MAIDIS BLOG

How Digital Transformation Can Improve Hospitals’ Operational Decisions

by Song-Hee Kim and Hummy Song

Many companies are interested in digital transformation — using digital technologies to create or modify business processes, culture, and customer experiences — to grow and stay ahead of the competition, and hospitals are no exception.

When people think about digital transformation in health care, they tend to focus on the idea of employing analytics to improve clinical decision-making. For example, with advances in computational science and machine learning, it’s become possible to deliver precision medicine, where therapies and interventions are tailored to each patient based on the individual’s genetic profile. Artificial intelligence (AI) algorithms are increasingly being used to improve the visual detection of signs of disease in fields such as radiology, dermatology, gastroenterology, ophthalmology, and pathology.

However, focusing on leveraging digital transformation solely to improve clinical decision making would be a mistake. Based on our research and that of others as well as the burgeoning advances in how hospitals are using data and technology, we believe that digital transformation has a substantial role to play in optimizing hospitals’ operational decision-making, which in turn can lead to improvements in the quality and efficiency of care and patients’ access to it.

Here are four key areas where hospitals can leverage digital transformation to improve operational decision-making: patient flow, staffing, scheduling, and supply chain management.

Patient Flow

As hospitals strive to provide the right care to the right patient at the right time, care providers need to do two things: evaluate patients’ needs accurately and manage hospital resources effectively. While providers are well trained to do the former, they’re usually not trained for the latter, which is a challenging task — especially given the strain on hospital capacity that’s all too common these days due to the pandemic.

At the hospital level, data-driven operational decision-support systems can provide valuable insights to aid in making these triage, admission, and discharge decisions. For example, when a patient arrives and the provider is unsure whether the patient should be sent to the ICU or a general ward, a decision-support algorithm can provide recommendations based on the predicted benefit of ICU admission for that particular patient. Research using patient-level operational data from more than 190,000 hospitalizations across 15 U.S. hospitals shows that when patients who had a clinical need for admission to the ICU are instead admitted to another part of the hospital (e.g., a general ward), this results in longer hospital stays and higher readmission rates.

When the capacity of the desired ICU is constrained, the provider may consider different options such as placing the patient in another unit (e.g., a surgical ICU instead of a medical ICU) or discharging patients who are currently in the ICU to make room for the new ones. Research using hospital operational data shows that both strategies have important tradeoffs and unintended consequences that should be accounted for. Decision-support algorithms can be designed to incorporate these tradeoffs, weigh the costs and benefits of the different choices, and provide appropriate recommendations.

Going beyond recommendations, algorithms can be leveraged to automate operational tasks. Research findings from a series of experiments where physicians and Amazon Mechanical Turk workers were asked to manage a simulated hospital unit shows that behavioral biases and cognition-driven decision errors may influence providers’ operational decisions. Decomposing these decisions into clinical and operational components and using algorithms to automate the operational component may ultimately lead to better outcomes.

At the ward level, machine learning and decision-support algorithms can also be used to predict the expected number of admissions, discharges, and transfers to and from the ward, which in turn can guide subsequent actions based on these predictions. This can facilitate the bed turnover process, leading to improved patient flow and reduced length of stay. The predictions for individual wards can serve as inputs to a hospital-wide bed management dashboard, which can be used not only to display the current status of each ward but also to provide predictions for the expected future status throughout the hospital.

For example, the Beth Israel Deaconess Medical Center in Boston, in collaboration with a team of operations researchers from MIT, has implemented prediction-informed dashboards to support admission and transfer decisions by displaying each ward’s current census as well as projected number of discharges. Similarly, Boston Children’s Hospital uses the Predictor of Patient Placement System, which allows the emergency department to know which patients are likely to be admitted to the hospital and to which ward. Hospital-wide bed management dashboards enable better planning and enhanced communication across the different wards and can be further developed to provide automated alerts about the system, such as when the average wait time for a new bed exceeds a predetermined threshold.

Staffing

Digital technologies can also help with the supply side when it comes to better managing capacity. Take, for example, nurse staffing, which accounts for a significant proportion of hospitals’ costs. Instead of relying on phone calls, text messages, and spreadsheets to make ad-hoc staffing decisions that often change at the very last minute, charge nurses and hospital administrators can utilize analytics to improve this process.

For example, algorithms can predict nurse absenteeism rates and the need for surge staffing to preemptively determine the right number of float nurses to call in. Research in emergency department operations shows that both can be modeled, even in environments where demand is highly uncertain. A key advantage is the ability of these systems to preempt and respond more quickly, which in turn can improve the consistency and predictability of the work schedule for nurses. This aspect is likely to be important as hospitals and other health care delivery organizations work on reducing notoriously high nurse-turnover rates: Research examining nursing turnover in one of the largest home health agencies in the United States shows that employer-driven inconsistency in workers’ schedules increases workers’ likelihood of quitting.

Analytics can also be leveraged to optimize team staffing. Hospitals rely on providers to work together effectively as a team, with team members spanning different roles and levels of experience. Research shows that the composition of care teams has a significant impact on performance. A study of emergency department teams collectively conducting more than 111,000 patient visits over the course of two years reveals that the differences in hierarchy and skill across attending physicians, nurses, and resident physicians lead to varying effects of being exposed to new team members when it comes to team performance.

Another study of cardiac surgery teams conducting more than 6,000 surgeries over seven years shows that it is important to account for the pairwise familiarity among team members — the number of past collaborations for all pairs within the team — because it has significant implications for team productivity. While it is nearly impossible to incorporate these takeaways when trying to staff teams manually, AI can easily incorporate these research insights to determine the optimal team composition of providers scheduled to work and provide recommendations on optimal staffing levels.

Scheduling

While many hospitals have moved to electronically capturing and storing patient records, the scheduling of various resources is still largely a manual process. This applies to the scheduling of surgical procedures in operating rooms, scans in radiology suites, and many others. This is another area where digital technologies can bring substantial improvements — not only by better predicting resource needs and effortlessly incorporating last-minute changes and cancelations but also by optimizing schedules based on the latest research.

For example, machine-learning algorithms can be used to better predict the duration of each procedure such as the length of a surgery or an MRI. At the Beth Israel Deaconess Medical Center, tools developed by Amazon are being used to book operating room times more precisely.

The expected duration is a function of not only patient characteristics and their clinical needs but also various operational factors. For example, researchers find that surgical procedure times tend to increase as a function of larger team sizes, higher workloads, and the sequence of the operation in the operating room. Algorithms are better equipped than humans to account for the effects of such operational factors in making predictions.

Machine learning can also be used to predict the required time that each patient should spend in the post-anesthesia care unit (PACU) following a surgery. Since PACU congestion often leads to delays in the operating room, this is another place where analytics can be used. For example, this study leverages analytics to optimally sequence surgical procedures to help prevent PACU congestion and minimize operating room delays.

Supply Chain Management In the United States, hospitals spent an average of $11.9 million each on medical and surgical supplies in 2018, accounting for up to one third of total operating expenses at some. Despite this, improving supply chain and inventory management is often not considered a high priority for hospitals, where providers tend to focus more on the processes surrounding direct patient care. Yet, having these supplies is necessary for delivering high-quality care.

Across many industries, digitally transforming the supply chain has been shown to reduce process costs by 50% and increase revenue by 20%; hospitals are no exception. By automating the process of collecting data, ordering, reconciling, and paying for medical, surgical, and pharmaceutical supplies, hospitals can reduce supply chain and inventory management-related costs. Due to the Covid-19 pandemic, improving agility and resilience to demand and supply-side shocks has become even more critical, and hospital managers are increasingly looking for ways to leverage data and technology to gain insight into inventory, pricing, lead times, and demand trends.

Radio-frequency identification (RFID) technologies and internet-connected trackers can be used to better track and locate supplies in real-time. For example, Mayo Clinic’s Saint Marys Hospital rolled out an RFID system for their emergency room operations in 2015, which led to improved care and patient experience as well as lower costs.

Pooling and coordinating supplies across different departments within a hospital can significantly reduce the amount of inventory required to meet a given service level. While physical centralization is one way to achieve this, information centralization, which can be easily achieved with a digitized supply-chain-management system, may be sufficient to reap the same benefits.

To make this type of digital transformation possible, hospitals must be intentional in the way they collect data and interact with their information technology systems. We have three prescriptions for how to go about this.

1. Collect the right data in the right format.

Start by identifying the pain points and the low-hanging fruit. When and where is data still collected and communicated offline? Can the fax become automatically captured and recorded in the electronic medical record (EMR) system? Can phone calls and text messages be reduced and replaced by electronic communications via the EMR?

When capturing data, be sure to capture operational characteristics in addition to clinical factors. Timestamps are a rich source of data that offer insight into hospital operations. Timestamps should be captured both when events occur (e.g., a bed is assigned to a patient, test results become available, or a patient is discharged) and when resources are requested (e.g., a bed is requested, a consult is requested, a test is ordered). Keeping track of the latter allows managers to understand the underlying demand for resources even if not all of the demands could be met, which allows for better planning for the future.

In addition to timestamps, be sure to also keep an accurate inventory of resources that gets updated in real time. Resources include not only medical, surgical, and pharmaceutical supplies that are ordered on a regular basis, but also beds, large equipment, and staff.

2. Set yourself up for scalability and interoperability.

From the outset, design the data-collection system with scalability and interoperability (the ability of different IT systems or equipment to exchange and make use of data) in mind. Standardize the input formats to minimize (or eliminate) the need for data cleaning and to enhance the quality of inputs into algorithms. Familiarize yourself with the four levels of interoperability and the established interoperability standards to set up a system that will facilitate health information exchange and data sharing.

Ultimately, having a uniform baseline data architecture and a standardized data format will allow for easier implementation and replicability of algorithmic tools across hospitals. In the United States, the Centers for Medicare & Medicaid Services (CMS) and many health care delivery organizations are looking to adopt Fast Healthcare Interoperability Resources (FHIR) standards.

3. Don’t lose sight of the human-algorithm interaction.

While algorithms can produce helpful predictions and recommendations, ultimately the decision-maker is the human. As a result, we must be cognizant of the widespread nature of algorithm aversion by decision-makers and aim to develop algorithms that are fair, explainable, prevent harm, and respect human autonomy so that the decision-maker can trust the algorithms. Furthermore, creating superb algorithms alone cannot improve hospital operations. Algorithms need to be carefully designed, implemented, and evaluated with the user in mind.

It’s also important to remember that health care is a knowledge-intensive industry. Care providers often possess a significant amount of local knowledge or expertise that algorithms fail to capture. Completely replacing human decision-makers by algorithms may not be the solution because incorporating human judgement and experience can often enhance the performance of algorithms.

Given the aging population, prevalence of chronic conditions, and advances in medicine, it has become more important than ever for hospitals to operate efficiently and effectively. Going forward, the key to improving operational decision-making will lie in their ability to leverage digital transformation.