An overview of the client

With four hospitals, twelve outpatient clinics, and a network of more than 300 affiliated physicians, HealthPlus is a regional healthcare network that serves 1.2 million people. HealthPlus, which was established in 1982 and has a reputation for providing high-quality treatment, was under growing pressure from bigger healthcare systems and shifting payment models that prioritized patient outcomes and value-based care.

The difficulty

In late 2021, when they came to us, HealthPlus was facing a number of serious issues that were affecting patient care as well as their capacity to be financially stable:

• Excessive Readmission Rates: HealthPlus was facing a 30-day readmission rate of 18% for specific diseases, which was significantly higher than the 13.8% national average. As a result, Medicare penalties topped $2.3 million per year.
• Inefficiencies in Resource Allocation: There were times when there was both understaffing and overstaffing since staffing and resource allocation were mostly determined by past trends rather than anticipated needs.
• Data Silos: It was practically hard to conduct a thorough study because vital patient and operational data was dispersed throughout several systems (EHR, billing, scheduling, etc.) with little integration.
• Reactive Care Model: This approach to care delivery was mostly reactive rather than proactive, and it had little capacity to identify patients who were at risk before their diseases deteriorated.
• The market position of HealthPlus was under threat from rival healthcare systems in nearby markets that were starting to use advanced analytics.

Their predicament was summed up by HealthPlus's Chief Medical Officer, Dr. Elena Patel: "We were starving for insights but drowning in data. Information was being generated by all these systems, but we were unable to make the connections necessary to enhance operations or care. We knew we wanted to go from reactive to proactive treatment, but we weren't sure how to get there, and our readmission penalties were increasing in the meanwhile."

How We Proceed

We created a tiered strategy for deploying predictive analytics that would yield both short-term gains and long-term transformation following a comprehensive evaluation of HealthPlus's data architecture, clinical workflows, and strategic goals:

Phase 1 (three months): Data Foundation and Governance

We started by building a strong data base that would serve as the basis for all upcoming analytics projects:

• Healthcare Data Lake: Combined data from all source systems using a data lake design that complies with HIPAA.
• Data Integration Framework: Created automated pipelines to load, extract, and transform data from scheduling, billing, EHR, and other systems.
• Program for Data Quality: Developed procedures to find and address problems with data quality at their root
• Clearly defined duties and responsibilities were established for a cross-functional data governance committee.
• Data inventory: To increase discoverability, a thorough inventory of the data assets that are available was created.

Close coordination with HealthPlus's clinical leadership and IT team was necessary during this phase to guarantee that the data foundation would satisfy both technical and clinical criteria. "The data governance work wasn't glamorous," says HealthPlus CIO Michael Chen, "but it was absolutely necessary. We had a single source of truth and assurance on the accuracy of our data for the first time."

Phase 2 of the Data Integration Workshop: Readmission Risk Prediction Model (four months)

After laying the data foundation, we concentrated on creating a predictive model to deal with high readmission rates, which were HealthPlus's biggest problem:

• Model Development: The 30-day readmission risk for patients with COPD, pneumonia, and congestive heart failure was predicted using a machine learning model.
• Feature Engineering: identified and included more than 200 factors related to clinical, demographic, and social determinants of health.
• Model Validation: By analyzing historical data, the model was validated and was able to predict high-risk patients with 83% accuracy.
• Integrating the model into clinical workflows allowed for automated notifications for high-risk patients and the presentation of risk scores in the EHR.
• Evidence-based intervention strategies for patients at varying risk levels were developed in collaboration with clinical teams.

Several iterations and close collaboration with doctors were necessary during the development phase to guarantee that the model would be accurate and useful. "We were first dubious about how a mathematical model could forecast something as complicated as readmissions," acknowledges Director of Care Management Dr. Sarah Johnson. "But when we saw how it incorporated factors we hadn't been systematically considering—like medication adherence patterns and social support—we began to see its value."

Phase 3: Five Months of Operational Analytics and Resource Optimization

To solve operational issues, we extended the analytics program, building on the readmission model's success:

• ED volume, admission rates, and duration of stay were predicted using patient flow analytics.
• Using a dynamic staffing model, staffing optimization forecasted staffing requirements by department, shift, and day of the week.
• Utilization of Resources Dashboard: Real-time dashboards displaying the present and anticipated use of resources were implemented.
• Tools for Capacity Planning: Created scenario planning tools to maximize equipment use and bed distribution
• Models were developed to forecast the financial effects of operational modifications through financial impact analysis.

Significant change management was needed during this phase as staff members adjusted to making decisions based on data. "At first, there was some resistance," says Chief Nursing Officer Jennifer Martinez. "Managers and nurses were accustomed to depending on their intuition and experience when making staffing decisions. We needed to prove that the analytics were adding to their judgment rather than taking its place."

Phase 4: Continuous Proactive Management of Population Health

We increased the program's scope to encourage proactive population health management once the early models showed promise:

• Chronic Disease Progression Models: These models were created to forecast the course of diabetes, high blood pressure, and other chronic illnesses.
• Analytics for Care Gaps: Analytics were used to determine whether patients needed to be seen for preventative screens or chronic illness care.
• The Social Determinants Integration: Improved models using data on socioeconomic determinants of health to find non-clinical care barriers
• Optimizing patient engagement: developed algorithms to forecast which engagement strategies would work best for certain patient groups.
• Findings Estimate: created instruments to forecast the probable results of various intervention tactics

Impact and Outcomes

HealthPlus has reaped substantial, quantifiable benefits from the use of predictive analytics:

Beyond these measurable advantages, HealthPlus's operational and service delivery methods have been completely changed by the predictive analytics program:

• Care teams now proactively contact high-risk patients before their diseases worsen, according to the proactive care model.
• Predictive insights are becoming more and more important in clinical and operational choices in a data-driven culture.
• Resource optimization has reduced waste and shortages by bringing staffing and resources into line with anticipated demand.
• Higher patient satisfaction ratings are the result of more effective operations and proactive care.
• Differentiation in the marketplace: HealthPlus is now regarded as a regional leader in data-driven medical care.

"The predictive analytics initiative has radically altered the way we practice medicine. Instead of merely responding to patients who come to us, we are now determining who need assistance before they are even aware of it. I just heard from one of our doctors about a patient who came in after we sent out our outreach call and was diagnosed with a dangerous disease that, had we not detected it in time, would have probably required an emergency hospitalization. We are reminded of the significance of this job by their stories."

— Dr. Elena Patel, HealthPlus's Chief Medical Officer

Lessons Learned

Other healthcare organizations starting predictive analytics projects can benefit from the following insightful findings from the HealthPlus implementation:

• Start with a Specific, High-Impact Clinical Problem: Getting started with a clear clinical problem (readmissions) allowed for rapid value demonstration and momentum building.
• Spend money on high-quality data: The caliber of the underlying data has a direct impact on the predictive models' usefulness and accuracy.
• Integrate Analytics into Processes: The impact of even the most precise forecasts is little if they are not smoothly incorporated into operational and clinical processes.
• Maintain equilibrium Intricacy and Interpretability: For clinicians to believe and act upon a model's predictions, they must comprehend why it is making them.
• Evaluate and Share the Impact: To maintain support and uptake, analytics efforts must be regularly measured and their impact communicated.

Most significantly, the project showed that a variety of technological know-how, clinical understanding, and change management abilities are necessary for predictive analytics to be successful in the healthcare industry. As Dr. Patel points out, "The technology was impressive, but what really made this successful was how it was implemented—with a deep understanding of our clinical workflows and culture."