Building a Successful Data Culture in the Post-Pandemic Healthcare Reality
Postponement of nonessential surgical procedures early in the coronavirus pandemic not only disrupted surgical care at U.S. hospitals, but also took away a large portion of hospitals’ total income, according to two recent studies. The University of Pennsylvania and Children’s Hospital of Philadelphia findings were presented at the virtual American College of Surgeons (ACS) Clinical Congress 2021. Nationwide, hospitals lost $1.53 billion from missed elective pediatric procedures alone, during the first three months of the pandemic — March to May in 2020. Thus, involvement of RCM BPO expertise is critical to fast-track revenue collections and enable providers to do what they do best: provide safe and effective care to patients.
Challenges in data science deployments
A machine-learning (ML) system is not just code; it is the combination of code, data, parameters, and training environment. Today, most healthcare organizations are looking at the potential of using AI/ML in helping achieving business outcomes and, despite significant investments, data science fails to make the desired impact. Key reasons include:
- Data science literacy at senior levels — Often, data science models do not survive the proof-of-concept (POC) stage, and they are sidelined due to a lack of fundamental data literacy at senior levels of the organization.
- Data findings conflict with intended course of action — At times, if data findings conflict with intended course of action usually derived based on past experience and gut feel, that could jeopardize the adoption of data science recommendations.
- Data availability, quality, and preparation issues — Collection of the required data is a challenging task. Data exists in both structured and unstructured format and is stored in various places with unique security and privacy issues. Data cleaning is a primary task before any form of analysis. Unstructured data or unformatted data, which may take most of the time for data cleaning, can be a reason for losing motivation. Insufficient data that is available for the analysis can also be a factor for failed AI/ML projects.
- Traditional RCM metrics are usually “lag indicators” — The primary KPIs by which practices measure revenue cycle performance are largely retrospective in nature: net collections, days in A/R, denial rates, and cost to collect, to name a few.
- Availability of skill-set — There is a profound lack of deep technical expertise and business acumen by data scientists in areas other than data science. You rarely find a data scientist who can take a step further to understand how a java backend application works and exchanges data. Strong business acumen (i.e., understanding of the provider RCM) is also critical to developing successful data science POCs.
- Analysis — Paralysis, challenges in building a compelling data story — There is a tendency to pick a large scope for deploying analytics and getting lost in the details. A lack of a clear data story telling leads to a vague explanation of analysis and recommendations. This delays / hampers the adoption of data culture in the organization.
Building a data culture
To arrive at the right business decisions, it’s critical that you start with the right business questions. Some pointers that could help build a successful data culture are as follows:
- Investing in data literacy across the organization. Specialized training should be offered just in time.
- Data-driven culture building. It starts at the very top. Include data in the decision-making process. Get in the habit of explaining analytical choices.
- Fix basic data-access issues quickly.
- Choose metrics with care. Move from retrospective to real-time reporting and action.
- Build a data science center of excellence (CoE). Invest in building a team with deep business acumen and data science skill sets.
- Make proofs of concept simple and robust, not fancy and brittle. Be willing to trade flexibility for consistency — at least in the short term.
Recent wins in data science deployments
Sagility has been able to build successful ML models to determine propensity of claim resolution and payments and optimize resolution efforts by identifying patterns of claims that could result in non-cash resolution. As a result of the deployment of Propensity-to-Pay Analytics Model, the Sagility team has delivered these breakthrough results for one leading national health system and longstanding client partner, for which Sagility manages multiple service lines, including Physician, Credit Balances, Payment Posting, and Day 1 A/R.
- Sagility helped the client save $.4 million, for a 15% increase in the cash collections on claims within 90 days from the date of placement.
- The client had benchmarked cash collections 0 – 90 days from the claim placement date at 25%. Sagility is now operating at 28.5%, which is well above the benchmark.
The potential for big data analytics / data science is huge in the future. Machine learning can also process the information much faster with its accelerated learning and advanced capabilities. Based on this, the time required for solving complex problems is significantly reduced. However, data is most useful when everyone has the ability to explore it, both individually and collaboratively, with other team members. The democratization of data and collaborative analytics are the future of business intelligence.