of trial protocols, corporate disclosures and regulatory filings.³⁸ This rigorous curation process captures decentralised features that may not be explicitly tagged in primary registries.

Variables

The dependent variables were trial duration metrics, specifically: CTD and the primary completion period-first patient in (PCD–FPI), based on ClinicalTrials.gov records.^{22 24} CTD was defined as the interval from the study start date (FPI) to last patient last visit (study completion date on ClinicalTrials.gov), whereas PCD–FPI was measured from the study start date (FPI) to the primary completion date (PCD). The latter represents the duration required to assess the primary efficacy outcomes (online supplemental figure 1). The independent variables were DCT adoption and its specific elements. For methodological transparency, raw DCT data³⁷ were regrouped or renamed into five functional categories: in-home devices, wearable devices, electronic data capture (EDC), remote trial visit and mobile technology, as documented in the mapping framework in online supplemental table 1.

Control variables—including clinical phase, trial size, demographics and trial characteristics—were selected based on previously reported associations with trial duration.²⁰,^{2224 39} To ensure objective classification, trial size and the number of investigational sites were categorised based on FDA-defined clinical trial phases.⁴² Specifically, the number of investigational sites was divided into three groups: <10, 10–99 and ≥100. These thresholds were informed by FDA clinical phase definitions⁴³ and industry performance benchmarks; fewer than 10 sites typically characterise exploratory studies, whereas 100 or more sites represent the operational complexity of large-scale, multinational pivotal trials.^{22 24} Additionally, the inclusion of older adults was defined as the participation of individuals aged 65 years and older, consistent with the standardised age group classifications provided by ClinicalTrials.gov.

The study period was categorised into three timeframes—pre-COVID-19 (2015–2019), COVID-19 (2020–2021) and post-COVID-19 (2022–2023)—to ensure sufficient statistical power and capture chronological shifts in DCT adoption. Furthermore, conditions were categorised into diabetes, dyslipidaemia, obesity or others. In cases where study titles contained overlapping terms, the final categorisation was determined by the primary endpoint to ensure consistent grouping.

Analysis

Statistical analysis was performed using IBM SPSS Statistics V.29.0 (IBM). Baseline characteristics were compared using independent samples t-tests for continuous variables and χ² tests for categorical variables. No formal adjustments for multiple comparisons were applied to the exploratory analysis of baseline characteristics. Statistical significance was defined as $p\mathrm{}\S amp;lt;\mathrm{}0.05$.

To ensure data integrity and minimise selection bias, we included all industry-initiated trials within the specified timeframe that met our eligibility criteria. Notably, no missing values were encountered during the data extraction process from GlobalData and ClinicalTrials.gov; all variables required for the study—including trial duration, clinical phase and participant demographics—were fully populated for the selected sample of 444 trials.

A global two-way analysis of variance (ANOVA) was initially performed to examine the main and interaction effects of DCT status and clinical phase on trial duration. Subsequently, individual two-way ANOVAs were conducted within the DCT subgroup (n=124) to assess differences in trial duration according to the five functional DCT categories across clinical phases. Furthermore, separate multiple linear regression models were employed for each phase to identify variables associated with trial duration. Potential confounding bias was addressed by adjusting for key variables in the multiple linear regression models. Finally, to ensure the robustness of our findings, a sensitivity analysis was conducted using PCD–FPI as an alternative dependent variable, verifying the consistency of findings related to DCT implementation during the primary efficacy evaluation period.

Results

Baseline characteristics

Among the 444 clinical trials analysed, phase 1 trials accounted for 31.5% (n=140), while phase 2 and phase 3 trials represented 34.9% (n=155) and 33.6% (n=149), respectively. The prevalence of DCTs significantly differed across clinical phases, with the adoption rate being lowest in phase 1 (11.4% of phase 1 trials) and highest in phase 3 (41.6% of phase 3 trials) ($p\mathrm{}\S amp;lt;\mathrm{}0.001$). Temporal analysis revealed a significant variation in DCT adoption across study periods (χ²$=\mathrm{}10.84,\mathrm{}\mathrm{}\mathrm{}p=\mathrm{}0.004$). The adoption rate peaked during the COVID-19 period (2020–2021) at 38.8%, compared with 27.1% in the pre-COVID-19 (2015–2019). However, no significant linear trend was observed over the entire 9-year period (Spearman’s $\rho =\mathrm{}0.007,\mathrm{}\mathrm{}\mathrm{}p=\mathrm{}0.891$), indicating that the observed increase was period-specific rather than sustained annually (online supplemental table 2).

In phases 2 and 3, the prevalence of diabetes was higher in DCTs than in non-DCTs. Furthermore, phase 3 DCTs involved a significantly larger number of investigational sites and a larger study population compared with non-DCTs (online supplemental table 3).

Trial duration

Baseline characteristics and primary outcomes (CTD and PCD–FPI) for phase 1, 2 and 3 trials are presented in tables13, respectively. Overall, while some baseline characteristics varied across phases, no significant differences in trial duration were observed between DCT and non-DCT groups at this descriptive level. When comparing CTD by phase, phase 1 DCTs had a mean duration of 12.5 months (SD=7.9) vs 11.7 months (SD=9.5) for non-DCTs. In phases 2 and 3, DCTs showed numerically shorter durations (phase 2: 17.3 vs 18.3 months; phase 3: 23.4 vs 25.0 months), although these differences did not reach statistical significance. Two-way ANOVA showed that clinical phase was significantly associated with trial duration $\left(F=\mathrm{}27.4,\mathrm{}\mathrm{}\mathrm{}p\S amp;lt;\mathrm{}0.001\right)$, while the main effect of DCT adoption $\left(F=\mathrm{}0.202,\mathrm{}\mathrm{}\mathrm{}p=\mathrm{}0.653\right)$ and its interaction with clinical phase $(F=0.243,p=0.784)$ were not significant (online supplemental table 3).

Table 1. Characteristics and durations of DCTs and non-DCTs in phase 1 metabolic disease trials.

	Phase 1 (n=140, 31.5%*)
	DCT (n=16, 11.4%†)		non-DCT (n=124, 88.6%†)
Condition/disease, n (%)	0.731
DM	6	(37.5%)	60	(48.4%)
Obesity	8	(50.0%)	44	(35.5%)
LD	1	(6.3%)	11	(8.9%)
Others	1	(6.3%)	9	(7.3%)
Sex of participants, n (%)	0.193
Single sex	0	(0.0%)	12	(9.7%)
Both	16	(100.0%)	112	(90.3%)
Older adult, n (%)	0.001**
Not included	1	(6.3%)	63	(50.8%)
Included	15	(93.8%)	61	(49.2%)
Child, n (%)	0.330
Not included	16	(100.0%)	117	(94.4%)
Included	0	(0.0%)	7	(5.6%)
Start period, n (%)	0.022*
Pre-COVID-19	7	(43.8%)	84	(67.7%)
COVID-19	7	(43.8%)	19	(15.3%)
Post-COVID-19	2	(12.5%)	21	(16.9%)
Country type, n (%)	0.595
US only	15	(93.8%)	111	(89.5%)
Multicountry	1	(6.3%)	13	(10.5%)
Trial size, n (%)	0.924
<80	13	(81.3%)	102	(82.3%)
80–299	3	(18.8%)	21	(16.9%)
≥300	0	(0.0%)	1	(0.8%)
No. of IS, n (%)	0.807
<10	15	(93.8%)	118	(95.2%)
10–99	1	(6.3%)	6	(4.8%)
≥100	0	(0.0%)	0	(0.0%)
Outcome, Mean±SD
CTD, months	12.5±7.9		11.7±9.5
P value	0.748
PCD–FPI, months	12.2±7.5		11.4±9.4
P value	0.753

Note: Values are presented as n (%) or mean±SD. Percentages may not sum to 100% due to rounding.

*p<0.05, **p<0.01, ***p<0.001.

Two-way ANOVA results for CTD and PCD–FPI are presented in online supplemental table 3.

^*% of the total sample (N=444); .

^†% within phase 1 (n=140).

ANOVA, analysis of variance; CTD, clinical trial duration; DCTs, decentralised clinical trials; DM, diabetes mellitus; IS, investigational site; LD, lipid disorder; PCD–FPI, primary completion date minus first patient in primary completion period.

Table 3. Characteristics and durations of DCTs and non-DCTs in phase 3 metabolic disease trials.

	Phase 3 (n=149, 33.6%*)
	DCT (n=62, 41.6%†)		non-DCT (n=87, 58.4%†)
Condition/disease, n (%)	<0.001
DM	53	(85.5%)	44	(50.6%)
Obesity	5	(8.1%)	14	(16.1%)
LD	4	(6.5%)	28	(32.2%)
Others	0	(0.0%)	1	(1.1%)
Sex of participants, n (%)	–
Single sex	0	(0.0%)	0	(0.0%)
Both	62	(100.0%)	87	(100.0%)
Older adult, n (%)	0.374
Not included	5	(8.1%)	11	(12.6%)
Included	57	(91.9%)	76	(87.4%)
Child, n (%)	0.369
Not included	54	(87.1%)	71	(81.6%)
Included	8	(12.9%)	16	(18.4%)
Start period, n (%)	0.064
Pre-COVID-19	48	(77.4%)	77	(88.5%)
COVID-19	13	(21.0%)	7	(8.0%)
Post-COVID-19	1	(1.6%)	3	(3.4%)
Country type, n (%)	0.204
US-only	5	(8.1%)	13	(14.9%)
Multicountry	57	(91.9%)	74	(85.1%)
Trial size, n (%)	0.010*
<80	2	(3.2%)	10	(11.5%)
80–299	7	(11.3%)	22	(25.3%)
≥300	53	(85.5%)	55	(63.2%)
No. of IS, n (%)	<0.001
<10	1	(1.6%)	13	(14.9%)
10–99	23	(37.1%)	48	(55.2%)
≥100	38	(61.3%)	26	(29.9%)
Outcome, mean±SD
CTD, mo	23.4±9.8		25.0±14.6
P value	0.449
PCD–FPI, months	20.5±8.0		22.8±14.2
P value	0.258

Note: Values are presented as n (%) or mean±SD. Percentages may not sum to 100% due to rounding.

*p<0.05, **p<0.01, ***p<0.001 (χ2 or t-test).

^*% of the total sample (N=444).

^†% within phase 3 (n=149).

CTD, clinical trial duration; DCTs, decentralised clinical trials; DM, diabetes mellitus; IS, investigational site; LD, lipid disorder; PCD–FPI, primary completion date minus first patient in primary completion period.

Regarding the primary completion period (PCD–FPI), phase 2 DCTs had a mean duration of 15.2 months (SD=9.2), compared with 16.2 months (SD=9.8) for non-DCTs (table 2). Phase 3 DCTs averaged 20.5 months (SD=8.0) vs 22.8 months (SD=14.2) for non-DCTs (table 3). Consistent with the CTD findings, clinical phase was significantly associated with the primary completion period$\left(F=\mathrm{}22.1,\mathrm{}\mathrm{}p\S amp;lt;\mathrm{}0.001\right)$, whereas neither DCT adoption $\left(F=\mathrm{}0.436,\mathrm{}\mathrm{}\mathrm{}p=\mathrm{}0.510\right)$ nor its interaction with clinical phase $(F=0.453,p=0.636)$ was statistically significant (online supplemental table 3).

Table 2. Characteristics and durations of DCTs and non-DCTs in phase 2 metabolic disease trials.

	Phase 2 (n=155, 34.9%*)
	DCT (n=46, 29.7%†)		non-DCT (n=109, 70.3%†)
Condition/disease, n (%)	0.016*
DM	35	(76.1%)	53	(48.6%)
Obesity	3	(6.5%)	21	(19.3%)
LD	7	(15.2%)	32	(29.4%)
Others	1	(2.2%)	3	(2.8%)
Sex of participants, n (%)	0.256
Single sex	0	(0.0%)	3	(2.8%)
Both	46	(100.0%)	106	(97.2%)
Older adult, n (%)	0.604
Not included	5	(10.9%)	9	(8.3%)
Included	41	(89.1%)	100	(91.7%)
Child, n (%)	0.766
Not included	44	(95.7%)	103	(94.5%)
Included	2	(4.3%)	6	(5.5%)
Start period, n (%)	0.284
Pre-COVID-19	32	(69.6%)	73	(67.0%)
COVID-19	13	(28.3%)	26	(23.9%)
Post-COVID-19	1	(2.2%)	10	(9.2%)
Country type, n (%)	0.676
US-only	27	(58.7%)	60	(55.0%)
Multicountry	19	(41.3%)	49	(45.0%)
Trial size, n (%)	0.474
<80	18	(39.1%)	36	(33.0%)
80–299	22	(47.8%)	50	(45.9%)
≥300	6	(13.0%)	23	(21.1%)
No. of IS, n (%)	0.662
<10	16	(34.8%)	34	(31.2%)
10–99	28	(60.9%)	66	(60.6%)
≥100	2	(4.3%)	9	(8.3%)
Outcome, mean±SD
CTD, mo	17.3±10.2		18.3±10.9
P value	0.605
PCD–FPI, months	15.2±9.2		16.2±9.8
P value	0.565

Note: Values are presented as n (%) or mean±SD. Percentages may not sum to 100% due to rounding.

*p<0.05, **p<0.01, ***p<0.001 (χ2 or t-test).

^*% of the total sample (N=444).

^†% within phase 2 (n=155).

CTD, clinical trial duration; DCTs, decentralised clinical trials; DM, diabetes mellitus; IS, investigational site; LD, lipid disorder; PCD–FPI, primary completion date minus first patient in primary completion period.

Type of decentralised elements

Beyond the overall prevalence across phases, the specific composition of decentralised elements also varied. Among the 124 trials identified as DCTs (27.9% of the total 444 trials), 82 (66.1%) used only a single type of decentralised element, while 42 (33.9%) incorporated two or more elements (the annual distribution of these trials across the study period is provided in online supplemental table 2). As detailed in online supplemental table 4, this variation was driven by the higher adoption of in-home devices—particularly blood glucose monitors—and EDC systems in later clinical stages ($p\mathrm{}\S amp;lt;\mathrm{}0.001$). In contrast, the use of multiple decentralised elements was exceptionally rare in phase 1 (2.1%, 3/140), reflecting a more conservative approach in early-phase research. Regarding the specific types of elements used across all 124 DCTs, remote trial visits (13.5%, n=60) were the most common, followed by in-home devices (10.8%, n=48) and wearable devices (10.4%, n=46) (online supplemental table 4). The variation in CTD according to these specific elements is summarised in online supplemental table 5.

To examine differences in trial duration according to these technologies, two-way ANOVAs were performed for each element type across clinical phases. As summarised in online supplemental table 5 (n=124 DCT trials only), no specific decentralised element showed a statistically significant main effect on duration ($p\mathrm{}\S amp;gt;\mathrm{}0.05\mathrm{}$ for all categories). For instance, phase 2 trials using wearable devices (13.8±8.4 months) or in-home devices (13.7±5.2 months) showed numerically shorter durations than those using remote trial visits (21.0±10.1 months), but these differences were not statistically significant after controlling for the clinical phase. For some elements, the small sample size further restricted meaningful comparison, and no significant interactions between decentralised elements and clinical phases were observed.

Association between DCT adoption and CTD

To identify phase-specific variables of trial duration, multiple linear regression analyses were conducted separately for each clinical phase (stratified analysis). The regression models for CTD demonstrated statistical significance across all phases: ($F=\mathrm{}4.50,\mathrm{}\mathrm{}p\S amp;lt;\mathrm{}0.001)\mathrm{}$ for phase 1); ($F=\mathrm{}2.26,\mathrm{}\mathrm{}p=\mathrm{}0.007)\mathrm{}$ for phase 2) and ($F=\mathrm{}5.82,\mathrm{}\mathrm{}p\S amp;lt;\mathrm{}0.001)\mathrm{}$ for phase 3). Consistent with the ANOVA results, variables related to DCTs—including DCT implementation status and specific decentralised elements—were not significantly associated with CTD across any phase ($p\S amp;gt;\mathrm{}0.05$). Among the control variables, the inclusion of older adults was significantly associated with a reduction in CTD, with standardised coefficients ($\mathrm{\beta }$) of $-0.19\mathrm{}(p=\mathrm{}0.02)\mathrm{}$ in phase 2 and $-0.39\mathrm{}(p\S amp;lt;\mathrm{}0.001)$ in phase 3. While the significance levels for this factor varied between the two duration metrics in phase 2, it remained significantly associated with CTD in phase 3 across both CTD and PCD–FPI analyses ($p\S amp;lt;\mathrm{}0.001$). Conversely, a higher number of investigational sites and multicountry trial status were associated with prolonged trial duration (table 4). These phase-specific variables and the lack of a significant relationship regarding DCT status remained consistent in the sensitivity analysis using PCD–FPI (online supplemental table 6), reinforcing the robustness of the primary findings.

Table 4. Multiple linear regression analysis of clinical trial duration across phases 1–3.

	Phase 1		Phase 2		Phase 3
	$\beta$	$p$	$\beta$	$p$	$\beta$	$p$
DCT	−0.01	0.90	0.08	0.45	−0.10	0.36
Multiple decentralised elements	−0.03	0.81	0.02	0.88	−0.07	0.48
Decentralised elements (RTV Ref.)
In-home device	0.02	0.83	−0.14	0.24	0.04	0.74
Wearable device	0.06	0.67	−0.14	0.28	0.07	0.47
EDC	—	—	0.04	0.61	0.02	0.79
Inclusion of older adult	−0.10	0.22	−0.19	0.02*	−0.39	<0.001
Multicountry	0.29	<0.01	0.09	0.31	0.20	<0.01
No. of IS (<10 Ref.)
10–99	0.19	0.03*	0.23	0.04*	0.43	<0.01
≥100	—	—	0.20	0.06	0.64	<0.001
Start period (pre-COVID-19 Ref.)
COVID-19	0.16	0.05*	−0.10	0.24	−0.04	0.60
Post-COVID-19	−0.10	0.21	−0.18	0.03*	−0.13	0.07
Trial size (<80 Ref.)
80–299	0.28	<0.001	−0.18	0.12	0.09	0.44
≥300	0.03	0.69	−0.05	0.70	−0.06	0.67
C/D (DM Ref.)
Obesity	0.18	0.03*	−0.02	0.85	0.15	0.05*
LD	0.11	0.15	0.00	0.99	0.17	0.03*
F	4.50		2.26		5.82
P value	<0.001		0.007		<0.001
R2	0.32		0.20		0.40
Adjusted R2	0.25		0.11		0.33

Separate phase-specific multiple linear regression models were fitted with CTD as the dependent variable. Reference categories were RTV, IS<10, pre-COVID-19, trial size <80 and DM. Variables with sparse observations (other diseases and mobile technology) were excluded from all models. Variables with zero observations within a given phase (eg, EDC and IS ≥100 in phase 1) were excluded from the corresponding phase-specific model. Similar results for PCD–FPI (online supplemental table 5)online supplemental table 5.

*p<0.05; **p<0.01; ***p<0.001.

C/D, condition/disease; CTD, clinical trial duration; DCT, decentralised clinical trial; DM, diabetes mellitus; EDC, electronic data capture; IS, investigational site; LD, lipid disorder; PCD-FPI, primary completion date minus first patient in primary completion period; RTV, remote trial visit; β, standardised regression coefficient.

Discussion

Our findings suggest two possible interpretations. First, the pharmaceutical industry may be in a transitional ‘learning phase’, where the full efficiency gains of DCTs have yet to be realised. Alternatively, the integration of decentralised elements may introduce logistical and technological complexity that counterbalances any time saved during recruitment. In this view, DCT adoption may not primarily serve as a tool for shortening trial timelines but rather as a means to enhance patient diversity and trial accessibility.

The FDA has highlighted the potential of decentralised approaches to bolster trial efficiency while fostering inclusivity. While some previous studies and industry reports have suggested that DCTs may reduce CTD,^{25 26 28 30} the present study found no statistically significant association between DCT status and trial duration. For phases 2 and 3 trials, shorter mean durations were observed for DCTs; however, these differences did not reach statistical significance, and multiple regression analyses likewise did not demonstrate a significant association. Sensitivity analysis using the PCD–FPI yielded consistent findings, with no statistically significant associations observed across all phases.

Notably, the inclusion of older adults was significantly associated with shorter CTD, particularly in phase 2 $\left(\mathrm{\beta }\mathrm{}=\mathrm{}-0.19,\mathrm{}\mathrm{}\mathrm{}p\S amp;lt;\mathrm{}0.05\right)$ and phase 3 ($\mathrm{\beta }\mathrm{}=\mathrm{}-0.39,\mathrm{}\mathrm{}\mathrm{}p\S amp;lt;\mathrm{}0.001$). This finding appears counterintuitive given the general perception that older populations may increase complexity due to comorbidities.²¹ Instead, their inclusion may be associated with shorter trial duration in metabolic trials, possibly reflecting higher disease prevalence or improved adherence. Importantly, this demographic factor did not alter the non-significant association between DCT status and trial duration, reinforcing the robustness of our findings across duration metrics and demographic characteristics. These results are consistent with our interpretation that the industry may currently be in a ‘learning phase’, in which potential time savings coexist with logistical complexities associated with early-stage integration.

In contrast to the non-significant association involving DCT status, our findings from metabolic disease trials indicate that decentralised elements such as remote trial visits, in-home medical devices and wearable technology were more frequently used, whereas EDC systems, including electronic Case Report Form (eCRF), eCOA and electronic Patient-Reported Outcome (ePRO), were less commonly used. The lower use of EDCs aligns with previous findings suggesting that therapeutic areas primarily reliant on physiological measurements (eg, cardiovascular and metabolic diseases) may not require extensive use of eCOA.⁴⁴ No significant differences in trial duration were observed according to the type of decentralised elements employed, and their associations with trial timelines also remained inconclusive.

Our temporal analysis supports this ‘learning phase’ hypothesis. The lack of a significant annual growth trend ($\rho =0.007,\mathrm{}\mathrm{}\mathrm{}p=\mathrm{}0.891$) may reflect the current stage of operational expertise in DCT implementation rather than sustained annual expansion. This may be further influenced by the temporal distribution of our dataset; over 70% of the analysed trials (n=321 out of 444) were initiated between 2015 and 2019. During this period, DCTs were in a nascent stage, lacking the standardised regulatory frameworks and integrated digital infrastructures^{1 12} that emerged more prominently after 2020. This early-stage implementation, where operational expertise was still maturing, may be related to the absence of statistically significant differences in duration.

The relationship between DCT implementation and trial characteristics appears multidimensional, varying by study objectives, therapeutic areas and implementation strategy. The FDA has highlighted the potential of DCTs to promote equitable access and participant accessibility and diversity. Additionally, trial sponsors may adopt DCTs for purposes beyond simply reducing trial duration. In line with recent definitions of trial validity, these decentralised approaches focus on enhancing the ‘applicability’ of results to real-world clinical settings—ensuring that trial outcomes are relevant to a broader patient population—rather than solely focusing on statistical representativeness, consistent with recent methodological frameworks.⁴⁵ While continuous data collection in DCTs may streamline certain processes, it can also increase participant burden and reduce retention rates depending on study procedures, trial content and the specific decentralised elements employed.

Therefore, although DCTs offer potential efficiency gains, these benefits must be carefully weighed against logistical complexity, regulatory hurdles and participant engagement challenges. In light of these factors, our study found no statistically significant association between DCT status and trial duration. These results indicate that as the industry progresses beyond the initial learning phase, sustained implementation and stakeholder training will be critical for detecting measurable differences in efficiency. Concurrently, stakeholders likely need more time to become accustomed to adopting and using DCTs effectively, considering the operational shifts required by their early-stage implementation. This will allow for more refined analyses of these associations as evidence accumulates.

Ideally, assessing the association between DCT adoption and trial duration would involve comparing the same study under different methodological conditions; however, practical limitations make this approach challenging. To address this, our study used multiple regression analysis to control for variables previously reported to be associated with duration. This approach enhanced the precision of our findings while acknowledging and accounting for the limitations inherent in comparing data across different studies.

Some studies suggest that DCTs may, in some cases, prolong trial durations due to technological and operational challenges.^{27 44} The successful implementation of DCTs may require the effective utilisation of new technologies,³⁴,³⁶ along with targeted investments in stakeholder training programmes and robust infrastructure^{23 46 47} development to ensure seamless and scalable adoption.⁴⁴ Therefore, the results of this study suggest that to optimise the implementation of DCT approaches, it is essential to adopt models tailored to specific study characteristics, patient profiles (eg, clinical phase, therapeutic area, intervention type and study objectives), and regulatory conditions. Furthermore, compared with conventional site-based clinical trials, DCTs require more frequent communication and interaction between researchers and participants to build trust, compensating for the lack of face-to-face engagement.^{23 27} Thus, effective implementation of DCTs necessitates dedicated training programmes for sponsors and investigators, incorporating case studies that illustrate best practices and highlight key differences between DCTs and site-based clinical trials.

Limitations and future research

This study acknowledges several limitations. First, a potential time-lag bias should be considered regarding the trials from 2022 to 2023. Since this study only included ‘completed’ trials, those initiated in recent years and still in progress were naturally excluded, explaining the smaller sample size for the most recent period. Second, selection and information biases are possible, as decentralisation status was determined based on secondary database information from ClinicalTrials.gov, which may lack the granularity of direct protocol characteristics. Third, although multiple confounders were adjusted for, fully controlling for differences in study design and operational environments across trials remained challenging. Residual confounding from unmeasured factors, such as sponsor-specific operational strategies, may still have affected the observed associations. Furthermore, as this study did not include patient and public involvement (PPI), there is a limit to understanding participant–side factors—whether facilitating or hindering—that may have been associated with the observed findings.

Future research should incorporate PPI to contextualise quantitative data with qualitative patient perspectives. Additionally, future efforts should aim to refine analytical methodologies and increase dataset granularity through enhanced metadata or integrated sponsor-level information. Examining the association between stakeholder training, regulatory readiness and infrastructure maturity across geographic regions will provide a more comprehensive understanding of the real-world effectiveness and scalability of decentralised clinical trial approaches.

Conclusions

This study represents one of the early empirical evaluations of the association between decentralised clinical trials and CTD using real-world data. Although prior literature has suggested that DCTs may be associated with shorter trial timelines, our analysis found no statistically significant difference in duration between DCTs and conventional site–based trials for metabolic diseases. This non-significant association remained consistent after adjustment for demographic and trial characteristics. These findings may reflect the current stage of DCT implementation, where operational complexities coexist with theoretical expectations of efficiency. Furthermore, although phase 2 and 3 DCTs exhibited numerically shorter mean durations, these differences were not statistically significant, and no statistically significant association was observed after adjustment for demographic and trial-specific characteristics.

Despite current uncertainties, this study provides empirical evidence of DCTs’ evolving role in clinical development and serves as a foundational reference for future research. By using publicly available data, the study enhances transparency and reproducibility, supporting broader efforts to evaluate and validate DCT strategies across different contexts. Continued stakeholder adaptation, methodological rigour in implementation and increased operational experience will be essential to better understand the role of decentralised trials in clinical trial efficiency, accessibility and patient-centredness.

Data availability statement

Data are available on reasonable request.