Um teste de envolvimento baseado em mensagens de texto para visitas de crianças saudáveis (HEDIS)

Study Overview:

This three-arm randomized clinical trial evaluates whether conversational AI-powered text message outreach with appointment scheduling assistance improves well-child visit completion rates among Medicaid-enrolled children aged 0-21 years in Virginia compared with automated text messages alone or traditional passive outreach. Randomization occurred June 1, 2025. Interventions were delivered June 9-July 14, 2025. Outcomes were observed through December 31, 2025 via administrative claims.

Purpose of the Study:

The primary purpose of this study is to determine whether conversational AI-powered appointment scheduling assistance, delivered via text message, improves well-child visit completion rates in Medicaid-enrolled children compared with automated SMS reminders alone or traditional passive outreach.

A secondary purpose is to evaluate whether AI-facilitated scheduling reduces staff time per completed appointment relative to traditional care team scheduling.

Research Questions:

Does automated SMS outreach improve well-child visit completion rates compared to traditional passive outreach? Does conversational AI-powered scheduling assistance lead to higher completion rates than automated SMS alone?

A Priori Hypotheses:

We hypothesized that automated SMS outreach would improve well-child visit completion rates relative to traditional passive outreach, and that conversational AI-powered scheduling assistance would further improve completion rates relative to automated SMS alone. We also hypothesized that AI-facilitated scheduling would reduce staff time per scheduled appointment relative to traditional care team scheduling.

Study Design:

This is a three-arm parallel-group superiority randomized clinical trial. Participants were randomized at the household level in a 1:1:1 ratio; all eligible children within the same household were assigned to the same arm to prevent intra-household contamination.

Arm 1 (Traditional Passive Outreach - Control): Participants received standard health plan outreach consisting of periodic mailed reminders and retained access to all standard appointment scheduling methods.

Arm 2 (Automated SMS): Participants received standardized automated text message reminders at predetermined intervals (initial contact, 2-week follow-up, 4-week follow-up). Messages provided general information about visit importance and instructions to call providers to schedule. No interactive or conversational capabilities were included.

Arm 3 (Automated SMS + Conversational AI Scheduling Assistance): Participants received automated text message reminders with response options. When families expressed interest, structured follow-up texts collected appointment preferences. An AI-powered telephone scheduling system then contacted the child's primary care provider directly to book appointments on behalf of families. The system escalated to human staff when clinic policies were incompatible with automated scheduling or technical failures occurred.

Participant Population:

The study enrolled Medicaid beneficiaries aged 0-21 years enrolled through a managed care organization in Virginia who had not completed a well-child visit in 2025. Eligible beneficiaries were identified, screened for exclusions (do-not-contact designation, inactive phone service), and randomized 1:1:1 at the household level. All randomized participants were included in the intention-to-treat analysis. Participant flow and baseline characteristics are reported in the Participant Flow and Baseline Characteristics modules.

Inclusion Criteria:

Medicaid beneficiary aged 0-21 years; had not completed a well-child visit in 2025; enrolled through the managed care organization; authorized for health plan outreach; valid mobile phone number capable of receiving SMS.

Exclusion Criteria:

Do-not-contact designation; inactive phone service; documented request for no contact; prior SMS well-child visit outreach earlier in the 2025 measurement year.

Study Procedures:

Identification of Eligible Participants: Medicaid beneficiaries aged 0-21 years with no completed well-child visit in 2025 were identified through HEDIS quality measure tracking.

Randomization: Participants were randomized at the household level in a 1:1:1 ratio using a computer-generated random sequence. All eligible children within the same household were assigned to the same arm.

Implementation of Outreach Strategy: Interventions were delivered June 9-July 14, 2025. Arm 1 received periodic mailed reminders. Arm 2 received standardized automated text message reminders at predetermined intervals. Arm 3 received automated text message reminders plus conversational AI-powered scheduling assistance; upon family request, an AI system contacted the child's primary care provider directly to book appointments.

Outcome Ascertainment: Well-child visit completion was ascertained through administrative claims using HEDIS technical specifications. Claims were extracted March 17, 2026, providing a 90-day run-out period.

Data Collection: Staff time per scheduled appointment was measured via time-motion analysis comparing AI-facilitated and traditional care team scheduling workflows.

Outcome Measures:

Primary Outcome: Completed well-child visit through December 31, 2025, ascertained through administrative claims using HEDIS technical specifications (binary outcome).

Secondary Outcome: Staff time in minutes per successfully scheduled appointment, comparing AI-facilitated scheduling versus traditional care team scheduling via time-motion analysis.

Tertiary Outcome (Arm 3 only): Percentage of automated scheduling attempts requiring human staff intervention.

Data Analysis:

The primary analysis compared well-child visit completion rates across arms using the χ² test of independence. Absolute risk differences, relative risks, number needed to treat, and 95% confidence intervals were calculated using the Wald method. A generalized estimating equations (GEE) model with binomial family, exchangeable correlation structure, and robust sandwich standard errors was fit to account for within-household correlation arising from household-level randomization. All analyses were intention-to-treat. Bonferroni correction was applied for multiple comparisons. Analyses were conducted in Python 3.10 (SciPy, StatsModels). All tests were two-sided with α=0.05.

Data Monitoring and Confidentiality:

All infrastructure was HIPAA-compliant and SOC 2 Type II audited. Business Associate Agreements were executed with all vendors prior to trial initiation. All protected health information was handled per HIPAA requirements and BAA terms. Individual participant data will not be made available due to patient privacy protections and managed care organization confidentiality requirements.

Risk/Benefit Assessment:

This study is categorized as minimal risk. Potential risks include: privacy breaches (minimal likelihood due to HIPAA-compliant, SOC 2 Type II audited infrastructure); patient annoyance with SMS messages (participants can opt out via do-not-contact designation); miscommunication of appointment details (booking accuracy verified through automated confirmation texts to families and cross-referencing against monthly claims data feeds).

Anticipated Benefits: Potential improvement in well-child visit completion rates and associated preventive care, developmental screening, and immunization delivery; potential reduction in family burden through direct AI-facilitated appointment scheduling; potential reduction in staff time per scheduled appointment relative to traditional care team workflows.

Subject Recruitment and Consent:

Participants were identified from existing Medicaid beneficiary lists where Waymark is authorized to conduct outreach. The trial protocol was approved with waiver of informed consent based on minimal risk determination, as interventions represent variations of standard health plan outreach practices. The trial was prospectively registered at ClinicalTrials.gov (NCT06698640) on November 18, 2024, prior to participant enrollment.