● Roundup: Analytics
point care teams in that direction. We can identify and target programs toward those with treatable conditions and better manage their utiliza- tion of services. T is is a perfect example of why predictive analytics isn’t nearly as suitable for most health systems as risk stratifi cation is. Our doctors are too busy taking care of patients who are already sick to guess who’s going to get sick tomorrow.
Anil Jain, M.D., FACP, Senior Vice President & Chief Medical Information Offi cer, Explorys Inc.
Patients are indeed unique individuals and certainly physiology,
environment, genetics and habits vary between them. However, what analyses of very large data sets have uncovered is that by looking statistically at the information a prediction model can be developed. Key pieces of information that impact a model may include patient demographic factors (age, gender), socioeconomic status (median household income, education), geographic factors, clinical diseases (diabetes, high blood pressure, cancer, etc.), habits (tobacco use, alcohol use, etc.), biometrics (height, weight, blood pressure, etc.), family history, surgical history and laboratory results (kidney func- tion, blood count, cholesterol, blood chemistry, etc.). For example, most health systems are attempting to reduce hospital readmissions (admissions to the hospital within 30 days of being discharged) to help improve quality while reducing cost. Explorys has developed a heart-failure hospital readmission model that uses more than 100 pieces of information to predict the chance that a patient will be
“readmitted” to the hospital within 30 days, allowing health systems care coordinators to intervene or the patient’s physicians to intensify treatment for their condition.
Dan Riskin, M.D., CEO, Health Fidelity Since each patient is diff erent and the population is becoming
more, rather than less complex, it’s not enough to build our analyt- ics systems off a weak claims data infrastructure. For example, the claims and EHR discrete data infrastructure will typically represent a complex patient as “hypertensive” and “diabetic.” T at type of record may have worked in 1993, but it will not work in 2013. T e patient must be represented with his or her full complexity
to support meaningful outcome prediction and quality measurement. T is means extracting the full clinical content from the EHR, putting it into a data warehouse and leveraging the discrete data elements as well as the narrative data elements. Only in this way can the “hy- pertensive diabetic” properly be recognized as “83 years old,” “living alone,” “smoking,” “well-controlled hypertension,” “poorly controlled diabetes mellitus” along with the hundred other features that represent clinical and social breadth for that individual patient. With this level of information, we will be able to predict outcomes better, measure trends, measure quality and drive improvement.
Eric Mueller, Director, Product Management, Lumeris
It is true that every patient is physiologically diff erent, but that does not negate the usefulness of individual patient data for predic- tive modeling. Biometric information such as blood pressure, BMI, blood sugar and LDL cholesterol levels combined with cost, quality and utilization information can contribute to a complete view of a patient’s healthcare history. Robust information at the patient level can contribute to a more complete view at the population level and ultimately lead to better care at the patient level. For example, after collecting all the necessary data, advanced ana-
lytics and reporting are used to help health plans identify trends – like high-utilizing patients – and correlate that back to specifi c conditions. If a health plan notices an increase in asthmatic EMR visits, they can use that information to implement a care management plan where physicians and their care teams complete assessments and care plans for each patient, educate the patients about their chronic conditions, stress the need to adhere to symptom response plans and schedule regular checkups. T e overall result: asthmatic ER utilization will decrease, and associated hospital admissions and readmissions will
16 January 2014
decrease as well. Because the health plan identifi ed a trend in their population, they could provide high-quality care for their individual asthmatic patients.
Bonnie Cassidy, Senior Director of HIM Innovation, Nuance
Medicine has never been a “one-size-fi ts-all” fi eld, but there are key pieces of patient data that can provide clear insight into those populations physicians are treating. Accurately capturing genomic information and data around the most frequent diagnoses and procedures will enable provider organizations to identify and better understand their patient populations’ unique needs and perhaps even the epidemiological factors of a community. T is is especially true for larger health organizations that have multiple hospitals or outpatient facilities; diff erent locations may not have the same population and, as a result, face unique challenges. Once health organizations better understand their patient popula- tions, they can begin to develop strategies that target specifi c areas of need in their communities. For instance, if clinical data shows a certain population has high occurrence and incidence of lung cancer, providers can off er educational opportunities and smoking cessation programs, while payers can off er wellness incentive programs. T is integrated strategy is driven by that initial data analysis on lung cancer.
Tony Jones, M.D., CMO, Philips Healthcare’s Patient Care and Clinical Informatics’ Business Unit
Predictive analytics can help drive down costs a couple of diff erent
ways. First, analyzing data for a specifi c patient population may help providers make faster treatment decisions at the point of care, which could mean fewer tests needed for a diagnosis and therefore, lower costs. Second, predicative analytics give hospitals a much stronger ability
to develop preventative and longer-term services customized for their patients. Aggregating retrospective and real-time clinical data can pres- ent a picture of the patient population a hospital provides care to, and can enable that hospital to design care that is catered to that patient population both now and in the future. In this case, money may not be directly saved, but it may be allocated more appropriately. Healthcare is already a data-rich environment. T e challenge is that much of that data (e.g., heart rate, ECG tracings, blood pressure, etc.) is displayed on monitors but not stored for real-time or future analysis. As a result, the ability to detect patterns is diminished through the loss of this valuable data. Capturing and storing this data, then combining it with radiology images, labs and patient history, provides a much richer data set and increases the likelihood of detecting meaningful patterns. T is forms the basis of predictive analytics and enables providers to treat minor medical conditions before they become major, expensive ones.
Karen Handmaker, MPP, Vice President, Population Health Strategies, Phytel
Care teams need current and trended information on every patient
to act eff ectively. Key pieces of information include chronic conditions and related lab results, BMI, preventive screenings, medication compli- ance and recent hospitalizations and ER visits. Using evidence-based protocols and comparative benchmarks, “analytics” can fl ag trends and results for each patient that help care team members identify patients who need immediate attention and determine what action to take. With access to the right information, practices can make progress on overall quality and cost performance for their populations.
HMT: How can administrators and clinicians alike identify and collect the right data they need on which to base decisions?
Todd Rothenhaus, Chief Medical Information Offi cer, athenahealth T ere is immense criticism of claims-based data analysis. But claims data represents the skeleton of health outcome analysis, and clinical data
HEALTH MANAGEMENT TECHNOLOGY www.healthmgttech.com
| Page 2
| Page 3
| Page 4
| Page 5
| Page 6
| Page 7
| Page 8
| Page 9
| Page 10
| Page 11
| Page 12
| Page 13
| Page 14
| Page 15
| Page 16
| Page 17
| Page 18
| Page 19
| Page 20
| Page 21
| Page 22
| Page 23
| Page 24
| Page 25
| Page 26
| Page 27
| Page 28