Rajesh Sabapathy, Director, Data Science, Optum Global Solutions India discusses methods to improve healthcare delivery and how the system can become more robust overall
As the world grapples with the rapid spread of COVID-19, there is an increase in awareness about the uses of healthcare data and analytics amongst the general public. Traditionally, users of predictive analytics in healthcare fall into three categories:
- Health insurance companies (payers)
- Hospitals or healthcare providers
- Government or Regulatory bodies
We will briefly touch upon seven high-impact predictive analytics use-cases for payers and providers, as the Government usually is a provider and / or a payer in most countries.
I. Health insurance companies (payers)
Minimising medical cost
Payment made on fraudulent claims drives up healthcare costs for everyone and needs to be curbed. Typically, payers use a lot of if-then rules to detect fraudulent claims. However, as fraud becomes more sophisticated, advanced predictive models are required which look for signals across hundreds of variables to detect fraud.
Another strategy in keeping medical cost down is to use disease prediction models in conjunction with population health management programmes. For example, a payer could use a predictive model to identify members who are pre-diabetic and have a high risk of turning diabetic. The most-at-risk members could then be enrolled into a diabetes management programme to reduce their diabetes risk. By keeping members healthy and reducing future high-cost claims, overall medical cost is lowered.
Minimising operational cost
Despite the usage of sophisticated automated claim adjudication systems, a good chunk of complicated claims still require manual processing. Predictive analytics could be used to detect incorrect manual processing. For example, denied claims which should have been paid. This helps avoid claim ‘rework’ and reduces operational cost.
Payers may use predictive analytics for targeted call routing to identify and prioritise members who need assistance the most, like families with special needs children. Calls from such groups could be diverted to a dedicated and trained team, who can then provide assistance regarding clinical or administrative needs. Such programmes reduce hassle for the target groups and provide superior customer experience.
Challenges for payers in harnessing the power of predictive analytics are:
- Consumption: Business and operations teams want to understand why a certain claim is suspected to be fraudulent or why a certain caller is expected to be a repeat caller; if the predictive model is not built with interpretability in mind, adoption suffers.
- Intervention: While predictive models can often identify objects of interest with high accuracy, the real-world intervention is still a business process. Shortcomings in design and execution of the business process will affect the overall project success.
- Attribution: It is difficult to attribute how much of the success is due to the power of a model versus the efficacy of the intervention.
II. Hospitals (providers)
Predictive analytics can help predict delays in operating rooms, which are high-cost physical assets, by using data on historical handoffs and delays. This can help improve the on-time metric significantly. Likewise, by using historical utilisation and breakdown data, models can be built to conduct proactive maintenance of high-cost equipment, like MRI machines, where downtime can prove expensive.
Improved health outcomes
Predictive analytics can be used for: assisting doctors with accurate diagnosis using models based on the patient’s medical history; predicting the risk of onset of in-hospital hard-to-treat infections like sepsis; predicting the risk of the patient being readmitted so that treatment protocol can be modified etc. Such use-cases gain a lot of importance under value-based care regimes where hospitals get paid only when the patient’s condition improves.
Revenue cycle management
Given the high cost of inpatient treatment, it’s essential for hospitals to get their claims paid on time, from a cash flow perspective. To avoid any back-and-forth on claims payments with insurance companies, hospitals can use predictive analytics on claims prior to submission to identify risk of payment being denied or delayed because of improper coding or insufficient information. Claims can thus be filed with the right information to ensure timely receipt of payment.
Supply chain management
About a third of a hospital’s operating cost stems for healthcare supply chain management. Modelling historical utilisation patterns allows a hospital to forecast demand ahead of time and manage inventory using just-in-time principles instead of overstocking. These models also allow procurement teams to negotiate better with suppliers and bring down unit costs.
For hospitals, specific challenges to solve are:
- Moral hazards: People take on more risks if they don’t have to bear the full costs of the risk. In a hospital context, if doctors rely on diagnosis from a model, they may take more risks with both diagnosis and treatment because they may believe that the model is accountable for the decision and not them.
- Bias: Models tend to reflect the bias inherent in the underlying data and also the choices made by the team that built the model. In certain situations, this may end up in diagnosis and treatment unfairly favouring one group of patients over others.