How a healthcare provider used data technology to significantly reduce the non-payment of medical bills by their patients.
Low correlation between risk scoring and bill default rate of patients, along with millions of data points across multiple variables, made predicting payment issues difficult.
dotData’s Feature Factory technology was applied to the data, allowing an increase in data size and the inclusion of multivariate data. ML was used to identify risk parameters.
With dotData, the company reduced non-payment by 25% and increased the accuracy of default prediction 400%. The client was able to stratify risk profiles, allowing a targeted outreach.
It can be difficult to accurately predict ongoing medical billing costs in America’s insurance-based healthcare system. Furthermore, patients’ ability to pay is notoriously challenging to assess, as is their risk of eventual default on payments, since this depends upon many factors beyond the purview of a healthcare provider.
From patients’ perspectives, it’s all too easy to fall into debt. A recent KFF poll found that over half of respondents had recently fallen into debt due to the cost of healthcare. In 2021, research by Stanford economist Neale Mahoney estimated healthcare debt at over $140 billion. As Mahoney put it, “When you think about financial distress — debt collectors calling and knocking on doors of households — our research shows that more than half the time now, it is about medical debt.”
The KFF’s Survey of Income and Program Participation in 2022 indicated the problem could be even worse, suggesting that Americans owe over $195 billion in medical debt and reporting that around 3 million people (around 1%) owe sums over $1000.
From a healthcare provider’s perspective, this scale of unmet financial commitments cannot be borne lightly. A 2018 study in Wisconsin found that lawsuits by healthcare providers over unpaid debts had increased by 37% since 2001. Many of the delays are understandable since insurers and patients often have a complex negotiation process regarding what’s covered by each policy.
However, it’s untenable for a provider to shoulder the burden of a mountain of unpaid debt on an ongoing basis. The COVID-19 pandemic exacerbated the existing problem, putting further pressure on hospitals’ Accounts Receivable departments. One survey revealed that around 48% of hospitals were dealing with a significant increase in patient debt. Additional changes to health insurance administration and an ongoing financial downturn further deepened this crisis. It became vital to improving financial risk management in patient bill payments.
The client, a major healthcare provider, already had predictive systems in place to assess which patients were likely to default on their bill payments. However, these methods did not perform particularly well, in part due to the conservative data pool and lack of depth within existing data.
However, expanding the data pool significantly might create untenably lengthy analysis processes, rendering the exercise impossible to integrate into existing administrative procedures.
Within a few months of patients receiving their first bill, the client discovered that 20% had already fallen into arrears. Although there were delays in processing bills due to medical coding complexity or error, the client wasn’t seeing a correlation between this and patient non-payment. The truth was much more complex.
The client also had trouble identifying which patients had a high risk of defaulting on their bill payments. However, the healthcare provider had a target in mind: to reduce late payments by 25%. This would bring their debt burden back within manageable parameters.
When patients default during an economic downturn following a global pandemic, the “optics” of aggressive debt recovery tactics aren’t on a healthcare provider’s side. Consider one article in Propublica, which notes that “the poor or uninsured often bear the brunt of such actions, said Christi Walsh, clinical director of health care and research policy at Johns Hopkins University.”
Given the potential bad press and moral complexity of pursuing bad debt, it was therefore imperative for the client to find a solution that allowed them to adopt subtler, more empathic outreach tactics.
If the client could identify high-risk patients before they default on payments, they could apply reimbursement plans far more manageable, protecting both revenue and their corporate reputation whilst being kinder to patients in challenging circumstances.
dotData was able to propose an effective data analytic approach. By applying predictive analytics and machine learning to a multifactorial and vast data pool, correlations would be found. dotData’s proposal was to increase the absolute size of the client’s data pool and the number of data sources drawn upon.
These new data sources included patient demographics, physician information, insurance details, and medical specialties. By adding these additional data points, dotData significantly increased the total count of data sources and individual data points, which now numbered in the millions. However, the volume of data remained within the capabilities of dotData’s AI systems.
The intention was to derive clear signals within the data so that accurate propensity-to-pay models could be created and broken down by patient cohort. By doing so, approaches to debt recovery could be tailored more effectively to each patient group.
Whereas previously, such an approach would have been prohibitively slow and complex, dotData was confident that its methods would provide the best opportunity to achieve the client’s ambition of a 25% reduction in overdue payments.
dotData’s Feature Factory was their chosen tool for data analytics due to its ability to run multifactorial pattern recognition on large data pools. The Feature Factory AI studied these higher dimensional datasets and quickly discerned new patterns of data which correlated with overdue payments.
While previous systems might have taken weeks to thoroughly comb through the required millions of data points, Feature Factory managed it within days. This gave dotData’s AI a significant advantage over legacy analytics – it could be repeated on an ongoing basis rather than applied as a “once and done” solution.
Among the important signals in the data were the following patterns:
These two insights, among others, made it easier for the client to derive bespoke advance strategies to contact patients at high risk of late – or non-payment to offer new payment plans or strategies. The revenue protection outcomes were significant, and the client achieved its stated aim of a 25% reduction in defaults.
dotData’s models achieved a 400% increase in accurately identifying patients who required follow-up to avoid falling behind in their payments. High-risk patients could be scored much more accurately in terms of default probability.
The client’s business team was able to integrate dotData’s results into their patient outreach tools and systems. The healthcare organization began proactively calling high-risk patients to discuss their bill repayments before any default occurred. In so doing, they would offer various repayment plans, making it easier for these struggling patients to afford their scheduled payments.
Other patients were identified early enough in their treatment that they could be persuaded to pay in situ during their visits rather than later following invoicing. dotData’s models accurately identified 50% of patients who would not pay in full before the due date of their debt. The client selected the most high-risk 10% for proactive follow-up from those identified patients.
The final effect of these procedural changes was a 25% reduction in overdue payments, constituting a considerable administrative saving. Furthermore, the process could be integrated permanently into the healthcare providers billing procedures.
Introduction Today, we announced the launch of dotData Insight, a new platform that leverages an…
Introduction Time-series modeling is a statistical technique used to analyze and predict the patterns and…
Introduction Time series modeling is one of the most impactful machine learning use cases with…
Introduction Building robust and reliable models in machine learning is of utmost importance for assured…
The past decade has seen rapid adoption of Artificial Intelligence (AI) and Machine Learning (ML)…
The world of enterprise data applications such as Business Intelligence (BI), Machine Learning (ML), and…