- Population health management – and specifically chronic disease management – depend on the ability of providers to identify patients at high risk of developing costly and harmful conditions such as diabetes, heart failure, and chronic kidney disease (CKD).
While basic risk stratification tasks can be performed though non-electronic means, such as patient questionnaires, manual chart reviews, and in-person assessments, the advent of big data analytics has drastically changed the way providers can develop risk scores, monitor patients, and even divide cohorts into extremely narrow subgroups to ensure precision care.
A new study published in the American Journal of Managed Care details how researchers at Carnegie Mellon University used machine learning algorithms to accurately predict future clinical pathways and best-practice treatment decisions for patients with chronic kidney disease. Using multidimensional and longitudinal EHR data, Yiye Zhang, MS, and Rema Padman, PhD, show that it is possible to leverage current sources of clinical big data to deliver standardized but personalized care.
Healthcare providers have long relied on clinical practice guidelines (CPGs) disseminated by professional societies and research groups, and may be hesitant to entrust their patients to the “black box” of algorithmically generated risk scores and clinical decision support suggestions, say Zhang and Padman.
However, “we are now in an era in which clinical interventions need to be personalized and predictive, and so should decision support recommendations,” the authors argue. To meet this objective, it is no longer sufficient to rely on CPGs, often created based on consensus opinions or randomized clinical trials that have strict enrollment criteria.”
“Rather, with the tremendous amount of data being accumulated in EHRs, healthcare service delivery can also benefit greatly from advanced statistical and machine learning models and algorithms that can learn potentially useful insights from large amounts of highly detailed data collected daily, as part of routine care delivered in multiple, diverse settings.”
To illustrate the effectiveness of machine learning for these applications, Zhang and Padman tackled patients with chronic kidney disease. The condition affects more than 26 million Americans, the study says, with a per person cost of more than $23,000 per year. An additional 73 million patients are at increased risk for developing the disease.
The researchers started by collecting EHR data on 664 patients. The data was time-stamped upon generation, allowing the team to organize events into sequences, and the sequences into common clinical treatment pathways.
By examining the timing of certain protocols, the researchers could distinguish between acute episodes, which may contain multiple visits, prescriptions, or tests within a two or three month period, and long-term chronic disease management, which may include follow-up visits at regularly scheduled six-month intervals. Events and test results can be correlated with specific diagnostic codes and laboratory results, which could allow the analysts to understand the effectiveness and relevance of certain treatment decisions, or why a patient may have developed complications.
The strategy helped to identify seven subgroups of chronic kidney disease patients, using laboratory observations and complications as metrics to categorize the participants. Further work will be required to refine the accuracy of these groups, but the paper illustrates how the underlying techniques can be useful for clinical care.
“This approach may facilitate timely extraction of potential new evidence that could become the basis for new clinical trials, and may also serve as ‘shared baselines’ to be used within a local practice for work flow and population health management,” the researchers suggest. “Patient-focused applications derived from our research, particularly those that visualize the clinical pathway and provide related patient-oriented recommendations and educational resources, may enhance patients’ understanding of their diseases and treatments, thus facilitating shared decision making.”
As EHR data becomes more ubiquitous, and data analytics become more adept at extracting actionable insights from the metadata included with entries into the electronic record, machine learning algorithms are likely to become more precise and accurate in their quest to stratify patients and predict adverse events.
“A crucial prerequisite for success in the application of advanced machine learning methods to healthcare delivery is data quality,” the researchers observe. “It is not uncommon for computational scientists to spend significant effort in cleaning EHR data before analysis. In addition, even after months of processing, there are often still missing data and errors, some arising from the mismatch between actual work flows and process assumptions, subjecting the analytical results to bias.”
“Such inefficiency can be minimized by careful observation and understanding of the care delivery context, and planning of the data storage with a range of options available depending on the data size.”
Zhang and Padman note that such risk identification strategies should be tested further with input from practicing clinicians, and may be applicable to other aspects of population health and chronic disease management.
“These methods and broad findings from EHR data are generalizable and can be adapted to other clinical conditions to support efficient review of treatments and outcomes and to aid clinical professionals and patients in making more informed treatment and management decisions,” they conclude.