Healthcare

What do you get?

Elevated patient care and operational efficiency with AI-driven insights.

What are we going to do for you?

Extract insights from patient dynamics to provide better care opportunities for patients using data analysis and predictive analytics.

Challenge

Healthcare providers often struggle with poor medical adherence to prescriptions, heightened readmission risks and fraudulent claims. Inefficient procurement processes can further hinder patient care.

Solution

Leveraging our expansive domain knowledge in this industry enables OmniData to provide solutions for predicting patient behavior, leading to strategies to unlock efficient care. Using Gen AI our solutions unlock efficiencies in procurement, fraud detection and optimized supply chains.

Predictive intervention

Predictive analytics for patient behavior and recommendations for driving behavior. Reduce the risk and costs associated with non-adherence to prescription.

Fraud Detect abnormal claims being submitted for processing from fraudulent activities and act in the moment.

Improved Procurement

Fulfill medical resource needs while reducing the risk of overstocking inventory levels.

Business Problem

The average hospital readmission rate in the US is 14.5%, with rates ranging between 11.2% and 22.3%. In 2022 Medicare assessed hospitals and found that 42% were penalized for readmission rates exceeding 30-day risk-standardized readmission rates. Preventable rehospitalization costs Medicare hundreds of millions dollars each year and can be avoided through better care.

Paying more attention to patients at risk of readmission during discharge can reduce the penalties significantly. How can patient behavior and profile data inform the classification of patients at risk?

Solution Approach

Using AI algorithms, specifically those associated with classification allows medical institutions and payers to identify clients at risk of readmission. Unlike current methods based on patient interaction only, AI models can ingest behavioral and patient profile data, and find meaningful predictive relationships between factors that shows risk of readmission. Furthermore, it provides business and patient specific actions, personalizing the experience and providing ranges of actions. For example, patients can be classified into Low, Medium and High Risk categories. Actions for these different categories might differ, for example sending discharge instructions via email for a low risk patient vs a nurse or medical practitioner having an in-person engagement and explaining the instructions in detail for high risk patients.

Using data about the patient including demographics, diagnosis and prescribed medicine, a classification model is built to identify patients at risk. OmniData's approach here is also to add a layer of optimization showing where to focus the manual intervention efforts to optimize for cost.

Value Impact

The value of having a robust solution to classify patients at risk is visible in the amount of dollars paid in penalties. Some studies show that upwards of $560M have been paid in penalties under the HRRP. The average penalty per hospital is roughly $217 000 per annum.

OmniData helps clients in the healthcare industry to identify patients at risk of readmission by applying AI methodologies on client data and providing solutions to aid in the decisioning process for healthcare providers to identify patients and provide relevant actions.

Technical Implementation: Data Sources

Target variables are binary in nature showing if a patient was readmitted within a 30-day period post discharge. Patient demographic data such as age, patient source of admission, diagnosis and prescribed medication are all combined to produce a predictive model.

Technical Implementation: Solution Features

  • Data preparation: Exploratory data analysis provides insights into the patterns and trends present. Data quality checks provides insights to data such as missing values, and indicates what the best approach would be to deal with these data points.
  • Feature engineering: Once the data is understood, feature engineering creates a new dimension on the raw data. Features are input variables that will be used in model building.
  • Model training: Features are used in a model training exercise where the most predictive features are selected. The model is also evaluated for performance based on criteria that suits the business problem
  • Model Persistence: Once a model has been selected as best of breed, that model is persisted for future inference and decision making.
  • Decisioning: A combination of business rules and model output is used to provide a decisioning platform to the client to manage patient readmission
  • Visualization: Results are visualized to show the impact, highlight possible future enhancements and track the value impact

Conclusion

Once a patient is discharged from the controlled environment of a hospital, communication and interaction reduce significantly. Many patients are unreachable on phones or email, and the ability to influence their actions diminishes. The most impactful actions are before discharge to ensure the right message is given to the right patient. OmniData's ability to use AI algorithms in this regard to classifying patients for personalized messaging and action reduced readmission risk and the associated costs with it.