Member Target Model – Clinical
The Final STAR ratings depend on the measure level compliance scores. Better the compliance, better is the rating. To improve the compliance rates, payers need to identify non-compliant members that are most likely to become compliant for the focus measure. To accurately recommend members and providers that have the highest likelihood of closing a specific gap, the Stars Optimizer platform ingests historic member and provider level data. Member level data includes utilization, cost, enrolment, prior gaps, and gap closure information. Other details like demographics, chronic conditions, plan benefits, risk scores, social determinants, interactions, etc. are also beneficial in getting to a more accurate target list. Based on the historical data for compliance for every focus measure, we can narrow down the member list who are most likely to turn compliant for the given year. The targeting algorithms helps figure out how to get to the recommended goal ratings for the focus measures. This will help in improving the compliance rates, and subsequently increase the final STAR ratings. It is executed Weekly/Monthly based on the data available at clients end, Works on HEDIS and PDE data source. Contract-measure level projections.
METHODOLOGY:
1. To generate member target lists, we use Decision Tree approach for observation model which uses observation data as input, and logistic regression approach for Member model which uses member demographic data as input.
2. Member model takes inputs as: Age, Gender, No of Inpatient Visits, No of Outpatient visits, Procedure count, Diagnosis Count
Member Observation Model takes inputs as: ResultCode, ValueSetName, Totalobscount, Duration, StandardMeasureName, TimeSinceLastObservation
3. Member logistic model predicts probability of being compliant for each member, then arrange the members in decreasing order of their probabilities and marks first N members as Target members., where N is taken as sum of Predicted probabilities for that contract-measure combination.
4. So, there will one logistic regression model per Contract-measure combination in Member demographic setup
5. There will be one Decision tree model per contract in Observation Member Setup.
Input variables : BIC (BI Clinical - Rule Engine) member data
Output Variables : Target member, likelihood of engagement
Metrics to Monitor
Statistical
|
:
|
Somers D |
Accuracy |
Precision and Recall |
Confusion Matrix |
F1 Score |
Roc and Auc |
Prevalence |
Detection Rate |
Balanced Accuracy |
Cohen's Kappa |
Concordance |
Gini Coefficent |
KS Statistic |
Youden's J Index
|
Business
|
:
|
Members Engaged
|
Infrastructure
|
:
|
Log Bytes |
Logging/User/IAMPolicy |
Logging/User/VPN |
CPU Utilization |
Memory Usage |
Error Count |
Prediction Count |
Prediction Latencies |
Private Endpoint Prediction Latencies |
Private Endpoint Response Count
|
Visit Model :
medictiv.citiustech.com