The effect of predictive analytics-driven interventions on healthcare utilization

Predictive analytics and early intervention in healthcare social work: a scoping review

This scoping review investigates the untapped potential of predictive analytics in healthcare social work, specifically targeting early intervention frameworks. Despite the escalating attention predictive analytics has garnered across multiple disciplines, its tailored application in social work remains notably sparse. This study endeavors to fill this lacuna by meticulously reviewing the extant literature and delineating the prospective advantages and inherent constraints of integrating predictive analytics into healthcare social work. The outcomes of this inquiry enrich the prevailing dialogue on the utility of predictive analytics in healthcare, offering indispensable perspectives for practitioners and policymakers in the social work domain.

The impact of integrated care on health care utilization and costs in a socially deprived urban area in Germany: A difference-in-differences approach within an event-study framework

We investigated the impact of an integrated care initiative in a socially deprived urban area in Germany. Using administrative data, we empirically assessed the causal effect of its two sub-interventions, which differed by the extent to which their instruments targeted the supply and demand side of healthcare provision. We addressed confounding using propensity score matching via the Super Learner machine learning algorithm. For our baseline model, we used a two-way fixed-effects difference-in-differences approach to identify causal effects. We then employed difference-in-differences analyses within an event-study framework to explore the heterogeneity of treatment effects over time, allowing us to disentangle the effects of the sub-interventions and improve causal interpretation and generalizability. The initiative led to a significant increase in hospital and emergency admissions and non-hospital outpatient visits, as well as inpatient, non-hospital outpatient, and total costs. Increased utilization may indicate that the intervention improved access to care or identified unmet need.

Recurrent heart failure hospitalizations in heart failure with preserved ejection fraction: an analysis of TOPCAT trial

AIMS

Recurrent heart failure hospitalization (HFH) is an important feature of the progression of heart failure (HF). In the Treatment of Preserved Cardiac Function Heart Failure with an Aldosterone Antagonist (TOPCAT) trial, we analysed risk factors for recurrent HFH events in HF patients with preserved ejection fraction (HFpEF) and developed a risk prediction model for recurrent HFH.

METHODS AND RESULTS

This analysis focused on the subset of TOPCAT participants enrolled in the Americas (n = 1767). Recurrent HFH was defined as two or more hospitalizations for HF during the follow-up period. Lasso regression and multivariate logistic regression were used to screen the risk factors, and the risk prediction model of recurrent HFH was established. During a median follow-up period of 3.4 (95% confidence interval: 3.3-3.6) years, 72.2% (542 of 751 total hospitalizations) of HFH events occurred in 9.4% (n = 163) of patients with recurrent HFHs. Patients in the recurrent HFH group had higher cardiovascular mortality rate [6.2 per 100 patient-years (PY) vs. 3.8 per 100 PY, P = 0.016] and all-cause mortality rate (10.0 per 100 PY vs. 6.8 per 100 PY, P = 0.015) than those in the non-recurrent HFH group. The model consisting of nine predictors has moderate predictive power for recurrent HFH events in patients with HFpEF (AUC = 0.75, Brier score = 0.08). Decision curve analysis showed a net clinical benefit from the application of the prediction model.

CONCLUSIONS

In patients with HFpEF, the majority of HFHs occur in a small proportion of patients with repeated hospitalizations, who typically have more comorbidities and are at higher risk of death. The predictive model developed in this study helps to identify patients at high risk of recurrent HFH.

DEPLOYR: a technical framework for deploying custom real-time machine learning models into the electronic medical record

OBJECTIVE

Heatlhcare institutions are establishing frameworks to govern and promote the implementation of accurate, actionable, and reliable machine learning models that integrate with clinical workflow. Such governance frameworks require an accompanying technical framework to deploy models in a resource efficient, safe and high-quality manner. Here we present DEPLOYR, a technical framework for enabling real-time deployment and monitoring of researcher-created models into a widely used electronic medical record system.

MATERIALS AND METHODS

We discuss core functionality and design decisions, including mechanisms to trigger inference based on actions within electronic medical record software, modules that collect real-time data to make inferences, mechanisms that close-the-loop by displaying inferences back to end-users within their workflow, monitoring modules that track performance of deployed models over time, silent deployment capabilities, and mechanisms to prospectively evaluate a deployed model's impact.

RESULTS

We demonstrate the use of DEPLOYR by silently deploying and prospectively evaluating 12 machine learning models trained using electronic medical record data that predict laboratory diagnostic results, triggered by clinician button-clicks in Stanford Health Care's electronic medical record.

DISCUSSION

Our study highlights the need and feasibility for such silent deployment, because prospectively measured performance varies from retrospective estimates. When possible, we recommend using prospectively estimated performance measures during silent trials to make final go decisions for model deployment.

CONCLUSION

Machine learning applications in healthcare are extensively researched, but successful translations to the bedside are rare. By describing DEPLOYR, we aim to inform machine learning deployment best practices and help bridge the model implementation gap.

Supply-side solutions targeting demand-side characteristics: causal effects of a chronic disease management program on adherence and health outcomes

We estimate the effects of a chronic disease management program (CDMP) which adapts various supply-side interventions to specific demand-side conditions (disease-staging) for patients with chronic kidney disease (CKD). Using a unique dataset on the entire population of the Emilia-Romagna region of Italy with hospital-diagnosed CKD, we estimate the causal effects of the CDMP on adherence indicators and health outcomes. As CKD is a progressive disease with clearly-defined disease stages and a treatment regimen that can be titrated by disease severity, we calculate dynamic, severity-specific, indicators of adherence as well as several long-term health outcomes. Our empirical work produces statistically significant and sizeable causal effects on many adherence and health outcome indicators across all CKD patients. More interestingly, we show that the CDMP produces larger effects on patients with early-stage CKD, which is at odds with some of the literature on CDMP that advocates intensifying interventions for high-cost (or late-stage) patients. Our results suggest that it may be more efficient to target early-stage patients to slow the deterioration of their health capital. The results contribute to a small, recent literature in health economics that focuses on the marginal effectiveness of CDMPs after controlling either for supply- or demand-side sources of heterogeneity.

The Potential For Bias In Machine Learning And Opportunities For Health Insurers To Address It

As the use of machine learning algorithms in health care continues to expand, there are growing concerns about equity, fairness, and bias in the ways in which machine learning models are developed and used in clinical and business decisions. We present a guide to the data ecosystem used by health insurers to highlight where bias can arise along machine learning pipelines. We suggest mechanisms for identifying and dealing with bias and discuss challenges and opportunities to increase fairness through analytics in the health insurance industry.

Patterns of healthcare utilisation in children and young people: a retrospective cohort study using routinely collected healthcare data in Northwest London

OBJECTIVES

With a growing role for health services in managing population health, there is a need for early identification of populations with high need. Segmentation approaches partition the population based on demographics, long-term conditions (LTCs) or healthcare utilisation but have mostly been applied to adults. Our study uses segmentation methods to distinguish patterns of healthcare utilisation in children and young people (CYP) and to explore predictors of segment membership.

DESIGN

A retrospective cohort study.

SETTING

Routinely collected primary and secondary healthcare data in Northwest London from the Discover database.

PARTICIPANTS

378 309 CYP aged 0-15 years registered to a general practice in Northwest London with 1 full year of follow-up.

PRIMARY AND SECONDARY OUTCOME MEASURES

Assignment of each participant to a segment defined by seven healthcare variables representing primary and secondary care attendances, and description of utilisation patterns by segment. Predictors of segment membership described by age, sex, ethnicity, deprivation and LTCs.

RESULTS

Participants were grouped into six segments based on healthcare utilisation. Three segments predominantly used primary care, two moderate utilisation segments differed in use of emergency or elective care, and a high utilisation segment, representing 16 632 (4.4%) children accounted for the highest mean presentations across all service types. The two smallest segments, representing 13.3% of the population, accounted for 62.5% of total costs. Younger age, residence in areas of higher deprivation and the presence of one or more LTCs were associated with membership of higher utilisation segments, but 75.0% of those in the highest utilisation segment had no LTC.

CONCLUSIONS

This article identifies six segments of healthcare utilisation in CYP and predictors of segment membership. Demographics and LTCs may not explain utilisation patterns as strongly as in adults, which may limit the use of routine data in predicting utilisation and suggest children have less well-defined trajectories of service use than adults.

Do payor-based outreach programs reduce medical cost and utilization?

There is growing interest in using predictive analytics to drive interventions that reduce avoidable healthcare utilization. This study evaluates the impact of such an intervention utilizing claims from 2013 to 2017 for high-risk Medicare Advantage patients with congestive heart failure. A predictive algorithm using clinical and nonclinical information produced a risk score ranking for health plan members in 10 separate waves between July 2013 and May 2015. Each wave was followed by an outreach intervention. The varying capacity for outreach across waves created a set of arbitrary intervention treatment cutoff points, separating treated and untreated members with very similar predicted risk scores. We estimate a difference-in-differences model to identify the effects of the intervention program among patients with a high score on care utilization. We find that enrollment in the intervention decreased the probability and number of hospitalizations (by 43% and 50%, respectively) and emergency room visits (10% and 14%, respectively), reduced the time until a primary care visit (8.2 days), and reduced total medical cost by $716 per month in the first 6 months following outreach.

Prior Heart Failure Hospitalization, Clinical Outcomes, and Response to Sacubitril/Valsartan Compared With Valsartan in HFpEF

BACKGROUND

The period shortly after hospitalization for heart failure (HF) represents a high-risk window for recurrent clinical events, including rehospitalization or death.

OBJECTIVES

This study sought to determine whether the efficacy and safety of sacubitril/valsartan varies in relation to the proximity to hospitalization for HF among patients with HF with preserved ejection fraction (HFpEF).

METHODS

In this post hoc analysis of PARAGON-HF (Prospective Comparison of ARNI [Angiotensin Receptor-Neprilysin Inhibitor] with ARB [Angiotensin Receptor Blocker] Global Outcomes in HFpEF), we assessed the risk of clinical events and response to sacubitril/valsartan in relation to time from last HF hospitalization among patients with HFpEF (≥45%). The primary outcome was composite total HF hospitalizations and cardiovascular death, analyzed by using a semiparametric proportional rates method, stratified by geographic region.

RESULTS

Of 4,796 validly randomized patients in PARAGON-HF, 622 (13%) were screened during hospitalization or within 30 days of prior hospitalization, 555 (12%) within 31 to 90 days, 435 (9%) within 91 to 180 days, and 694 (14%) after 180 days; 2,490 (52%) were never previously hospitalized. Over a median follow-up of 35 months, risk of total HF hospitalizations and cardiovascular death was inversely and nonlinearly associated with timing from prior HF hospitalization (p < 0.001). There was a gradient in relative risk reduction in primary events with sacubitril/valsartan from patients hospitalized within 30 days (rate ratio: 0.73; 95% confidence interval: 0.53 to 0.99) to patients never hospitalized (rate ratio: 1.00; 95% confidence interval: 0.80 to 1.24; trend in relative risk reduction: p_interaction = 0.15). With valsartan alone, the rate of total primary events was 26.7 (≤30 days), 24.2 (31 to 90 days), 20.7 (91 to 180 days), 15.7 (>180 days), and 7.9 (not previously hospitalized) per 100 patient-years. Compared with valsartan, absolute risk reductions with sacubitril/valsartan were more prominent in patients enrolled early after hospitalization: 6.4% (≤30 days), 4.6% (31 to 90 days), and 3.4% (91 to 180 days), whereas no risk reduction was observed in patients screened >180 days or who were never hospitalized (trend in absolute risk reduction: p_interaction = 0.050).

CONCLUSIONS

Recent hospitalization for HFpEF identifies patients at high risk for near-term clinical progression. In the PARAGON-HF trial, the relative and absolute benefits of sacubitril/valsartan compared with valsartan in HFpEF appear to be amplified when initiated in the high-risk window after hospitalization and warrant prospective validation. (PARAGON-HF; NCT01920711).