This blog post is informative in nature and is not related to Caradigm products.
As I wrote about previously, the volume of medical data vastly exceeds the ability of humans to process it without the aid of computers. But through the use of real-time and predictive analytics, that data can become a powerful resource. Real-time and predictive analytics enable organizations to turn their data stores into knowledge and insight with speed and accuracy that no human could ever achieve. And that insight could enable a different action to be taken – potentially improving an outcome, saving money or even saving lives.
I believe we need to think about how a new generation of decision support in the form of real-time analytics can be surfaced for immediate application. And importantly, these analytics should allow users to look at individual situations using a population-based context. Think about a patient who presents with a fever and shortness of breath. Early efforts at clinical decision support may have flashed an alert on the screen that says “consider pneumonia” or “consider ordering a CXR.” In response the clinician enthusiastically (sic) responds, “Duh.”
But imagine a more robust system, where the clinician is provided with the following information: “Of the last 100 patients seen in your hospital with this age group, complaint, and vital signs, 61% had a URI, 32% had pneumonia, 4% had pulmonary embolism, and 4% had other diagnoses. The 1-month mortality was 3.2%. Admission rate was 24%. And by the way, there has been a recent outbreak of atypical pneumonia among patients from Community Long Term Care, so public health requests this patient have a urinary antigen test for Legionella.” Now the physician is able to ask the computerized system, “show me the last 100 patients seen by me with these complaints,” and can then explore the data across her own personal cohort of patients and determine her own admission rate, bounce-back rate, or other parameters. Now decisions can be made for an individual patient using a population-based context.
Emerging real-time analytics have the potential to provide this kind of significant value. The idea is to enable depth of insight across a broad distributed set of data — at a minimum, across a hospital and its many departments and affiliated clinics, but even better across multiple hospitals and multiple outpatient medical practices, where traditionally an observable pattern can be obscured by the sheer number of patients being treated and the episodic nature of medical care today.
Predictive analytics work in a similar way, but with a focus on analyzing data and providing probabilities of various outcomes in the future. In the scenario above, imagine the decision support system provides additional information: “Of the patients with this symptom complex discharged from the emergency department, the predicted 72-hour return rate is 15%. Factors positively correlated with 72-hour return rate include: patient age, pulse oximetry on initial visit < 95%, absence of primary care follow-up visit within 48 hours, and absence of case management and/or visiting nurse within 72 hours.” Based on this type of knowledge, clinicians can further analyze the patient’s situation and make an effort to modify the key risk factors to increase the probability of successful outpatient care. Staff can proactively develop care transition plans and strategies to engage patients, families, case managers and visiting nurses as appropriate for follow-up care.
An organization can also use predictive analytics across populations for multiple purposes. For example, to determine how soon a larger or new facility will be needed to serve a growing patient population. Predictive analytics can be used to inform provider payment structures, reduce fraud, and identify specific future challenges for programs such as Medicare and Medicaid.
So both real-time and predictive analytics take the power of data beyond mere storage, categorization and traditional periodic reporting. Key medical insights can seem like the proverbial needle in a haystack; just increasing the amount of data that can be stored is the equivalent of adding more hay, but real-time and predictive analytics serve as a powerful magnet to expedite the search for the needle.
For these analytics to reach their potential, systems need to bring together all data across the organization and in many cases across multiple organizations, liberating data stores that have been historically siloed in legacy systems. In my next post, I will discuss a surround-and-supplement strategy for upgrading health IT systems so that organizations can take advantage of their legacy data, moving to a new generation of systems that serve future needs at a manageable cost, and without throwing away past investments.
What are your data challenges? What are the questions that you would like your data to more readily answer for you? If you have analytics in place, what problems have you solved through their use — and especially, has your organization improved outcomes? I’d like to hear about it.