National implementation of a digital Parkinson’s disease screening programme in Thailand: reach, adoption, and real-world performance of the CheckPD app Authors: Roongroj Bhidayasiri, Saisamorn Phumphid, Jirada Sringean, Chanawat Anan, Teerawat Laosombut, Jeeranun Meesri, Sunaree Wekhinhiran, Natthaphat Leabthong, Onanong Phokaewvarangkul, Decho Surangsrirat, Peerapon Vateekul, Zar Kyi Win, Nurhussneeda Jeh-voh, Pattamon Panyakaew, Warongporn Phuenpathom, Suppata Maytharakcheep, Priya Jagota, Varis Ratanasirisawad, Surasa Khongprasert, Watchara Rattanachaisit, Tittaya Prasertpan, Ornanong Udomsirithamrong, Appasone Phoumindr, Nontakorn Likhitwitayawuid, Thanachporn Saengmanee, Piyaporn Rattanajun, Kornake Horaruengdecha, Narumol Sukmueng, Punkorn Dornbunlon, Werasit Sittitrai, Sukanda Vorachetbancha, Wittaya Chanchalong, Aekamorn Phonsrithong, Pinit Kullavanijaya, Jintakan Sricholwattana, Vittratorn Chirapravati, Pradhana Chariyavilaskul, Katevisuth Sukpisan, Jadej Thammatacharee, Athaporn Limpanyalers Published: 2026 ABSTRACT AbstractBackgroundParkinson’s disease (PD) remains underdiagnosed in Thailand, and its rising prevalence presents a growing challenge for the healthcare system. The previously validated CheckPD digital population screening platform has been implemented nationally in collaboration with the Thai Red Cross Society (TRCS) and the National Health Security Office (NHSO), enabling integration of digital PD risk screening into preventive health frameworks.ObjectiveTo evaluate the early phase of a national rollout of the CheckPD platform, focusing on population reach, adoption, predictive performance, exploratory usability, and implementation factors influencing scalability across diverse real-world settings.MethodsThis RE-AIM-guided implementation study in 10 Thai provinces assessed reach, adoption, completion, system performance and positive predictive value among neurologist-evaluated screen-positive participants. Preliminary usability was assessed in 30 post-screening completers using the SUS and UEQ-S. Supplementary implementation feedback was collected from Village Health Volunteers and public health officers.ResultsBetween January 2024 and October 2025, 13,381 out of 18,520 users completed screening across 10 provinces (completion rate: 72.3%). The mean SUS score was 83, with a 92% first-time task completion rate. Programme reach was achieved through multiple channels, including Village Health Volunteers (6,742 participants), community field campaigns (5,207), facilitated online training initiatives (3,448), and self-initiated app downloads (3,123). When compared with neurologists’ diagnoses among 730 screen-positive participants who underwent evaluation, the screening demonstrated a positive predictive value of 81.23% (593/730; 95% CI 78.39%–84.07%). Key facilitators of implementation included TRCS endorsement and network support, community volunteer engagement, and user-centred app design. Exploratory multivariable logistic regression analysis identified educational attainment and geographic context as significant predictors of screening completion, with higher educational attainment and residence outside Bangkok associated with a higher likelihood of completing the screening workflow.ConclusionsThe CheckPD programme demonstrates that national-scale digital screening for neurological disorders is feasible in a low-to-middle-income country when embedded within trusted institutions, supported by community networks, and aligned with data protection standards. Thailand’s experience provides an early, promising, and potentially scalable model for implementing population-level improvements in brain health by enabling earlier detection and assessment of individuals at risk, in alignment with the World Health Organization’s Brain Health framework.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12889-026-27654-0.Keywords: Parkinson’s disease, Thailand, Digital interventions, Non-communicable diseases, Healthcare policy Background Parkinson’s disease is a significant public health issue Parkinson’s disease (PD) is one of the fastest-growing neurological disorders worldwide and represents an increasing public health challenge in ageing societies [1]. Its global prevalence has increased markedly over the past three decades, driven largely by population ageing and longer disease duration, with additional contributions from environmental risk factors such as exposure to pesticides and other pollutants [2–4]. Projection studies have estimated that by 2050, globally 25.2 million people will be living with PD, representing a 112% increase from 2021, and this rise is expected to be more pronounced among low-to-middle-income countries (LMICs) [5, 6]. In Thailand, epidemiological studies using data from the national Parkinson’s Disease Registry have established that PD is generally under-reported but have estimated an age-adjusted prevalence of 424.57 PD cases/100,000 of the population, with a significantly higher prevalence in rural versus urban areas [7]. PD is associated with progressive motor and non-motor disability, reduced quality of life, and substantial long-term healthcare and societal costs [8, 9]. Despite the availability of effective symptomatic treatments, delays in receiving a definitive diagnosis of PD are common, limiting opportunities for early intervention, risk modification, and timely access to specialist care [10]. These challenges are particularly pronounced in LMICs where healthcare resources and specialist capacity may be lacking, and where disparities in access to these services exist between urban and rural areas [11]. In Thailand, for example, definitive diagnosis of PD is often delayed by 2–3 years, partly due to limited availability of movement disorders specialists [7, 11]. Currently, Thailand has 926 neurologists, however most are located in larger provinces, such as Bangkok, Chiang Mai, and Chonburi, with far fewer in rural areas, resulting in inequitable access to specialist care [12–14]. Prompt diagnosis of PD and referral for specialist care at its early stages is essential to avoid people living with the consequences of debilitating motor and non-motor PD symptoms, to maximise their long-term outcomes, and to minimise the impact on the healthcare system. Earlier identification of individuals at increased risk of PD may facilitate more timely specialist assessment, earlier initiation of symptomatic treatment, and improved care planning. Although direct cost-effectiveness evidence for population-level PD screening remains limited, the burden of PD increases substantially with advancing disease severity, with worse quality of life and markedly higher healthcare and societal costs in more advanced stages [8, 15, 16]. Delays in diagnosis therefore represent a missed opportunity to intervene earlier in the disease course [17, 18]. Prompt diagnosis of PD and referral for specialist care at its early stages are essential to reduce the impact of disabling motor and non-motor symptoms, optimise long-term outcomes, and minimise the overall burden on patients, caregivers, and the healthcare system. Gaps in current PD detection and prevention strategies When PD remains undiagnosed and untreated, progression to advanced stages is associated with profound deterioration in quality of life for patients and caregivers, alongside escalating requirements for long-term supportive and social care. Correspondingly, the economic burden on healthcare systems and society increases substantially, with costs of care for advanced PD far exceeding those associated with early detection and management [8, 15, 16, 19]. The rising demand for timely PD diagnosis and effective treatment poses growing challenges to national health systems, economic sustainability, and workforce capacity. These pressures underscore the urgent need to shift from predominantly reactive care models towards preventive, population-based strategies that facilitate earlier risk identification and intervention at scale [20, 21]. In line with recommendations from the Global Burden of Disease Study and related initiatives, prioritising prevention of non-communicable diseases (NCDs), including PD, alongside the development of clear treatment pathways and the strengthening of health system infrastructure, is essential to mitigate long-term neurological disease burden [22, 23]. Thailand operates a predominantly publicly financed universal health coverage system. Since 2002, the population has been covered through three main public insurance schemes: the Civil Servant Medical Benefit Scheme for government employees and dependents, the Social Security Scheme for formal-sector employees, and the Universal Coverage Scheme for those not covered by the other two schemes, who comprise the majority of the population [11, 24, 25]. Care is delivered mainly through public primary care and district/provincial hospital networks, with contracted private providers playing a larger role in urban settings; under the Universal Coverage Scheme, patients generally register with a local provider network for outpatient, preventive, and referral care, with little or no point-of-care payment. While the timely identification of PD across populations seems like a logical approach to the problem, there are often significant barriers to its practical implementation at a primary and secondary care level within national healthcare systems and infrastructures [26]. In Thailand, the majority of the population lives in remote or rural areas so traditional face-to-face assessment in a clinic is not always feasible and can be resource intensive, plus specialist PD expertise may not be easily available [11]. It is clear that in order to address these challenges, there is a need for simple, accessible, and preventive approaches that can be deployed nationally and equitably across both urban and rural populations. Population-level digital health interventions provide a scalable mechanism to address these implementation barriers by enabling systematic early risk detection while supporting more efficient allocation of specialist and healthcare resources [27]. Digital screening as a public health strategy Digital tools and innovative technologies powered by artificial intelligence (AI) and machine learning are increasingly being adopted across multiple areas of medicine, including PD, to support clinicians in diagnosis, monitoring, and treatment decision-making [28–31]. The use of AI-supported digital tools for population health screening offers substantial advantages over traditional methods in terms of speed, scalability, remote accessibility, and systematic data collection to better quantify disease burden. As a result, these approaches have particular value for LMICs, especially in remote or underserved areas where healthcare resources are constrained [32–34]. Globally, a wide range of digital health screening tools has been evaluated across diverse disease contexts [35]. Examples include the UK’s National Health Service digital portal, which enables citizens to access health services, including screening, and self-management resources [36] and India’s MadhuNETrAI, an AI-driven mobile application developed for national diabetic retinopathy screening [37]. The benefits of digital health tools have also been demonstrated in mental health and cognitive or neurocognitive population screening [38–40], with the World Health Organization (WHO)’s Mental Health Gap Action Programme (e-mhGAP) aiming to leverage such tools to scale services for mental, neurological, and substance use disorders globally. Within Southeast Asia, initiatives in Indonesia, Vietnam, Malaysia, and Thailand illustrate how mobile- and tablet-based platforms are being adopted for screening of chronic diseases and geriatric health conditions, helping to overcome longstanding barriers related to access, workforce limitations, and geographic inequities [41–44]. Thailand already has a well-established example of large-scale, AI-enabled population screening through its national diabetic retinopathy programme. In this programme, a deep-learning system has been prospectively deployed across multiple primary care sites to deliver real-time diabetic retinopathy detection with specialist-level accuracy, demonstrating both feasibility and effectiveness within a middle-income country health system [45]. Thailand’s national diabetic retinopathy programme has also been supported by human-centred implementation research and economic evaluation. The programme is established across primary care units nationwide, and a Thai cost-utility analysis reported that deep-learning-assisted screening achieved similar quality-adjusted life-years compared with trained human graders, while reducing societal costs by approximately US$2.70 per patient and generating positive net monetary benefit [46]. This initiative highlights how AI-supported digital screening can be successfully integrated into routine workflows, particularly in settings with limited specialist availability. Against this backdrop, the implementation of digital PD risk screening platforms, such as CheckPD, represents a logical extension of established national digital screening strategies, applying similar principles of early identification, task-shifting, and scalability to the growing neurological disease burden. Thailand’s digital readiness and constraints Thailand has high levels of internet and smartphone use, making app-based screening a feasible approach for large-scale public health deployment [47, 48]. However, digital participation remains uneven, particularly among some older adults and individuals with lower digital literacy, which may limit equitable uptake of fully self-administered tools [49–51]. For this reason, the CheckPD implementation model incorporated both self-administered and assisted pathways, including support from Village Health Volunteers (VHVs), who are already embedded within Thailand’s primary healthcare system [52, 53] (Table 1). Table 1Statistics on internet use, device access and device use in ThailandIndicatorValueInternet accessInternet users (% of population)91.2% (65.4 million people)Mobile phone subscriptions (% of population)139% (99.5 million numbers)Average daily internet use7 h 54 minInternet use via mobile phones5 h per day (63.3%)Main devices for internet accessSmartphones99.6%Desktops16.7%Laptops10.6%Tablets2.8%Common internet access locationsHome95.9%Work38.3%Private centers27.5%Public centers14.2%Note: Totals exceed 100% as users can use more than one device and more than one access location Statistics on internet use, device access and device use in Thailand Note: Totals exceed 100% as users can use more than one device and more than one access location The CheckPD platform and its benefits as a national screening tool The digital screening platform evaluated in this study, hereafter referred to as CheckPD, is a multimodal, AI-assisted mobile application that has been specifically developed for population-level PD screening in Thailand. The broader CheckPD programme was registered on ClinicalTrials.gov as an observational patient registry (NCT06609681; actual study start 1 January 2024; first submitted 16 September 2024; first posted 24 September 2024). Because public registration occurred after study initiation but during the ongoing implementation period, we report the registration dates explicitly and do not describe the study as prospectively registered. The protocol for national digital screening using CheckPD has been described previously and established the suitability of CheckPD for national rollout as a Royal Commemorative Project in honour of Her Royal Highness the Princess, in collaboration with the Thai Red Cross Society (TRCS) and the National Health Security Office (NHSO) [54]. CheckPD employs AI-driven multimodal assessment methods that integrate a structured questionnaire on PD symptoms and risk factors with digital analyses of tapping, tremor, gait, and voice parameters, and is implemented within a governance framework aligned with Thailand’s Personal Data Protection Act (PDPA). Although many digital tools for PD have been reported, most have been developed for research, symptom monitoring, or digital biomarker validation rather than for end-to-end population screening integrated with health-system workflows (Supplementary Data 1) [55–57]. In contrast, CheckPD was purposefully designed from inception as an accessible, human-centred design-driven, informed digital screening platform capable of operating at a national scale in real-world settings [54]. Rather than focusing on a single symptom domain, the application integrates multimodal assessments and is embedded within Thailand’s healthcare infrastructure through a formal memorandum of understanding with the NHSO. This enables clearly defined referral pathways following risk identification. When screening results indicate an elevated risk of PD, the app automatically directs users to registered hospitals within the national health system for further clinical evaluation. A comparative overview of CheckPD and selected global PD digital platforms, highlighting differences in purpose, evidence base, scalability, and system integration, is presented in Supplementary Data 1. Objectives of this study This study aimed to evaluate the early implementation of a phased national rollout of the CheckPD digital screening programme across 10 provinces in Thailand, focusing on reach, adoption, real-world predictive performance, exploratory usability, and implementation factors relevant to scale-up in a LMIC. Methods Study design and implementation framework This was a pragmatic implementation study with descriptive quantitative analyses and supplementary qualitative implementation feedback, evaluating the early phase of national deployment of a digital PD screening programme in Thailand between January 2024 and October 2025. The study was designed as a pragmatic, real-world evaluation, embedded within routine public health activities, rather than in a controlled experimental setting, in order to reflect actual conditions of use at scale. Evaluation was guided by the RE-AIM framework (Reach, Effectiveness, Adoption, Implementation, and Maintenance), which is widely used to assess the public health impact, usability, and scalability of complex interventions [58–60]. The RE-AIM framework enabled systematic assessment of population reach, predictive performance under real-world conditions, adoption across provinces and delivery pathways, fidelity of implementation, and early indicators of sustainability. National rollout and programme governance were coordinated by the TRCS in collaboration with the NHSO. The NHSO is Thailand’s public agency responsible for ensuring equitable access to essential health services across the continuum of care, including health promotion, disease prevention, treatment, and rehabilitation, primarily through strategic financing and reimbursement of healthcare providers. Early risk identification through population-based screening initiatives such as CheckPD aligns with NHSO’s mandate by enabling more timely and effective treatment, optimising resource allocation, and improving long-term health outcomes. The programme was aligned with Thailand’s preventive health agenda and NCD strategies, facilitating integration with existing community health infrastructure and referral pathways. Study setting Implementation was conducted across 10 provinces selected to represent Thailand’s geographic, demographic, and socioeconomic diversity, including urban, peri-urban, and rural contexts (Fig. 1). Participating provinces included Bangkok, Nonthaburi, Pathum Thani, Samut Prakan, Chonburi, Nakhon Pathom, Phetchaburi, Nakhon Sawan, Chiang Rai, and Sisaket. These provinces span central, northern, and northeastern regions and vary in population density, healthcare access, and digital connectivity. Screening activities were delivered through multiple community-based and institutional settings, including community centres, temples, district hospitals, public health outreach campaigns, and personal mobile devices. This multi-setting approach was intended to maximise inclusion of both digitally experienced users and individuals with limited digital literacy or constrained internet access, particularly older adults and residents of rural communities. Geographic distribution of CheckPD screening participants and neurologist availability across the first 10 provinces included in the national rollout. The number of neurologists per province is shown to provide clinical service context for the participating provinces This figure illustrates the spatial distribution of CheckPD screening participants across Thailand among the first 10 provinces included in the analysis. Provinces are shaded according to the number of screened participants, with darker shading indicating higher participation levels. The accompanying table summarises the absolute number and proportion of participants by province, while a second table indicates the number of neurologists available within each province. Participants and eligibility Adults aged 40 years or older residing in Thailand were eligible to participate in the screening programme, reflecting epidemiological evidence that the incidence of PD, while relatively low in early adulthood, begins to rise from the fourth decade of life and increases progressively with age. In addition, longitudinal studies indicate that prodromal features of PD, including non-motor symptoms such as hyposmia, Rapid Eye Movement (REM) sleep behaviour disorder, and subtle motor changes, can emerge years to decades before clinical diagnosis, often during midlife [54, 61, 62]. Participation was voluntary and required provision of digital informed consent within the mobile application prior to initiation of screening. There were no exclusion criteria based on gender, educational attainment, occupation, or underlying health status, consistent with the population-level screening intent of the programme. Individuals who did not complete the consent process, who withdrew prior to submission of screening data, or who experienced technical failure that prevented capture of evaluable data were excluded from analytic datasets. No financial or material incentives were provided. Implementation and recruitment strategies The CheckPD screening programme was delivered using four coordinated implementation pathways designed to maximise reach while accommodating varying levels of digital literacy and access: (1) Field screening campaigns organised by provincial public health offices and conducted as one-day or multi-day outreach events; (2) Online training–activated screening sessions, in which trained community leaders or health workers facilitated local screening following remote instruction; (3) Community screening led by VHVs, with a focus on older adults and rural communities; and (4) Self-screening via app download, available nationwide throughout the study period (Fig. 2). The self-download pathway was promoted through Thai Red Cross communications, provincial public health announcements, social media posts, QR codes at screen events, and word-of-mouth referrals. Volunteers and public health officers involved in field and community screening received standardised training from the TRCS, delivered through online modules and written manuals. Training covered consent procedures, step-by-step screening workflows, device setup, and basic troubleshooting to ensure consistency, data quality, and participant safety across provinces and implementation settings. Four implementation pathways of the CheckPD Parkinson’s disease (PD) screening programme. a Field screening conducted through community outreach activities with on-site facilitation and referral pathways; (b) online training–facilitated screening sessions enabling coordinated group participation; (c) Village Health Volunteer (VHV)–assisted community screening, particularly supporting older adults and individuals with limited digital literacy; and (d) nationwide self-administered screening via direct app download without supervision. These pathways were designed to accommodate diverse user needs and implementation contexts The CheckPD platform and screening workflow The CheckPD screening workflow comprises four sequential components designed for rapid, scalable population screening (Fig. 3). (1) App access, user registration, and digital informed consent, where participants securely register on the CheckPD mobile application and provide electronic consent prior to screening. (2) Guided digital motor and voice assessments, including finger tapping for bradykinesia, tremor assessment, gait and balance evaluation, and structured speech recording, with in-app instructions to ensure standardised task execution. (3) Structured symptom and risk-factor assessment, consisting of a questionnaire capturing motor symptoms, non-motor symptoms, and established risk factors for PD. (4) Automated data processing and risk classification, in which collected data are analysed using predefined algorithms to generate an individualised PD risk profile. The entire screening process is designed to be completed within approximately 8–10 min, supporting feasibility for large-scale, population-level deployment. Overview of the CheckPD digital screening workflow. The process comprises four sequential components: (1) app access and user registration; (2) digital motor and voice performance assessments, including tapping, tremor, gait, balance, and speech tasks; (3) symptom and risk factor evaluation through structured questionnaires; and (4) automated risk classification using the integrated AI algorithm, with personalised feedback and recommendation for further evaluation where appropriate To enhance usability and accessibility, the application incorporates Thai-language localisation, audio-visual guidance, enlarged interface elements, and a linear, stepwise workflow. Although the screening workflow was presented in a stepwise sequence, participants were allowed to skip individual tasks or discontinue specific components while continuing the assessment, reflecting the pragmatic, user-centred design of the platform for real-world deployment. These design features were intentionally implemented to reduce cognitive load, support users with varying levels of digital literacy, and minimise task-related errors, particularly among older adults and individuals with motor or cognitive challenges. In addition, a step-by-step video tutorial demonstrating how to use the application and complete each screening task is provided to further support user onboarding and correct task execution (Supplementary data 1). Outcomes and evaluation measures Primary outcomes were operationalised according to the RE-AIM framework. Reach was defined as the number of app downloads, initiated screenings, evaluable users, and completed screenings. Effectiveness was defined narrowly as the positive predictive value (PPV) of the CheckPD risk classification among screen-positive participants who subsequently underwent neurological evaluation; sensitivity, specificity, and negative predictive value could not be estimated because screen-negative participants were not routinely clinically verified. Adoption was defined as uptake across participating provinces and delivery pathways. Implementation was defined as fidelity and feasibility of delivery, including full-protocol completion, task-level completion and dropout, and system performance indicators such as crash rate and offline success rate. Usability and user experience were assessed separately as exploratory post-screening acceptability indicators in a small subsample of screening completers. Maintenance was assessed only as an early indicator of sustainability, including continued programme operation over time and institutional integration with the Thai Red Cross Society and National Health Security Office. Usability was assessed using the System Usability Scale (SUS) and the short User Experience Questionnaire (UEQ-S), capturing both pragmatic and hedonic dimensions of user experience [63, 64]. UEQ-S results were calculated according to standard UEQ-S scoring procedures and presented on the original scale from − 3 (most negative evaluation) to + 3 (most positive evaluation). External benchmark categories were not applied because the available UEQ benchmark dataset includes heterogeneous interactive products and may not provide a directly comparable reference for a neurological screening application [65]. Secondary outcomes included AI-based PD risk classification, system performance indicators (including application crash rate and offline success rate), and qualitative feedback from volunteers and public health officers regarding usability, user trust, and integration into existing workflows. Supplementary qualitative implementation feedback was collected from 30 VHVs and 30 public health officers who participated in field or community implementation activities. Participants were selected purposively to capture perspectives from both community volunteer and public health implementation roles across participating sites. Semi-structured interviews were conducted in Thai after implementation activities using a brief topic guide covering perceived ease of use, user trust, barriers to completion, workflow integration, referral processes, and suggestions for improvement. Responses were documented as structured field notes rather than verbatim transcripts. Analysis used directed content analysis informed by the RE-AIM framework and the study’s implementation objectives. Two researchers independently reviewed the field notes, coded responses into categories relating to facilitators, barriers, workflow integration, and improvement needs, and resolved discrepancies through discussion (JM and SW). Because these data were collected as pragmatic implementation feedback rather than as a standalone in-depth qualitative study, formal thematic saturation and theory generation were not sought. Statistical analysis and data visualisation Quantitative analyses were conducted to characterise programme reach, user engagement, and task completion patterns during real-world implementation of the CheckPD platform. Descriptive statistics were used throughout, consistent with the pragmatic and implementation-focused aims of the study. Continuous variables are reported as means with standard deviations, and categorical variables as counts and percentages. To examine age-related patterns of engagement, participants were stratified into five predefined age groups: <40, 40–49, 50–59, 60–69, and ≥ 70 years. For each age group, task outcomes were categorised as complete, incomplete, or unattempted. The number and proportion of participants in each category were calculated such that percentages within each age group totalled 100%. These analyses were used to assess age-related differences in task adherence and dropout across the screening workflow. Task-level engagement was further analysed by calculating completion rates for each individual screening task within the CheckPD application. Each task was coded as a binary outcome, with completion coded as ‘1’ and non-completion coded as ‘0’. For bilateral tasks (e.g. dual finger tapping and pinch-to-size), a task was considered complete only if both left- and right-sided components were successfully completed. Task completion rates were expressed as the percentage of participants who completed each task. Tasks were ranked from highest to lowest completion rate and visualised using funnel plots to illustrate cumulative attrition and identify stages of the screening workflow associated with the greatest participant dropout. These analyses were intended to highlight usability- and implementation-related barriers, particularly for tasks requiring greater sensor precision, balance stability, or environmental setup. In addition, multivariable logistic regression analysis was performed to explore factors associated with completion of the full screening protocol. The dependent variable was task completion status (completed versus not completed). Independent variables included selected sociodemographic and contextual factors, including educational attainment and province of residence. Geographic classification was defined as residence in Bangkok versus outside Bangkok. This categorisation was selected because the present study represents the early implementation phase across the first 10 participating provinces, and was intended to capture differences in implementation context and delivery pathways rather than to formally distinguish rural–urban populations. The study was not powered or designed to formally assess urban–rural differences, as the primary aim was to evaluate implementation across the first wave of participating provinces. Results are presented as odds ratios with 95% confidence intervals. Model fit was assessed using the Omnibus test of model coefficients, Nagelkerke pseudo R², and the Hosmer–Lemeshow goodness-of-fit test. Descriptive analyses and data visualisation were conducted using Python version 3.12.10, with pandas for data handling and matplotlib for visualisation. The exploratory multivariable logistic regression analysis was conducted using IBM SPSS Statistics version 23.0 (IBM Corp., Chicago, IL, USA). A two-tailed p-value < 0.05 was considered statistically significant. No hypothesis-driven inferential statistical testing beyond this exploratory regression analysis was undertaken, as the primary objective of the study was to evaluate real-world feasibility, engagement, and implementation performance, rather than to establish causal relationships. Analyses of AI predictive performance, including calculation of PPV based on neurologist verification, were conducted separately with 95% confidence interval (CI) and are described in the following section. All subgroup comparisons presented outside the logistic regression model are descriptive and should not be interpreted as formal hypothesis-testing results. AI evaluation and clinical verification CheckPD classifies users as ‘normal’ or ‘at risk’ using an ensemble machine-learning model trained on multimodal input data derived from questionnaire responses and sensor-based motor and voice tasks, with the method described previously [54]. Participants classified as at risk were advised to seek further clinical evaluation through standard healthcare pathways. AI performance metrics were calculated only amongst participants who were flagged as at risk and subsequently underwent neurological assessment, considered as the gold standard for diagnosis of PD. Participants classified as screen-negative were not routinely clinically verified as part of this implementation programme; therefore, population-level diagnostic accuracy could not be estimated. Usability and user experience assessment Usability and user experience were assessed descriptively in a non-probability, post-screening subsample of participants who had completed the full CheckPD workflow. A total of 30 participants from multiple implementation pathways were selected and invited to complete the usability assessment, and all invited participants responded (response rate 100%). This subsample was intended to provide preliminary u sability and acceptability signals among screening completers rather than a statistically powered or population-representative estimate of usability across the full implementation cohort. Validated Thai versions of the SUS and the UEQ-S were administered. UEQ-S results were calculated according to standard UEQ-S scoring procedures and presented descriptively on the original − 3 to + 3 scale. Participants who did not complete screening were not eligible for this usability substudy, and their usability barriers could not be directly assessed. Data sources, management, and governance Data for this implementation study were derived from the CheckPD digital screening platform and its associated clinical screening and analytics infrastructure, as illustrated in the system architecture and data flow diagram (Fig. 4). The CheckPD ecosystem comprises three core components: (1) the CheckPD mobile application for population-level data capture, (2) a web-based screening module for clinical and questionnaire-based assessments, and (3) a Business Intelligence (BI) dashboard for monitoring programme performance and implementation outcomes. System architecture and data flow of the CheckPD platform. This diagram illustrates the integrated data architecture of the CheckPD ecosystem, comprising the CheckPD mobile application, a web-based clinical screening module, and a Business Intelligence (BI) dashboard. Multimodal screening data collected via the mobile application are stored in a secure NoSQL database operated by the Thai Red Cross Society, with selected structured data migrated to a SQL Data Warehouse managed by ChulaPD for analysis and reporting. Clinical diagnostic labels assigned by authorised neurologists are stored exclusively within the SQL environment. Aggregated data from both components are visualised through the BI dashboard to support programme monitoring, implementation evaluation, and governance The CheckPD mobile application serves as the primary source of multimodal screening data. During routine use, participants generate real-time data including demographic information, questionnaire responses, motor assessments (finger tapping, tremor, gait, and balance), and speech recordings. Structured application data are stored as JavaScript Object Notation (JSON) files, while speech recordings are stored as waveform audio (WAV) files. All raw application data are housed within a NoSQL database operated by the TRCS, which functions as the secure primary data repository. To ensure data integrity and continuity, daily backups are performed through a dedicated web application, with archived copies stored on encrypted cloud storage and local hard disk drives. However, no structured data were collected from users who exited prior to generating evaluable screening records; therefore, early-stage attrition could not be quantitatively analysed. For secondary use, analysis, and reporting, selected structured data, primarily demographic and screening-related variables, are transferred from the TRCS NoSQL database to a structured SQL Data Warehouse managed internally by a team at Chulalongkorn Centre of Excellence for Parkinson’s Disease and Related Disorders (ChulaPD; www.chulapd.org). Data migration occurs on a daily and weekly schedule, depending on data type and operational requirements. An automated migration pipeline transfers JSON and WAV files using three unique identifiers: a general system identifier, the Thai national identification number, and a record-specific identifier. This automated process includes built-in error detection, with notifications issued to the ChulaPD data team in the event of transfer failures or inconsistencies. Only records with neurologist-confirmed diagnostic labels (PD or control) are included in the analytic migration workflow. Clinical diagnostic data are generated through a dedicated physician-facing web application. Authorised neurologists review participant-level screening outputs and assign diagnostic classifications, categorised as ‘PD’ or ‘not PD’. Diagnostic access and labelling privileges are restricted to approved personnel, and all labelled records are stored exclusively within the SQL Data Warehouse. This separation of raw data storage and curated diagnostic datasets supports data minimisation, traceability, and role-based access control. Programme monitoring and implementation evaluation are supported by the BI dashboard, which integrates data from both the CheckPD and Screening components. Using visual analytics tools, including Looker Studio, the dashboard provides near real-time summaries of application downloads, screening completion rates, diagnostic distributions, task-level adherence, and system usage patterns. This infrastructure enables continuous monitoring of reach, adoption, implementation fidelity, and system performance throughout the national rollout. Data governance across the CheckPD platform is designed to comply with Thailand’s PDPA. Governance mechanisms include role-based access control, separation of identifiable and analytic datasets, controlled data migration, and secure storage managed by national and academic partners. Collectively, this architecture supports secure, scalable, and auditable use of digital health data for population-level PD screening and implementation research. Ethical considerations and data protection The TRCS acted as the designated data controller, with ChulaPD serving as the data processor. Prior to initiating any screening or assessment, all participants were required to provide explicit approval for data collection through a digital informed consent process embedded within the CheckPD platform, in accordance with approvals granted by the relevant ethics and regulatory bodies. All personal and health-related data were protected through end-to-end encryption, employing Advanced Encryption Standard 256-bit (AES-256) encryption for data at rest and Transport Layer Security (TLS) version 1.3 for data in transit. Data were stored on an ISO/IEC 27,001-certified server infrastructure operated by the TRCS. Analytical activities were conducted exclusively on de-identified datasets under approval from the relevant ethics committees. Ethics approval was obtained from the Institutional Review Board of the Faculty of Medicine, Chulalongkorn University (IRB No. 0804/65), as well as from the Central Research Ethics Committee (CREC) under the approved protocol entitled ‘Flipping the Paradigm of Parkinson’s Disease: A Model of National ‘Eat, Move, Sleep’ Digital Interventions to Prevent or Slow the Rise of Non-Communicable Diseases in Thailand’ (CREC No. CREC023/68BR-MED06; Certificate Number COA-CREC105/2025). In accordance with Thailand’s PDPA (B.E. 2562, 2019), Data Protection Impact Assessments were performed on an annual basis to evaluate and mitigate potential risks related to data processing and security. Reporting standards and implementation framework alignment This study was designed and reported in alignment with principles from the RE-AIM framework for implementation research, with explicit reporting of reach, adoption, implementation fidelity, and early indicators of maintenance. Elements of the STROBE guidelines for observational studies were followed where applicable, including transparent reporting of study setting, participant eligibility, data sources, and outcome measures [66]. Given the pragmatic, implementation-focused nature of the programme, no randomisation or hypothesis-driven comparisons were undertaken, and reporting emphasised real-world feasibility, usability, and scalability rather than causal inference. Results Participant demographics and CheckPD app usage Between January 2024 and October 2025, the CheckPD platform recorded 30,327 application downloads nationwide. Of these, 21,882 users initiated screening and generated evaluable data, and 18,520 users from the first 10 provinces met the criteria for inclusion in the present analysis based on completed implementation governance and stable data pipelines (Fig. 5). Therefore, all completion, demographic, pathway-specific, and regression analyses in this manuscript are based on the 18,520 users from the first 10 provinces, unless otherwise specified. Flow of user engagement, screening completion, risk classification, and clinical follow-up in the CheckPD programme. This flowchart shows progression from app download to screening initiation, completion, AI risk classification, and neurological follow-up among participants in the 10 included provinces Participants in the analytic cohort had a mean age of 56.91 ± 11.89 years and were predominantly female, comprising 14,004 women (76.0%) and 4,460 men (24.0%), with sex unspecified for 56 participants (< 1%) (Table 2). Several lifestyle and risk-factor fields showed high levels of missingness because these items were not required for completion of the core screening workflow. These variables are therefore presented descriptively and were not used as primary implementation indicators. Overall, 13,381 participants (72.3%) completed all screening tasks, whereas 5,139 (27.7%) had incomplete assessments. Because this data arises from province- and pathway-specific rollout activities, it represents an implementation sample rather than a representative population of all Thai adults at risk for PD. Table 2Demographic characteristics of CheckPD users from the first 10 provinces included in the national screening rollout (n = 18,520)Demographic dataFrequency (%)Age, mean ± SD56.91 ± 11.89Gender Male4,460 (24%) Female14,004 (76%) Not specified56 (< 1%)Education Bachelor’s degree or higher2,681 (15%) Below Bachelor’s degree12,679 (68%) Not specified3,160 (17%)Province Nakhon Pathom5,135 (28%) Phetchaburi576 (3%) Sisaket1,051 (6%) Nakhon Sawan5,054 (27%) Chiang Rai3,118 (17%) Nonthaburi825 (4%) Pathum Thani893 (5%) Samut Prakan458 (2%) Bangkok1,319 (7%) Chon Buri91 (< 1%)Occupation Civil servant (non-teaching)481 (3%) Farmer5,502 (30%) Merchant / Trader2,795 (15%) Petroleum / oil exposure occupation9 (< 1%) Private sector employee503 (3%) Professional athlete1 (< 1%) Retired / Not employed2,203 (12%) State enterprise employee121 (1%) Teacher / Lecturer56 (< 1%) Other2,981 (16%) Not specified3,868 (20%)Marital status Single3,081 (17%) Married / Partnered9,687 (52%) Divorced848 (5%) Widowed1,720 (9%) Not specified3,184 (17%)Income No income759 (4%) Less than 5,000 THB3,999 (22%) 5,000–10,000 THB5,403 (29%) 10,001–20,000 THB1,820 (10%) 20,001–30,000 THB562 (3%) 30,001–40,000 THB300 (2%) More than 40,000 THB610 (3%) Not specified5,067 (27%)History of smoking Yes121 (1%) No1,853 (10%) Not specified16,546 (89%)Alcohol drinking Yes428 (2%) No1,544 (8%) Not specified16,548 (89%)Coffee drinking Yes1,215 (7%) No762 (4%) Not specified16,543 (89%)Milk/dairy milk product consumption Yes747 (4%) No1,230 (7%) Not specified16,543 (89%)Regular exercise (more than 150 min/week) Yes1,086 (6%) No891 (5%) Not specified16,543 (89%)Insecticide exposure Yes288 (2%) No1,689 (9%) Not specified16,543 (89%)History of narcotic use Yes41 (< 1%) No1,936 (11%) Not specified16,543 (89%)History of severe head injury Yes70 (< 1%) No1,906 (10%) Not specified16,544 (89%)Note: THB, Thai Baht. Smoking, alcohol intake, coffee intake, dairy intake, exercise, insecticide exposure, narcotic use, and severe head injury were optional; percentages of “not specified” therefore reflect implementation-stage data capture rather than true population absence of these risk factors Demographic characteristics of CheckPD users from the first 10 provinces included in the national screening rollout (n = 18,520) Note: THB, Thai Baht. Smoking, alcohol intake, coffee intake, dairy intake, exercise, insecticide exposure, narcotic use, and severe head injury were optional; percentages of “not specified” therefore reflect implementation-stage data capture rather than true population absence of these risk factors Amongst the 13,381 participants who completed screening, 1,761 (13.2%) were classified by the CheckPD algorithm as at risk and 11,620 (86.8%) as normal (Fig. 5). Of those classified as at risk, 730 underwent in-person neurological evaluation by the time of analysis. Among this evaluated subgroup, 593 were clinically diagnosed with PD and 137 were not diagnosed with PD, corresponding to a PPV for those classified as ‘at risk’ of 81.23% (95% CI:78.39%-84.07%) (Supplementary Data 3). The remaining 1,031 participants with an at-risk screening result had no documented neurological assessment at the time of analysis. To better understand factors associated with screening completion, we compared participants who completed all screening tasks with those who had incomplete assessments (Supplementary Data 4). Participants who completed screening were younger, had higher educational attainment, and showed differences in geographic distribution and occupation compared with those who did not complete the assessment. In particular, participants with a Bachelor’s degree or higher were more likely to complete screening, while participants residing in Bangkok were less likely to complete the full workflow. CheckPD app performance Using the RE-AIM framework, the programme demonstrated substantial reach, multi-pathway adoption across 10 pilot provinces, moderate-to-high implementation fidelity with 72% full-protocol completion, favourable preliminary usability signals among screening completers, and early signs of institutional maintenance through continuing operation and integration with national partners. Reach: Overall programme reach, defined as the proportion of screened individuals relative to app downloads, was 72.15%, indicating high population engagement during this early phase of national implementation. Effectiveness: The CheckPD app showed a positive predictive value of 81.23% based on those 730 users who were considered ‘at-risk’ according to the app’s AI prediction during screening versus the subsequent neurologist’s diagnosis. Adoption: Most CheckPD implementation activities occurred in 2025, following a preparatory period in 2024 dedicated to programme planning, system development, stakeholder engagement, and team readiness. Adoption of the CheckPD platform increased primarily during organised implementation periods, with screening activity remaining low outside these intervals (Fig. 6). Distinct spikes in participation were observed following online training sessions and field-based screening activities across provinces, indicating that these facilitated activities directly contributed to increased app uptake and screening completion. The highest levels of screening activity were recorded between May and September 2025, during which weekly screening volumes exceeded 1,000 participants during peak periods (Fig. 6). Temporal trends in daily CheckPD screening activity in relation to implementation strategies. This figure displays daily numbers of completed CheckPD screenings over time, overlaid with the timing of key implementation activities across provinces. Coloured markers indicate the initiation of different implementation strategies, including online training sessions (red), field screening events (blue), and Village Health Volunteer (VHV) led screening activities (grey), while the black line represents total daily screening volume. Distinct peaks in screening activity correspond closely with organised implementation events, particularly during periods of intensive outreach between May and September 2025, demonstrating the direct impact of facilitated implementation strategies on user uptake and screening participation intensive outreach between May and September 2025, demonstrating the direct impact of facilitated implementation strategies on user uptake and screening participation. Geographic distribution of screening participation varied across provinces (Fig. 1). The highest levels of participation were observed in Nakhon Pathom, while Chonburi recorded the lowest uptake. Provinces with more than 500 screened participants were predominantly located in the central, northern, and northeastern regions of Thailand, reflecting regional differences in adoption associated with local implementation intensity and engagement strategies (Fig. 1). Implementation: Levels of participation and test completion within the CheckPD programme varied across provinces and by implementation method (Fig. 7). Across all settings, 13,381 of 18,520 participants (72.3%) completed the full screening protocol, while 5,139 (27.7%) had incomplete assessments. Volunteer-facilitated and organised implementation strategies accounted for the majority of engagement within the 18,520-person analytic cohort: 6,742 participants were screened through VHV-assisted pathways, 5,207 through community-based field screening campaigns, 3,448 through online training–activated screening sessions, and 3,123 through self-initiated app download. These pathway counts refer to all participants in the 10-province analytic cohort, irrespective of whether they completed the full screening workflow. Completion rates were consistently higher in structured, facilitated settings, particularly those involving VHVs and online training, compared with self-download pathways in several provinces (Fig. 7, Supplementary Data 5). Participation and screening completion outcomes of the CheckPD programme by province and implementation method among users from the first 10 provinces Table 3Distribution of CheckPD screening algorithm risk classifications amongst participants with completed assessments, by province and implementation methodProvinceImplementation methodParticipants withcompleted testsAI prediction resultAt riskNormalNakhon PathomField Screening593 (100%)104 (18%)489 (82%)Online Training1,319 (100%)95 (7%)1,224 (93%)VHVs1,663 (100%)141 (8%)1,522 (92%)Self-download296 (100%)15 (5%)281 (95%)PhetchaburiField Screening269 (100%)30 (11%)239 (89%)Online Training49 (100%)4 (8%)45 (92%)Self-download35 (100%)8 (23%)27 (77%)SisaketField Screening384 (100%)51 (13%)333 (87%)Online Training1 (100%)0 (0%)1 (100%)Self-download251 (100%)25 (10%)226 (90%)Nakhon SawanField Screening919 (100%)142 (15%)777 (85%)Online Training670 (100%)51 (8%)619 (92%)VHVs1,896 (100%)131 (7%)1,765 (93%)Self-download281 (100%)17 (6%)264 (94%)Chiang RaiField Screening471 (100%)90 (19%)381 (81%)Online Training98 (100%)7 (7%)91 (93%)VHVs1,571 (100%)230 (15%)1,341 (85%)Self-download33 (100%)11 (33%)22 (67%)NonthaburiField Screening299 (100%)40 (13%)259 (87%)Online Training69 (100%)7 (10%)62 (90%)Self-download206 (100%)49 (24%)157 (76%)Pathum ThaniField Screening244 (100%)32 (13%)212 (87%)Online Training206 (100%)11 (5%)195 (95%)Self-download201 (100%)31 (15%)170 (85%)Samut PrakanField Screening303 (100%)125 (41%)178 (59%)Self-download66 (100%)23 (35%)43 (65%)BangkokSelf-download753 (100%)161 (21%)592 (79%)ChulaPD Clinic170 (100%)116 (68%)54 (32%)ChonburiField Screening0 (0%)0 (0%)0 (0%)Self-download65 (100%)14 (22%)51 (78%)Total13,381 (100%)1,761 (13%)11,620 (87%)Note: VHVs, Village Health Volunteers. “ChulaPD Clinic” refers to clinic-facilitated screening conducted through the ChulaPD clinic using the online training–activated workflow and is included within the online training category in aggregate implementation-pathway counts Distribution of CheckPD screening algorithm risk classifications amongst participants with completed assessments, by province and implementation method Note: VHVs, Village Health Volunteers. “ChulaPD Clinic” refers to clinic-facilitated screening conducted through the ChulaPD clinic using the online training–activated workflow and is included within the online training category in aggregate implementation-pathway counts Task completion declined progressively across older age groups (Fig. 8). At the task level, questionnaire and voice tasks showed the highest completion rates, whereas gait/walk, balance, and pinch-to-size showed the lowest completion rates (Fig. 9). Funnel plot analyses further highlighted task-specific differences in completion, with questionnaire-based and voice assessments achieving the highest completion rates, while motor tasks requiring greater physical stability, precise sensor positioning, or environmental setup, such as gait, balance, and certain tremor assessments, showed lower completion rates (Fig. 9). Age-stratified completion rates of the CheckPD app screening tasks. This figure illustrates screening task completion rates across predefined age groups among users from the first 10 provinces included in the analysis. Completion rates declined progressively with increasing age, indicating age-related differences in engagement and ability to complete all components of the digital screening workflow Task-specific completion rates in the CheckPD screening workflow. This funnel plot shows the proportion of participants completing each screening task, with higher completion for questionnaire and voice-based assessments and lower completion for more complex motor tasks, reflecting increasing task burden in real-world, unsupervised use Multivariable logistic regression analysis identified educational attainment and geographic context as significant independent predictors of screening completion (Table 4). Participants with a Bachelor’s degree or higher were more likely to complete all screening tasks compared to those with lower educational attainment (odds ratio [OR] = 1.528, 95% CI 1.380–1.691). In contrast, residence in Bangkok was associated with a lower likelihood of completion (OR = 0.755, 95% CI 0.662–0.860). This geographic grouping reflects implementation-phase site categorisation rather than a strict rural–urban classification. Table 4Logistic regression analysis of user parameters that influenced the likelihood of completing the CheckPD app screening testsPredictorsExp (B)p-valueEducational status – Bachelor’s degree1.528 (95%CI 1.380–1.691)< 0.001*Residence in Bangkok0.755 (95%CI 0.662–0.860)< 0.001*Model summaryOmnibus tests of Model coefficients: Chi-square17.393< 0.001*Nagelkerke R Square0.006Hosmer and Lemeshow Test0.365Note: * is statistically significant. The pseudo R-square from the Nagelkerke R-square model determined the percentage of variance. Nagelkerke R-squared – an approximate measure of the proportion of explained variation. Hosmer and Lemeshow Test – the goodness of fit in the logistic regression model Logistic regression analysis of user parameters that influenced the likelihood of completing the CheckPD app screening tests Note: * is statistically significant. The pseudo R-square from the Nagelkerke R-square model determined the percentage of variance. Nagelkerke R-squared – an approximate measure of the proportion of explained variation. Hosmer and Lemeshow Test – the goodness of fit in the logistic regression model Usability outcomes In the exploratory post-screening usability substudy, 30 screening completers completed the SUS and UEQ-S assessments. The mean (± SD) SUS score was 83 ± 9, indicating favourable perceived usability among this selected subgroup. Among first-time users aged 40–70 years in this substudy, the in-app task completion rate was 92%, and the median total test time was 8 min. These findings should be interpreted as preliminary usability signals among workflow completers rather than representative estimates for all CheckPD users. Qualitative feedback indicated that participants found the CheckPD application easy to understand and trustworthy, particularly due to the presence of Thai Red Cross branding. User experience (UX) For the same 30 respondents, the mean (± SD) UEQ-S scores were 1.8 ± 1.0 for efficiency and 1.7 ± 1.2 for interestingness. The mean pragmatic quality score was 1.6 ± 1.0 and the mean hedonic quality score was 1.5 ± 1.1, yielding an overall mean user experience score of 1.6 ± 0.9. All mean UEQ-S scale scores were positive on the standard − 3 to + 3 scale, indicating favourable user experience among respondents in the usability substudy (Fig. 10). No external benchmark category was assigned because no directly comparable benchmark dataset of neurological screening tools was available. UEQ-S pragmatic, hedonic, and overall scores for CheckPD on the standard UEQ-S scale. Scores are presented on the standard range from − 3 (most negative evaluation) to + 3 (most positive evaluation). The marker shows the observed mean score for CheckPD, and the error bars indicate its confidence interval. The horizontal zero line represents a neutral evaluation Qualitative feedback from participants and implementation stakeholders provided additional context to these findings, highlighting ease of use, perceived trust in the platform, and the value of assisted screening pathways (e.g. VHV support) as facilitators of engagement in real-world settings. Content analysis of implementation feedback identified four recurring implementation categories. First, VHVs and public health officers described Thai Red Cross and local public-health endorsement as important contributors to user trust and willingness to participate. Second, assisted screening pathways were perceived as important for supporting older adults and individuals with limited digital confidence, particularly during registration, task instruction, and troubleshooting. Third, implementers noted that motor tasks requiring balance, walking space, or precise phone positioning were more likely to require assistance than questionnaire or voice tasks. Fourth, public health officers highlighted referral coordination after an at-risk result as an important determinant of downstream programme impact. These categories were consistent with quantitative findings showing higher engagement during facilitated implementation activities and lower completion for more complex motor tasks. System performance Results for CheckPD app stability and response time confirmed that it is suitable for field operations. System logs confirmed a crash rate of < 1% and an offline success rate of 93%, enabling reliable operation in remote settings. Volunteer throughput was ~ 12 participants/hour. Discussion This implementation study demonstrates that a digital PD screening programme can be deployed at scale within a national healthcare context, achieving substantial reach, high user engagement, and favourable preliminary usability signals across diverse real-world settings. Beyond prior hospital-based validation studies, which established the diagnostic performance of the CheckPD algorithm under controlled conditions [54], the present findings provide important evidence on how such a tool performs during routine implementation. In particular, the results highlight key operational factors that influence success at scale, including user engagement with different task types, variability in completion across demographic groups, and the need for effective linkage between digital screening and clinical follow-up. Overall findings for primary and secondary outcome measures This real-world implementation study indicates that the CheckPD app is a feasible national PD screening platform for Thailand. Across RE-AIM dimensions, the programme demonstrated substantial reach and adoption, together with high levels of task engagement and completion. Furthermore, the AI-driven app achieved a high positive predictive value (PPV) of 81.23% (95% CI: 78.39%–84.07%) when compared with neurologists’ diagnoses. However, this study does not attempt to estimate the overall diagnostic accuracy of CheckPD across the entire screened population. Rather, it provides a real-world PPV estimate specifically among participants who completed screening, were classified as at-risk, and subsequently underwent neurological evaluation. Therefore, while sensitivity, specificity, and negative predictive value cannot be determined from the current dataset, these findings robustly support the programme’s ability to enrich for likely PD cases during clinical follow-up. A key implementation challenge identified in this study is the gap between risk identification and clinical follow-up. Although 1,761 participants were classified as at risk, only 730 (41%) underwent neurological evaluation at the time of analysis. This incomplete linkage to care represents a critical limitation of real-world effectiveness, as the benefits of early detection depend on timely diagnostic confirmation and subsequent management. The observed follow-up gap likely reflects practical constraints such as access to specialist services, participant readiness, and variability in referral pathways. Addressing this implementation gap will be essential for maximising the clinical and public health impact of digital screening programmes at scale. The observed reduction in PPV in this study compared with the original hospital-based validation study’s reported diagnostic accuracy of 91% [54] confirms a broader and well-recognised phenomenon in AI deployment, a so-called ‘translation gap’ or ‘deployment gap’. The lower performance observed in routine implementation compared with earlier validation is consistent with the broader challenge of translating AI tools from controlled development settings into real-world use [67, 68]. In addition, machine-learning research has identified ‘under-specification’, the tendency for equally good models on training data to behave very differently once deployed outside the development domain, as a fundamental reason for unpredictability in real-world use [69]. These observations strongly support the need for adaptive validation and continuous post-deployment monitoring when scaling diagnostic AI tools [70]. In the present programme, participants used different devices and completed tasks in variable community settings, often without direct supervision and with incomplete task data. These conditions are more representative of operational deployment than hospital-based validation. Despite these challenges, the algorithm retained useful case-enrichment performance in real-world use, supporting continued clinical utility while underscoring the need for ongoing monitoring during scale-up [54]. The performance of CheckPD during national implementation was influenced by several real-world factors that differed substantially from the controlled environment of the original validation study. Unlike structured testing performed under supervision, screening in community settings relies on participants’ ability to follow instructions independently, which introduces variability related to digital literacy and user skill, a known barrier to effective engagement with digital health technologies and outcomes across diverse populations [71, 72]. Device heterogeneity, including differences in smartphone hardware, operating systems, sensor quality, and environmental conditions, has been shown to affect sensor data quality collected in real-world settings, potentially introducing inconsistencies in health-related assessments derived from consumer-owned devices [73]. In addition, incomplete task completion, network connectivity issues, and variable levels of user engagement can lead to missing data and reduced evaluable datasets, reflecting known challenges in remote and unsupervised digital data collection [74]. In this study, follow-up was more commonly achieved in facilitated implementation pathways, where referral processes were actively coordinated, compared with self-directed screening. These implementation realities likely contribute to deviations from controlled-study performance metrics, but they also reflect the authentic operational environment in which CheckPD is intended to function. Additional comparative analyses between participants with complete and incomplete screening highlight important implementation considerations. Lower completion among older individuals and those with lower educational attainment likely reflects differences in digital health literacy and ability to engage with multi-step app-based workflows. The observed lower completion among Bangkok residents, compared with participants outside Bangkok, may reflect differences in implementation context, including lower exposure to facilitated screening pathways (e.g., community or VHV-supported activities) and greater reliance on self-directed use. These findings reinforce the importance of assisted screening models and tailored engagement strategies to improve completion rates across diverse populations. Strengths of CheckPD and implementation success factors Key facilitators of the positive results obtained with CheckPD in this study included TRCS endorsement and network, community volunteer involvement, and its user-friendly design. While other PD mobile applications have been developed primarily for symptom tracking and patient support (Supplementary Data 1), there is limited evidence regarding digital tools designed for national population-level screening. To our knowledge, CheckPD represents the first national, app-based population screening programme for PD, implemented within a universal healthcare coverage system. CheckPD differs from existing digital tools for PD in both its scope and means of deployment. A key strength of the app is that it employs multimodal analysis to evaluate a combination of different data inputs which previous studies have shown can provide valuable additional information leading to greater accuracy when compared to analysis of individual modalities alone [75, 76]. Challenges and limitations This study has several limitations. First, verification bias is likely because only a subset of screen-positive participants underwent neurological evaluation, while screen-negative participants were not routinely clinically verified. Second, the participant profile reflects real-world programme uptake rather than population-representative sampling. In particular, the marked female predominance and province-specific implementation pathways limit direct inference about national PD-risk distribution. In addition, because this study represents the early phase of implementation across a limited number of provinces, it was not designed to formally evaluate urban–rural differences, and such effects cannot be reliably determined from the current dataset. The 10 included provinces were early implementation sites selected on the basis of programme governance readiness and stable data pipelines rather than random national sampling. They were concentrated in central, northern, and northeastern Thailand, and did not include southern provinces during this phase, which potentially limit generalisability. Third, a proportion of participants did not complete the full screening workflow, and reasons for early withdrawal were not systematically captured. This limits our ability to fully characterise barriers to completion and may introduce bias if individuals who discontinued differed systematically from those who completed the assessment. The participant profile in this study reflects real-world programme uptake rather than population-representative sampling. Descriptive analyses showed a marked female predominance in participation and age-related gradients in task completion. In contrast, the exploratory multivariable logistic regression identified two variables significantly associated with completion of the full screening protocol: higher educational attainment and residence outside Bangkok (Table 4). Large population surveys of mobile health (mHealth) apps have also demonstrated systematic gender, age and education differences in app use and sustained engagement [77]. Our study attracted a higher proportion of female app users than men. This pattern probably reflects gender differences in health-seeking behavior and willingness to engage with preventive digital tools, rather than underlying disease epidemiology. The over-representation of women among CheckPD users contrasts with the slightly higher prevalence of PD in men [78]. Similar gender imbalances have been reported across internet- and app-based preventive and mental-health interventions, where evidence suggests that women may disproportionately engage with preventive health and health monitoring technologies, and that digital health innovations can positively affect women’s access to health care and self-care [79–81]. Our findings therefore probably reflect gendered patterns of digital health engagement rather than true PD risk, and raise the possibility that some higher-risk men are under-screened by the current digital strategy. The observed age-related patterns of test completion and dropout raise the possibility that some higher-risk individuals, particularly older men, may be under-screened by the current predominantly app-based strategy [82–84]. Future iterations of CheckPD should therefore consider targeted outreach and tailored messaging to improve engagement among older men and other under-represented groups. These patterns also suggest a dual mechanism of attrition: younger adults (often at lower PD risk) were more likely to skip individual tasks, whereas older adults, the primary target population for PD screening, were more likely to initiate screening but not complete the full assessment workflow. This pattern may reflect motivational dropout amongst younger users and capability- or usability-related dropout amongst older users, consistent with usability research demonstrating that age-related cognitive, sensory, and visuomotor changes can hinder effective interaction with complex digital interfaces and sensor-based tasks. Consequently, datasets used for training and validating digital biomarkers may over-represent younger, digitally proficient, and more adherent users, while under-representing frailer or more impaired older adults. This form of informative missingness represents an important limitation, as it may bias estimates of model performance when digital screening tools are deployed in routine, unsupervised real-world settings [85]. Our findings regarding educational level likely reflect the importance of digital health literacy for effective engagement with the CheckPD application. The lower odds of completion among participants residing in Bangkok should be interpreted in light of the study design. Geographic categorisation was based on Bangkok versus outside Bangkok to reflect differences in implementation context during the early rollout phase across the first 10 provinces, rather than to formally distinguish rural–urban populations. National data indicate that digital access, including smartphone ownership and internet connectivity, is high across Thailand, with only modest differences between urban and rural areas [86]. This suggests that disparities in digital infrastructure alone are unlikely to explain the observed differences in completion. Instead, these findings may reflect variation in implementation approach, with participants outside Bangkok more likely to engage through structured or facilitated pathways (e.g. community or VHV-supported screening), whereas participants in Bangkok relied more on self-directed use. These results highlight the importance of context-specific implementation strategies in optimising engagement in digital health programmes. The usability and user experience findings should be interpreted cautiously because they were derived from a small, non-probability subsample of 30 participants who had completed the full screening workflow. Although the response rate among invited participants was 100%, the sample was not designed or powered to provide a representative estimate of usability across the full implementation cohort. Individuals who were unable or unwilling to complete screening were not included, which may overestimate perceived usability and limits inference regarding barriers experienced by non-completers. The qualitative component should also be interpreted as supplementary implementation feedback rather than a standalone qualitative study. Data were recorded as structured field notes rather than verbatim transcripts, and interviews focused on VHVs and public health officers rather than participants who discontinued screening. As a result, these findings provide useful contextual explanation of implementation processes but should not be interpreted as a comprehensive thematic analysis of participant experience. Assessing the impact of missing data is critical for evaluating app-based screening tools in the field. However, in this real-world implementation, 72% of participants completed all assessment tasks, a substantial level of engagement for a multi-task, unsupervised workflow deployed at scale. This compares favourably with broader mHealth and digital phenotyping literature, where sustained adherence often declines outside controlled research environments and completion rates above 70% are at the higher end of reported benchmarks [87]. For context, while intensive interventions with specialised design elements can reach ‘exceptionally high’ daily completion rates of 84%, typical real-world Ecological Momentary Assessment studies report averages closer to 67.2% [88]. Despite exceeding these benchmarks, our data revealed distinct patterns of incomplete engagement. Completion rates declined as participant age increased, with a notable divergence in behaviour: while younger participants more frequently left specific tasks unattempted, older adults were more likely to terminate assessments mid-sequence. Funnel analyses further clarified this attrition, highlighting cumulative dropout at transitions to complex or sensor-intensive motor tasks. This suggests that a subset of higher-burden activities accounted for a disproportionate share of data loss, underscoring the delicate balance between task complexity and sustained user adherence in remote screening. These patterns are consistent with findings from smartphone-based digital phenotyping studies, in which missing sensor and task data are common and vary widely between individuals, even within well-controlled research protocols [89, 90]. Analyses of large digital phenotyping cohorts, such as the mindLAMP platform, similarly show that adherence decreases as task frequency and complexity increase, with high-friction tasks disproportionately driving disengagement [91, 92]. Systematic reviews of mHealth interventions consistently identify participant attrition and engagement loss as key challenges for real-world implementation [93]. Conclusions The CheckPD programme has demonstrated a successful transition from a validated digital prototype to a scalable, real-world national public health intervention for PD. Its strong real-world performance, reflected in a high PPV of the screening algorithm, combined with a human-centred design approach and robust ethical governance, positions Thailand’s implementation as a credible reference model for digital neurology and population-level brain health initiatives globally. By enabling early identification of individuals at increased risk of PD and establishing clear pathways for clinical follow-up, CheckPD supports timely intervention and behaviour-focused risk modification. This approach aligns closely with emerging prevention-oriented strategies for PD and with the WHO’s Brain Health framework, highlighting the potential of integrated digital screening programmes to contribute meaningfully to neurological disease prevention at scale. Supplementary Information Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Supplementary Material 4. Supplementary Material 5. Acknowledgements This national Parkinson’s disease screening initiative was conducted in honour of Her Royal Highness Princess Maha Chakri Sirindhorn, in recognition of her longstanding commitment to public health, medical education, and the well-being of the Thai population. The authors also gratefully acknowledge the collaboration with the National Health Security Office (NHSO), which was formally established under a Memorandum of Understanding (MOU) signed on 7 May 2024, enabling integration of the CheckPD screening programme within Thailand’s national health system and preventive health framework.The research team would like to express their sincere appreciation to all individuals and organisations involved in this national programme, including local collaborators, provincial coordinators, and administrative and public health personnel, whose dedication and support were essential to the successful implementation of the project. We also gratefully acknowledge the cooperation of government agencies, provincial public health offices, hospitals, and provincial Red Cross chapters across all participating provinces.This project was implemented through the collaboration of the following provinces and their affiliated institutions: Nakhon Sawan Province: Governor of Nakhon Sawan phase 1 (Mr. Thawee Sermphakdikul), President of the Nakhon Sawan Provincial Red Cross Chapter (Mrs. Waraporn Sermphakdikul )and Governor of Nakhon Sawan and President of the Nakhon Sawan Provincial Red Cross Chapter phase 2 (Ms. Chutiphon Sechang), Nakhon Sawan Provincial Public Health Office (Dr. Amnart Noikham , Mr.Teera Kangkhetkron , Miss Nattaya Pattaveenontawong and Miss Wilawan Nunart) Sawanpracharak Hospital (Rattikorn Thungsuk, Chattama Chairat and Mrs.Tassanee Tabthimthai).Nakhon Pathom Province: Governor of Nakhon Pathom and President of the Nakhon Pathom Provincial Red Cross Chapter (Ms. Arocha Nantamontry), Nakhon Pathom Hospital (Dr. Surachai Chokkrchitchai), Sam Phran Hospital (Dr. Tinnakorn Chuenchom), Nakhon Pathom Provincial Public Health Office (Mr.Suphat Katanyutita and Mr.Tawatchai Naksrisung).Nonthaburi Province: Governor of Nonthaburi (Mr. Kiattisak Trongsiri), President of the Nonthaburi Provincial Red Cross Chapter (Mrs. Phonsri Tongsiri), Nonthaburi Provincial Public Health Office (Dr. Paripon Juljerm and Mrs. Sineenart Rattanapunpanit) Pranangklao Hospital (Dr. Sakol Sookprome).Chiang Rai Province: Governor of Chiang Rai (Mr. Charin Tongsuk), President of the Chiang Rai Provincial Red Cross Chapter (Mrs. Sineenat Thongsuk) Chiang Rai Provincial Public Health Office (Dr. Ekkachai Kumlue) Chiang Rai Prachanukroh Hospital (Dr. Achara Laongnualpanich, Miss Netphit Khamhoi and Mrs. Jintana Yoongrum) Mayor of Chiang Rai City Municipality (Mr. Wanchai Jongsutanamanee).Samut Prakan Province: Governor of Samut Prakan (Mr. Suphamit Chinnasri), President of the Samut Prakan Provincial Red Cross Chapter (Miss Orawan Chinnasri), Samut Prakan Provincial Red Cross Chapter (Mrs. Kanchana Pansiri and Mr. Somsak Kaewsana) Samut Prakan Provincial Public Health Office (Mr. Prapart Phookduang and Mrs. Walaipun Sumritwatcharsai) Samut Prakan Hospital (Dr. Pimwalai Chulapimphan).Phetchaburi Province: Governor of Phetchaburi (Pol. Lt. Col. Phopchanok Chalanukhro), President of the Phetchaburi Provincial Red Cross Chapter (Mrs. Natthinee Kongbuchakiat), Phetchaburi Provincial Public Health Office (Mr. Chartchai Kitiyanun, Miss Sununtinee Rungsiriwattanakij and Miss Phiyawan Phobata) Phra Chom Klao Hospital (Dr. Attasit Nawaapisak).Sisaket Province: Governor of Sisaket (Mr. Anupong Suksamonit), President of the Sisaket Provincial Red Cross Chapter (Ms. Phakanan Sila-art), Sisaket Provincial Public Health Office (Dr.Thanong Veerasangpong Ms.Sineenuch Wirasaengpong and Ms. Mali Supatti), Sisaket Hospital (Dr.Sitthipan Janpong).Pathum Thani Province: Governor of Pathum Thani (Mr. Somkid Chanthamruek), President of the Pathum Thani Provincial Red Cross Chapter (Asst. Prof. Dr. Sasithorn Sujarittanakarn) Pathum Thani Provincial Public Health Office (Dr. Non Jindavech and Mr. Apichon Jeensavake) Pathum Thani Hospital (Dr. Saralee Chindamang and Ms. Kamolruetai Suphakawanich). Authors’ contributions R.B. wrote the main manuscript text and J.S. and T.L. prepared figures and tables. All authors contributed to editorial changes in the manuscript. All authors read and approved the final manuscript. Funding This research was funded by the following grants; the Thailand Science Research and Innovation Fund (Program Management Unit for Competitiveness, C01F670185), the National Economic and Social Development Council, Thailand Center for Excellence for Life Sciences (TC (ERP) 31/2568), National Research Council of Thailand (N42A680591, N35E680087), Ratchadapiseksompotch Fund (RA-MF-22/69) and the Center of Excellence grant (CE68_028_3000_004), both awarded by Chulalongkorn University, and the Thai Red Cross Education and Research Committee Fund. Data availability Supplementary materials relevant to this study are available via the ChulaPD Screening Data Repository at: https://drive.google.com/drive/folders/1UqVXnM6HEsLGmdLu2gjng96AhRaQ2ios Declarations Ethics approval and consent to participate The protocol for this study was approved by the Institutional Review Board of the Faculty of Medicine, Chulalongkorn University (IRB No. 0804/65), as well as from the Central Research Ethics Committee (CREC) under the approved protocol entitled ‘Flipping the Paradigm of Parkinson’s Disease: A Model of National ‘Eat, Move, Sleep’ Digital Interventions to Prevent or Slow the Rise of Non-Communicable Diseases in Thailand’ (CREC No. CREC023/68BR-MED06; Certificate Number COA-CREC105/2025). Digital informed consent was obtained from all participants. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Supplementary Material 4. Supplementary Material 5. Data Availability Statement Supplementary materials relevant to this study are available via the ChulaPD Screening Data Repository at: https://drive.google.com/drive/folders/1UqVXnM6HEsLGmdLu2gjng96AhRaQ2ios