A randomized trial examining the effect of predictive analytics and tailored interventions on the cost of care

COVID-19 has accelerated digital medicine adoption, with telehealth use surging to mitigate healthcare system burdens while maintaining quality and safety. Within this expanding landscape, Personal Emergency Response Systems (PERS) enable older adults to live independently by providing access to immediate assistance. Recent enhancements include predictive models that use PERS data to identify patients at risk of emergency department (ED) transports, enabling proactive intervention. The present study embedded such predictive analytics into a Stepped-Care intervention (predictive model-driven risk identification followed by nurse-led, tailored interventions) to evaluate its impact on healthcare utilization and costs in older patients using PERS. Many healthcare organizations target the highest-cost patients (e.g., top 5% of the cost pyramid), but such programs have not consistently generated the expected cost savings. A prior longitudinal retrospective study found that the middle-cost segment persistently accounted for the highest costs over five years, with the greatest annualized cost increases. Guided by this, the current randomized controlled trial focuses on the middle segment to test whether a Stepped-Care intervention can reduce healthcare utilization and costs. The objective is to evaluate the impact of the Stepped-Care approach on healthcare costs (primary focus here) among older adults in the middle segment of the cost pyramid; clinical utilization outcomes are summarized from a companion paper and briefly reported in the results.

The paper situates its contribution within several strands of literature: (1) Population health programs often target the top 5% of high-cost patients, yet evidence shows mixed cost-savings despite clinical improvements (e.g., McWilliams & Schwartz; Peikes et al.; VA intensive outpatient program). (2) A prior longitudinal retrospective analysis by the authors’ group showed the middle segment of the cost pyramid was persistently the costliest and had the largest annualized cost increases, motivating focus on this segment. (3) Effective care models for older adults with multimorbidity include Guided Care, Project BOOST, the Transitional Care Model, and Mobile Integrated Healthcare, which have demonstrated variable net savings (e.g., $1,364 to $4,027 per patient in some contexts). (4) National analyses (CMS, AHA) indicate a disproportionate share of total costs are driven by relatively few inpatient encounters, underscoring the importance of interventions that reduce hospitalizations/readmissions. (5) The rapid expansion of telehealth, catalyzed by COVID-19, along with supportive policy changes, provides a conducive environment for team-based, digitally enabled care models such as the studied Stepped-Care approach.

Design: Two-arm randomized controlled trial with 1:1 allocation to intervention (IG) or control (CG). Randomization used a computerized random-number generator; allocation concealment via opaque envelopes opened after consent. Study Period: 9 months total—3-month observation (to collect PERS data for predictive modeling) followed by 6-month intervention. Enrollment occurred May 2017–July 2018. Participants: MGB patients receiving home care from Partners HealthCare at Home (PHH), aged ≥65, English-speaking, and in the middle segment (6th–50th percentile) of the cost pyramid based on total healthcare costs in the prior fiscal year (2016 for 2017 enrollees; 2017 for 2018 enrollees), determined from historical data. Exclusions: Implanted devices (potential interference with PERS), dementia/Alzheimer’s/psychiatric illness (anxiety disorder or psychosis), inpatient length-of-stay >30 days during intervention period, discharge to long-term skilled nursing facility, or missing identifiers preventing linkage of EHR and PERS data for predictive scoring. All provided written informed consent. Intervention: All participants received PERS and instructions to use the help button or call 911 as needed. CG received care-as-usual throughout. IG received a Stepped-Care intervention during the 6-month intervention: Step 1—daily 30-day ED transport risk scoring via a predictive model using PERS utilization/events/outcomes, self-reported medical conditions, and demographics; AUC 0.78. High-risk IG patients triggered Step 2—nurse-led triage and tailored care plan following a telephone needs assessment (covering general health, respiratory symptoms, physical activity, ADLs, pain, bladder control). Tailored components could include follow-up calls with feedback, patient education (4-week program), home visits, PCP or outpatient visits, and/or telemonitoring, as clinically indicated. Outcomes and Data Sources: Primary focus in this paper is cost outcomes; utilization outcomes summarized from companion paper. Costs defined as the sum of variable and fixed costs for direct and indirect patient care for all inpatient and outpatient encounters occurring between days 1–180 of the intervention period. Inpatient encounters were hospital admissions; outpatient encounters included ambulatory/medical/surgical encounters without an overnight stay. Cost and utilization data sourced from the MGB Enterprise Data Warehouse (billing and internal hospital cost data, not insurance claims or payments). Baseline demographics, needs, and comorbidities collected via enrollment and needs assessment questionnaires and stored in REDCap. Statistical Analysis: Intention-to-treat. Descriptive statistics compared groups at baseline. Event counts (e.g., inpatient/outpatient encounters) modeled via Poisson regression. Costs modeled with linear regression to evaluate accumulated cost trends within and between groups and to estimate annualized per-patient costs. Justification for linear regression based on central limit theorem approximation of summed exponential (Gamma) costs; Gamma models produced similar inferences. Sensitivity analyses using ANOVA assessed need to adjust for demographics; baseline models (group and intervention duration) were sufficient (no p<0.05). Power: For the primary utilization outcome (ED encounters), power 0.80, alpha 0.05, effect size 0.35; calculated 160 per arm, adjusted to total N=370 for 15% attrition.

Participants: 4957 assessed; 370 randomized (CG n=189, IG n=181). After exclusions (missing data, LOS>30 days), 331 analyzed (CG n=172 [91%], IG n=159 [88%]). Baseline: No significant differences in demographics, comorbidities, or pre-intervention 90-day utilization/costs. Population median age 80; 67% female; 85% white; common comorbidities: hypertension 60%, inflammatory pain disorders 58%, hyperlipidemia 36%, cancer 31%; 62% with ≥3 conditions. Clinical utilization (from companion paper): • ED encounter rate: 15% decrease (p=0.291, ns). • 90-day readmissions: 68% fewer in IG (p=0.007); proportion with 90-day readmission reduced 76% (9.9% CG vs 2.5% IG, p=0.011). • 180-day readmissions: 53% fewer in IG (p=0.020). • EMS encounters: 49% fewer in IG (p=0.006). • Other outcomes trending lower in IG but not significant: inpatient encounters (−14%, p=0.438), 30-day readmissions (−57%, p=0.083), ED transports (−33%, p=0.153). Cost outcomes (180-day intervention; USD): Within-group totals: • CG total costs accumulated to $1,548,355 (inpatient $1,021,437 [66%]; outpatient $526,918 [34%]); mean per patient $9,002 (SD $22,047); inpatient $5,939 (SD $16,962); outpatient $3,063 (SD $8,993). • IG total costs accumulated to $1,142,090 (inpatient $656,188 [57%]; outpatient $485,902 [43%]); mean per patient $7,182 (SD $13,304); inpatient $4,127 (SD $9,503); outpatient $3,056 (SD $7,716). Cost trends (linear regression on accumulated daily costs): • CG daily total cost increase: $8.5K/day (p<0.001), driven by inpatient $5.9K/day (p<0.001) and outpatient $2.6K/day (p<0.001). • IG daily total cost increase: $6.5K/day (p<0.001), driven by inpatient $3.7K/day (p<0.001) and outpatient $2.8K/day (p<0.001). Between-group differences: • Daily total cost reduction in IG vs CG: −$2,049/day (p=0.0094). • Per-patient daily total cost reduction IG vs CG: −$9/day (p=0.0402). Annualized per-patient costs (from models): • Total: CG $17.7K (95% CI $16.0K–$19.4K) vs IG $14.2K ($12.5K–$15.9K); reduction $3.5K (−20%), p=0.04. • Inpatient: CG $11.8K ($10.0K–$13.7K) vs IG $8.1K ($6.3K–$10.0K); reduction $3.7K (−31%), p=0.02. • Outpatient: CG $5.8K ($5.2K–$6.4K) vs IG $6.1K ($5.5K–$6.7K); +4% (ns), p=0.10. Encounter distributions and unit costs: • CG: 58 inpatient (5.9%) and 924 outpatient (94.1%) encounters; inpatient comprised 66% of total cost; mean cost per inpatient encounter $17,611 (SD $16,241); per outpatient encounter $570 (SD $1,237). • IG: 46 inpatient (4.8%) and 922 outpatient (95.2%) encounters; inpatient comprised 57% of total cost; mean cost per inpatient encounter $14,265 (SD $11,116); per outpatient encounter $527 (SD $1,378). High utilizer subgroup: • Similar proportions with any inpatient encounter (CG 21.5% vs IG 23.3%, p=0.801). • Marked reduction in multiple inpatient utilizers in IG (CG 15 patients [8.7%] with 36 encounters vs IG 6 patients [3.8%] with 15 encounters). • Multiple-encounter costs: CG $740.1K (72% of CG inpatient costs) vs IG $156.5K (24% of IG inpatient costs). Diagnosis categories (CCSR): • Top categories (e.g., valve disorders, heart failure, septicemia, GI hemorrhage) accounted for ~27.4% of inpatient costs; top 10 of 62 categories accounted for ~46.7% of inpatient costs.

The Stepped-Care intervention combining predictive analytics with nurse-led tailored interventions significantly reduced annualized total healthcare costs per patient by 20%, primarily through a 31% reduction in inpatient costs. Outpatient costs were similar between groups, indicating the intervention did not increase outpatient spending sufficiently to offset savings from reduced hospitalization-related costs. The findings align with national evidence that a small proportion of inpatient encounters drive a large share of total costs; by reducing multiple hospitalizations among high utilizers, the intervention effectively targeted the key cost drivers in the middle segment of the cost pyramid. Compared with prior programs focused on high-cost patients, this study demonstrates meaningful cost reduction in the often-overlooked middle-cost segment, complementing literature on Guided Care, Transitional Care, and other care management models. In the context of rapid telehealth expansion and supportive policy changes, integrating predictive risk stratification with proactive, team-based care can support population health management for older adults at home, potentially improving outcomes while reducing costs.

This randomized controlled trial shows that a Stepped-Care model leveraging predictive analytics and tailored nurse-led interventions can reduce healthcare costs in older adults in the middle segment of the cost pyramid, chiefly by lowering inpatient utilization and costs, particularly among high utilizers. The approach fits within evolving, team-based digital medicine models and can help health systems provide high-quality, home- and community-based care in the post-COVID era while alleviating pressures on hospital resources. The results support broader implementation and evaluation of predictive, telehealth-enabled population health programs; future work could assess generalizability across more diverse populations and integrate costs captured outside a single health system.

Generalizability may be limited as the cohort was predominantly older, female, white, living alone, and highly educated. Cost and utilization data were limited to encounters within the Mass General Brigham system; encounters outside the system were not captured. Some patients had missing data, and exclusions (e.g., LOS >30 days) were applied; these factors may affect the magnitude of observed cost differences.