AI News, Sign Up Successful – Please Check Your Inbox
- On Sunday, September 30, 2018
- By Read More
Sign Up Successful – Please Check Your Inbox
We believe the key to unlocking the full potential of big data in healthcare lies at the intersection of data science and medical science.
The Lumiata Medical Graph analyzes hundreds of healthcare data sets within the context of clinical practice and the world’s medical knowledge, and maps out insights on current and future health trajectories of individuals.
Using Risk Scores, Stratification for Population Health Management
Population health management requires providers to maintain a delicate balance between taking a long view of generalized patient trends and focusing personal attention on the individual and the distinctive circumstances that will influence her journey towards better health.
By using big data and large sample sizes to better understand patterns of what is likely to happen to individuals, organizations can develop insights into how each unique patient is progressing along common disease trajectories and plan their interventions accordingly.
However, successful population health initiatives have a few commonalities, including a firm grasp on the basics of big data analytics, information governance, and how to formulate and leverage the risk scoring and risk stratification techniques that help providers organize the complex needs of their patients.
Organizations can develop risk scores by examining large cohorts of patients with similar characteristics, extracting key clinical and lifestyle indicators from those cases, and using algorithms to chart how those factors influence ultimate outcomes.
risk score may indicate the likelihood of a single event, such as a hospital readmission within the next six months, while a risk stratification framework may combine several individual risk scores to create a broader profile of a patient and his or her complex, ongoing needs.
Healthcare payers and providers both use risk scores to estimate costs, target interventions, gauge a patient’s health literacy and lifestyle choices, and try to prevent patients from developing more serious conditions that could result in higher spending and worse outcomes.
Using accessible metrics such as age, diabetes status, smoking history, cholesterol levels, and body mass index (BMI), patients can easily compute their CVD risk scores themselves through a number of freely available online calculators.
In addition to changing how providers view heart disease, the theory of predictive analytics based on patient characteristics now underpins the majority of systemic reform efforts and is making the transition to value-based reimbursement a feasible task.
All three require the ability to stratify patients by risk in order to identify and address high-priority issues that impact larger groups of patients, forestall or avoid costly events, and ensure that individual needs are met in a timely and efficient manner.
For providers participating in value-based care arrangements, which pair financial risk with clinical outcomes, success in these key areas can help to avoid penalties for quality and stay on the positive side of shared savings or bonus payments.
This task requires a familiarity with basic data science techniques, including how to identify relevant data sources, how to extract and normalize this data, and how to create a methodology for synthesizing information into a standardized risk score.
When choosing a population health management package, KLAS recommends that providers ensure the offering can meet identified core needs, gather and analyze data from disparate sources with few aggregation problems, and provide cost-effective and workflow-friendly results that will help to achieve a return on the investment.
Regardless of whether an organization is looking to purchase a vendor solution or develop their population health management tools in-house, providers must have a firm grasp on what data is accessible, how clean, complete, and accurate it is, and what it can tell clinicians about the patient story.
At Community Care North Carolina, a state-wide care management initiative covering the majority of Medicaid patients in the region, claims data combines with a number of other information sources to give more than 1800 primary care practices increased insight into patient risk and population health.
“We are also using laboratory data and state vital statistics data to build a shared utility for the whole state to use in ways that target patients more intelligently for specificcare management interventionsand speed up the cycle of quality improvement.
However, while EHRs can be a powerful tool for predicting a specific outcome for an individual, they can fall short of meeting the larger population health management needs of accountable care organizations or providers in the same local area where each member is using a different electronic health record system.
AAMC also suggests that providers look to take advantage of additional clinical data sources to feed risk analytics, including local disease registries, pharmacy data, lab data, and real-time data from home monitoring devices, wearables and Internet of Things tools, or bedside monitoring equipment in the hospital.
Environmental factors, including socioeconomic circumstances, behavioral health, and patient personality and lifestyle choices, produce significant impacts on long-term outcomes, and are necessary components of a risk stratification framework that will provide rich, comprehensive, and holistic predictions into patient health.
In Brownsville, Texas, for example, community health worker supplement clinical care while also collecting and reporting on environmental data that would otherwise be inaccessible/ These community advocates deliver data on the availability of public transportation, housing instability, and religious practices while connecting to patients in their own environments with cultural sensitivity and a personal touch.
Creating risk scores that include a blend of social, behavioral, and clinical data will help to give providers the actionable, 360-degree insight necessary to identify patients in need of proactive, preventive care while meeting value-based reimbursement requirements andimprovingoutcomes across the care continuum.
What is Individualized Health?
Individualized health is a scientific approach to health promotion and disease management that acknowledges that people vary in their circumstances, preferences, and in their optimal path to full health.
no two individuals react alike and behave alike under the abnormal conditions which we know as disease.”* A primary goal of individualized health is that each person’s health decisions are fully informed by scientific evidence specific to their condition so that better outcomes are achieved at more affordable costs.
The idea is to reference each individual patient against a population of otherwise similar patients to answer clinical questions about that patient’s health state, disease trajectory, and likely response to treatment.
In prostate cancer evaluation, for example, a new patient’s risk of having an aggressive tumor can be predicted by comparing him to a subset of patients with similar age, PSA trajectory, prostate density, and previous biopsy results for whom prostatectomy revealed whether the tumor was aggressive or not.
For example, to reliably and accurately predict if a certain patient is likely to be diagnosed with an aggressive form of breast cancer, a woman’s genetics, proteomics profile, family history, lifestyle choices, and demographic characteristics should be jointly considered.
This is partially because it is burdensome and costly to collect all of the necessary information on study participants and partially because it requires sophisticated statistical models that can incorporate diverse information and produce accurate, consistent, and interpretable results.
The successful implementation of learning health systems requires the extraction of data from clinical notes, the retrieval of research quality data from electronic health records, the acquisition of patient consent, and the protection of patient privacy, all of which require technical and ethical solutions.
These tools can come in many forms, including apps, wearable devices, electronic medical record dashboards, and statistical software packages that can be used by both clinicians and researchers.
By creating better ways to measure health and disease, researchers aim to gain a more insight into how people experience disease, leading to the development of more effective health interventions.
Through enhanced integration and analysis of data and the development of better measurements, individualized health researchers are creating a foundation on which to develop tools that will ultimately improve health-related decision-making.
The tools created by researchers are diverse, ranging from risk prediction scores and statistical software to apps, wearable devices, and electronic medical record dashboards.
Creating and using software that enables interoperability and integration among the various electronic health record systems is an important step in providing researchers a more comprehensive view of patients’
major challenge facing individualized health is ensuring that the vast array of tools being developed to improve health are accessible to clinicians, patients, and public health professionals in a timely and effective manner.
To facilitate this, clinicians can collaborate with researchers to identify pressing medical needs, aid in the design of tool interfaces, and consult on the integration of tools into the clinical workflow.
Clinicians can also promote the availability of high-quality data by working with researchers to optimize electronic health records for research purposes and by encouraging their patients to participate in research studies.
Public health practitioners Public health practitioners can collaborate with communities and individuals to prioritize public health needs and determine which of these priorities are amenable to a tailored, individualized health approach.
Researchers should also collaborate extensively with clinicians, public health practitioners, and patients to identify the most important needs that can be addressed by individualized health and ensure that the tools they create are relevant and practical.
Penn Signals is a collaborative data science platform developed by the Penn Medicine data science team that combines clinical data at scale with big data to allow researchers to explore solutions, allow developers to develop predictive applications, and providea platform for deployment.
According to the Centers for Disease Control (CDC), sepsis (blood infection) affects more than a million Americans annually and is the ninth leading cause of disease-related deaths and the #1 cause of deaths in intensive care units.1 Worldwide incidents exceed 20 million cases a year, and mortality due to septicshock may approach 50 percent, even in industrialized countries.2 The mortality rate is approximately 40 percent in adults and 25 percent in children.3 Treatment guidelines call for the administration of broad-spectrum antibiotics within the first hourfollowing recognition of septic shock.
Prompt antimicrobial therapy is critically important, as the risk of dying increases by approximately 10 percent for every hour of delay in receiving antibiotics.4 Traditional methods of sepsis identification generally only detect about half of the cases, and even thendetection typically occurs just two hours before the patient succumbs to septic shock.
Heart failure (the inability of the heart to pump enough blood to meet the needs of the body and lungs) is amazingly common, affecting 5.8 million people in the United States5 and about two percent of adults worldwide.6 It is the number one reason that people in the developed world—especially those over 65—are admitted to the hospital.7 Even though treatment itself is relatively inexpensive and typically involves lifestyle modifications(reduced smoking, physical exercise, dietary changes), as well as medications, treatment costs are high,mostly because of hospitalization, with estimates exceeding $35 billion per year in the United States.8 Exacerbating those costs is the fact that a quarter of patients hospitalized in the United States are readmitted within 30 days, and half seek readmission within six months after treatment.9 Penn Medicine deduced that an improved algorithm that could correctly identify patients and get them on the right regimen earlier could reduce treatment costs, reduce hospital readmissions, and improve patient care.
For the sepsis pilot, which involved 150 of Penn Medicine’s 1200 to 1500 total patients, Penn Medicine was able to correctly identify about 85 percent of sepsis cases (up from 50 percent), and made such identifications as much as 30 hours before onset of septic shock (as opposed to just two hours prior, using traditional identification methods).
From Insight to Action Once the data science team was able to produce reliable predictions, their next challenge was to figure out how to deploy this in real time so that a care provider at the point of care can get information when it’s most meaningful in a way that they can act on it.
The combined clinical and data science team created a dashboard that reports how the care team is acting in the care pathway, how to reduce readmission rates, how toimprove quality of life for patients, etc.
They want to take the best solutions that depend on open source technology, that depend on big data techniques, and deploy them in an infrastructure that has tohave 24/7 reliability and in a way that doesn’t take months and months to do.
To do this, Penn Medicine is excited about the potential of using TAP to allow clinicians and data scientists to have a development, test, and QA environment, where they can quickly build, quickly explore, and quickly deploy new predictive analytics applications.
The goal of this project was to improve Penn Medicine’s ability to predict patients who are at risk of being re-admitted to the hospital for heart failure in the next 30 or 90 days of discharge by using a patient’s medication history as an added predictor.
Summary According to a recent Gartner Group study of nearly 300 respondents, 65 percent cited their inability to identify the value of a big data system as the single greatest barrier that prevents them from adopting their own big data solutions.10 This implies a need for education and some guidance to help others make that first step, or at least prove the utility of such a deployment.
Race, Ethnicity, and Language Data: Standardization for Health Care Quality Improvement
While a range of health and health care entities collect data, the data do not flow among these entities in a cohesive or standardized way.
Although hospitals, community health centers (CHCs), physician practices, health plans, and local, state, and federal agencies can all play key roles by incorporating race, ethnicity, and language data into existing data collection and quality reporting efforts, each faces opportunities and challenges in attempting to achieve this objective.
Health care involves a diverse set of public and private data collection systems, including health surveys, administrative enrollment and billing records, and medical records, used by various entities, including hospitals, CHCs, physicians, and health plans.
No one of the entities in Figure 5-1 has the capability by itself to gather data on race, ethnicity, and language for the entire population of patients, nor does any single entity currently collect all health data on individual patients.
Thus there is a need for better integration and sharing of race, ethnicity, and language data within and across health care entities and even (in the absence of suitable information technology [IT] processes) within a single entity.
Health information technology (Health IT) may have the potential to improve the collection and exchange of self-reported race, ethnicity, and language data, as these data could be included, for example, in an individual's personal health record (PHR) and then utilized in electronic health record (EHR) and other data systems.1 There is little reliable evidence, though, on the adoption rates of EHRs (Jha et al., 2009).
Until data are better integrated across entities, some redundancy will remain in the collection of race, ethnicity, and language data from patients and enrollees, and equivalently stratified data will remain unavailable for comparison purposes unless entities adopt a nationally standardized approach.
Because hospitals tend to have information systems for data collection and reporting, staff who are used to collecting registration and admissions data, and an organizational culture that is familiar with the tools of quality improvement, they are relatively well positioned to collect patients' demographic data.
Thus, while hospitals are an important component of the health care system and represent a major percentage of health care expenditures, they are only one element of the system for collecting and reporting race, ethnicity, and language data.
A 2006 National Public Health and Hospitals Institute (NPHHI) survey asked hospitals that collected race and ethnicity data whether they used the data to assess and compare quality of care, utilization of health services, health outcomes, or patient satisfaction across their different patient populations.
Systems changes can involve training a large number (possibly hundreds) of hospital registration/admission staff (many of whom may be off site) and modifying practice management and EHR systems to ensure that proper and consistent data fields are in place across multiple departments and units that serve as patient entry points.
The ten hospitals in the collaborative initially cited the data collection requirement as one of the greatest challenges of the program, yet once they focused their efforts on these goals, they were able to bring together key stakeholders within each institution, implement needed IT changes, and train staff.
CHCs are front-line providers of care for underserved and disadvantaged groups (Taylor, 2004) and therefore are good settings for implementing quality improvement strategies aimed at reducing racial and ethnic disparities in care.
Yet while CHCs serve diverse patient populations and, as organizations, understand the importance of demographic data for improving the quality of care, the accuracy of the race, ethnicity, and language data they collect may be limited (Maizlish and Herrera, 2006).
less is known, however, about the extent to which CHCs consistently collect patient race and ethnicity data beyond the basic Office of Management and Budget (OMB) categories included in their national Uniform Data System (HRSA, 2009).5 Like hospitals, CHCs face challenges to collecting data, such as the need to train staff, the need to modify existing Health IT systems, and the need to ensure interoperability between the practice management systems where demographic data are collected and recorded and the EHR systems where the demographic data can be linked to clinical data for quality improvement purposes.
The structure and capabilities of primary and specialty care entities vary tremendously, ranging from large groups or health centers with highly structured staff and advanced information systems to solo physician practices with correspondingly small staff.
Indeed, Henry Ford Medical Group has collected race and ethnicity data for more than twenty years, and the Palo Alto Medical Foundation, a multispecialty provider group with several clinics, has recently begun to collect race and ethnicity data for use in analyses of disparities (Palaniappan et al., 2009).
Primary care sites typically do not have structured information available about care provided at other locations, so their ability to analyze data on quality of care by race, ethnicity, and language is generally limited to measures involving routine prevention and primary care.
Data on race, ethnicity, and language need collected in these settings could be useful throughout the health care system if mechanisms were in place for sharing the data with other entities (e.g., health plans) that have an ongoing obligation and infrastructure for analysis of data on quality of care which can be stratified by race, ethnicity, and language need and can look at episodes of care and care coordination.
Multispecialty group practices, which provide a range of primary care, specialty care, inpatient care, and other services, may be in a strong position to collect race, ethnicity, and language data because they have regular contact with large numbers of patients over long periods of time, can place the data collection in the context of improvement of care rather than administration of health insurance benefits, and typically have the necessary staff and other forms of infrastructure (e.g., a shared EHR system at all care sites).
Health plans, including Medicaid managed care and Medicare Advantage plans, have the capabilities necessary to systematically compile and manage race, ethnicity, and language data, and thus have roles to play in quality improvement (Rosenthal et al., 2009).
NPHC viewed direct data collection as the gold standard since this method supports interventions and direct outreach to individuals, but NHPC members realized that obtaining data through direct methods can take years to achieve in a health plan setting (Lurie, 2009).
Likewise, the limited success of Aetna with data collection (Box 5-4) after several years of concerted effort suggests that the upper limit of data collection by health plans with presently known direct methods may be far below the level necessary for identifying disparities in quality of care through stratified analysis, for example, of Healthcare Effectiveness Data and Information Set (HEDIS) data.
Despite informing members of how data will be used, plans may also face internal legal concerns about taking on unnecessary liability through threats of legal action due to misperceptions regarding the purposes of collection.
Moreover, conducting surveys of representative population-based samples in diverse settings requires an assessment of the need for in-language interviews (Ponce et al., 2006), balanced by the costs associated with high-quality translations and trained bilingual interviewers.
Yet oversampling incurs costs associated with the rarity of the population and the expense of the survey modality (e.g., the marginal cost of adding one more Samoan respondent would be greater for in-person household interviews than for telephone interviews).
Given the limitations of survey sampling, administrative databases offer the potential to collect data on higher numbers of smaller ethnic groups and make statistically reliable analytic comparisons across groups (e.g., a hospital administrative database versus a sample of hospital patients).
The above discussion of challenges faced by various health and health care entities highlights how important it is for data capture and quality to overcome Health IT constraints and minimize respondent and organizational resistance.
Until such integration is achieved, enhancing legacy Health IT systems, implementing staff training, and educating patients and communities about the reasons for and importance of collecting these data can help improve data collection processes.
An individual may self-identify in one clinical setting according to a limited set of choices, whereas another setting may offer more detailed, specific response options, or the individual's race may have been observed rather than requested and then recorded by an intake worker.
If a provider, for example, collects these data through self-report and hospital records involve observer assignment, then favoring the self-reported data from the provider setting would make sense if the data were linked and conflicting data were found.
Thus, a central system may be able to send data on a patient's race, ethnicity, and language to affiliated outpatient settings, but data collected in outpatient settings may not flow back to the central system (Hasnain-Wynia et al., 2004).
However, clinical performance data may be captured in an another system, meaning that race, ethnicity, and language data in the practice management system need to be imported into the EHR system to produce quality measures stratified by these variables.
While transitioning from legacy Health IT systems to newer systems is challenging, especially in physician practices (Zandieh et al., 2008), the American Recovery and Reinvestment Act of 200910 provides stimuli for moving forward with national standard Health IT systems.
Questions for requesting these data may introduce response bias, in the absence of adequate staff training.11 Before embarking on formally training staff to collect data, each entity needs to assess its data collection practices and delineate what is being done currently and what will change.
- On Tuesday, October 22, 2019
Understanding Medical Data from Real World Settings | Bristol-Myers Squibb
Learn how data collected from routine medical practice settings can help healthcare providers better understand how a treatment may work for patients.
Healthcare Data Mining with Matrix Models (Part 1)
Authors: Joel Dudley, Icahn School of Medicine at Mount Sinai Ping Zhang, IBM Thomas J. Watson Research Center Fei Wang, Department of Healthcare Policy ...
ACHILLES - A platform for exploring and visualizing clinical data summary statistics
Abstract: This presentation introduces ACHILLES (Automated Characterization of Health Information at Large-scale Longitudinal Evidence Systems), ...
The Age of Data-Driven Medicine: Mining the Patient-Feature Matrix
Dr. Paea LePendu, PCCI SMU Computer Science Seminar Electronic health records contain data about patients that include laboratory test results, medication ...
Using JReview to Analyze Clinical and Pharmacovigilance Data in Disparate Systems
Learn how JReview can be used to analyze clinical and pharmacovigilance data. -- Sponsors and CROs naturally rely on various clinical and safety systems ...
Unhealthy Politics: The Battle over Evidence-Based Medicine
Skip ahead to main speaker at 6:03 The U.S. medical system is touted as the most advanced in the world, yet many common treatments are not based on sound ...
Talks@12: Data Science & Medicine
Innovations in ways to compile, assess and act on the ever-increasing quantities of health data are changing the practice and police of medicine. Statisticians ...
The Future of eClinical Solutions and Integrating Systems and Data | Medidata
Robert Musterer is president of ER Squared, an eClinical consulting company partnering with Zifo Technologies, and presented at the recent Medidata ...
The Third Industrial Revolution: A Radical New Sharing Economy
The global economy is in crisis. The exponential exhaustion of natural resources, declining productivity, slow growth, rising unemployment, and steep inequality, ...
Dr. Susan Murphy: "Wearable Tech in Healthcare" | Talks at Google
University of Michigan Professor of Statistics & Psychiatry, Susan Murphy, discusses her research on improving sequential, individualized, decision making in ...