Big Data in medicine is the data analysis of large amounts of patient medical information to improve medical care. Analysis of data may be used to pick the best intervention for a patient based upon care outcomes of similar patients or to evaluate physicians.
The primary source of patient medical information is patient medical records, where a medical record reports what happens during an encounter, where an encounter is defined as either an outpatient visit or a hospital stay. Often the data analysis is done after the medical records have been moved to a database where each patient’s identity has been removed, a process called “de-identification.”
A medical record includes the patient’s complaint, past histories of associated medical conditions, tests given, diagnosis and plan of care, where the plan of care identifies interventions such as medications prescribed and procedures. Other medical records record what happens during procedures.
A primary purpose of a medical record is to create a legal document that records what happens during an encounter. Accordingly, medical records for encounters are signed off and can no longer be changed once signed off. An addendum medical record can be issued correcting information in a medical record, but this is not often done.
There are problems with the use of medical records for Big Data:
1. Medical records exist in different medical organizations: A patient may have been seen in many different medical organizations, with some medical information unavailable for analysis.
2. Inaccurate diagnoses and treatments: Medical records contain preliminary and differential diagnoses and do not always identify or reach confirmed diagnoses. Sometimes there are treatments based upon non-confirmed diagnoses. There may be upcoding for financial reasons.
3. Medical information during an encounter often comes from patients and may not match more accurate information in previous medical records: The reliability of information from patients, such as past history of a medical condition and current medications, can be suspect as humans often have bad memories and patients may not be that knowledgeable about medical care.
4. Missing information: Physicians may not collect the necessary biomarker information to do a data analysis.
5. Difficulty in identifying outcomes: Outcomes of an intervention may not be recorded yet, or it may be unclear that an outcome is related to an intervention.
To provide medical information in addition to medical records for Big Data, I propose that there be longitudinal histories for significant medical conditions for patients, for short longitudinal histories. A longitudinal history for a medical condition would identify follow-on events to the initial medical condition, with these events being follow-on medical conditions and interventions.
For example, for a severe knee fracture, there could be follow-up surgery, pain and arthritis, opioid abuse, a knee replacement and a follow-up knee replacement. For each event would be the date of event, patient age, relevant biomarkers, associated encounters and for each encounter associated medical records.
A longitudinal history would be created within electronic medical record (EMR) systems used by physicians seeing the patients. Unlike for medical records, physicians can update and correct longitudinal histories, and patients, along with a professional in medicine, can audit and suggest changes to this information. Note that for longitudinal histories to be possible, physicians must work differently than they do today, initiating, updating and correcting longitudinal histories.
For interventions and medical conditions, medical research would identify biomarkers to collect that can help predict follow-on medical conditions. For example, a cataract is detected in a patient, and the patient will have cataract surgery. Potential bad follow-on medical conditions to the cataract surgery that could result are a capsular rupture or a detached retina. Before the cataract surgery, biomarkers that can be used to predict these two events should be collected by the ophthalmologist.
A physician providing care to a patient would identify if a new medical condition or intervention is related to the initial or follow-on medical condition and if so, add it as an event in the longitudinal history.
In some cases, based upon medical research, the system could do this association; for example, if a detached retina occurred after cataract surgery, then the retinal detachment would be added as a possible follow-on medical condition for the longitudinal cataract history. The system would be able to identify the probability that the retinal detachment occurred as a result of the cataract surgery or independently, based upon information from similar patients, comparing probabilities of a retinal detachment for similar patients who do and do not have a preceding cataract surgery.
The physician performing any procedure could be identified, which is necessary to evaluate physicians in doing a procedure.
Other information that could be collected over the life of the longitudinal history could be disability measurements related to the initial medical condition, measurements identifying any disability of the patient.
One such disability measure is EQ-5D that develops a disability measure from 0.0 (total disability) to 1.0 (no disability) based upon 5 separate measures: mobility, self-care, usual activities, pain/discomfort and anxiety/depression.
Given longitudinal histories for many patients, the following can be determined:
- Predicting the typical results of an intervention for a given patient.
- Predicting the probability of a future outcome for a given patient.
- Selecting the best intervention for the given patient.
- Evaluating physicians based upon outcomes of interventions (such as evaluation of ophthalmologists doing cataract surgeries).
- Ad hoc studies
Besides providing additional medical information for Big Data, longitudinal histories additionally enhance patient care. They likely produce more accurate histories of a patient’s medical conditions than histories in encounters, which now commonly come from the patient.
For a longitudinal history for a patient’s medical condition to be possible, there must be a way to present such a history to all physicians seeing the patient.
Michael R. McGuire is the author of A Blueprint for Medicine.
Image credit: Shutterstock.com