- Retrospective Performance Measures and Predictive Analytics
- Regulatory Reporting, Compliance and Transparency
- Quality Improvement
- Clinical Benchmarking
- Clinical Decision Support
- Population Health Management
- Comparative Effectiveness
- Public Health Surveillance
There are many types of clinical analytics, varying by:
- The data involved
- The users of the information
- The clinical actions or management decisions that the analysis supports
Examples of data can range from:
- Today’s lab test results for patients on a nursing floor
- Monthly utilization rates for specific medical devices at a surgical center
- Annual rates of unfilled prescriptions for all patients in a geographic area.
Examples of users making decisions based on clinical analytics can include:
- Nurses prioritizing their care activities for a shift
- A chief of surgery working with his medical staff to define best practice standards
- A public health official identifying neighborhoods with problems of access to healthcare services
Clinical analytics solutions and projects can also be categorized as either retrospective (looking back) or prospective (looking forward).
Performance measurement systems are typically retrospective, aggregating and analyzing information about numerous patients over an extended period of time. This helps managers make decisions about improving the organization’s performance or a population’s health.
Clinical decision support (CDS) systems are often prospective or predictive, analyzing detailed information about individual patients and offering guidance about the care to be provided to them in the near future.
Prospective CDS tools can be:
- Rule-based, encoding knowledge and logic carefully designed by clinical experts.
- Based on mathematical models that create a risk score using statistical correlations or machine learning technologies (e.g., neural network models).
Types of Performance Measures
Performance measures are often categorized based on whether they analyze an input to the care process or an output of the care process.
Types of performance measure include:
- Cost, Resource and Efficiency Measures track the specific resources used (e.g., CAT scans or drugs or surgical procedures), the rates at which those resources are consumed and the payments or costs for healthcare services.
- Process Measures track the level of compliance with specific standards of care (e.g., DVT prophylaxis for stroke patients in the hospital, periodic mammograms for women within a specific age range).
- Structural Measures provide information about how the care delivery system operates (e.g., volume of specific surgical procedures performed by a physician, the percent of prescriptions transmitted electronically).
- Outcomes Measures track the impact of care delivery on patient health and monitor items (e.g., mortality rates, control of blood pressure levels, ability to perform activities of daily living). Outcomes also include Patient Safety Measures, which track items such as hospital-acquired infections, adverse drug events and patient falls.
- Patient Experience Measures are largely based on patient survey questions. They track patient perceptions and satisfaction levels.
Hospitals and other healthcare providers are required to submit numerous clinical data to various government agencies and accrediting organizations in order to:
- Comply with legal mandates, to qualify as a Medicare provider
- Maintain accreditation
- Obtain various financial incentives offered by Medicare, Medicaid and other payers
Many other types of clinical analytics rely in part on repurposing the data that providers already collect for regulatory reporting. A subset of the data that providers submit to the federal government is made available to the public on websites such as Hospital Compare.
A dramatic increase in the scope of these required data submissions, the associated financial incentives and the scrutiny of publicly reported measures have been critical factors in increasing the focus of management on clinical performance measurement. Important measures based on regulatory data submissions include:
- Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) , a national survey that asks patients about their experiences during a recent hospital stay. Patient experience measures based on HCAPS are published on Hospital Compare and are included in a formula that determines payments to hospitals under Medicare’s Value-Based Purchasing (VBP) program.
- Core Measures are Process Measures that are abstracted from medical records and submitted to Medicare and to the Joint Commission. Some of these measures are published on Hospital Compare and are included in the formulas for Medicare’s VBP program. Additional information is available in the CMS website and on QualityNet.
- Physician Quality Reporting System (PQRS) measures are mandated by CMS, which will soon start imposing financial penalties on physicians who do not participate. The American Medical Association website provides information about PQRS.
- Meaningful Use data submissions include numerous structural measures related to adoption and use of electronic health records systems (EHR), as well as a number of Clinical Quality Measure (CQMs) (mostly process measures). These data submissions are required in order for "Eligible Hospitals" and "Eligible Providers" to receive stimulus payments under the health reform provisions of the American Recovery and Reinvestment Act of 2009 (ARRA).
- Ongoing Professional Practice Evaluation (OPPE) is a requirement for hospital accreditation set by the Joint Commission. While OPPE does not require regular data submission to an external organization, it does mandate that the hospital must have an ongoing process in place to review the performance of physicians who have privileges to practice in the hospital. It also specifies the types of measures that must be included. This accreditation requirement has spurred many hospitals to adopt clinical analytics solutions that can help them evaluate physician performance.
Back to top >
Hospitals routinely create reports on physician performance from a variety of financial and clinical perspectives. Since physicians direct most of the clinical activity in the hospital, they have an enormous influence on the overall performance of the hospital itself. In the past, "physician report cards" have been a source of friction between hospital administrators and community physicians, due to a focus on the reduction of hospital costs rather than on helping physicians improve performance along dimensions that are important to them and that they feel are under their control.
In recent years, many hospitals have learned that they can more effectively engage with their physicians on performance improvement by focusing first on clinical outcome measurements for the care provided to patients, looking at cost reduction as a secondary effect of quality improvement.
Payers also evaluate provider performance, often by focusing on population health measures that evaluate primary care physicians based on the provision of preventive care and on the management of chronic disease. Payers also evaluate surgeons and other specialists based on the cost and outcomes of episodes of care, including post-discharge rehabilitation resources, readmissions and other follow-up care.
Payers may use physician performance metrics to:
- Choose which physicians to include in their provider networks
- Assign physicians to a higher or lower tier in their networks
- Pay quality bonuses to strong performers.
Methodologies are challenging for truly meaningful measures of physician performance due to the complexity of healthcare operations and the wide variation in the condition of patients being treated. Even attribution of the right patients to a physician can be a tricky task in situations where multiple physicians are consulting on the care of an individual. Risk adjustment methodologies are essential to meaningful physician performance measures since some physicians do treat sicker patients and can be expected to have fewer positive outcomes as a result.
Back to top >
Quality improvement projects are arguably the most exciting and valuable focus of clinical analytics today. Clinical data analysis is an essential part of process improvement, whether a formal methodology such as Lean or Six Sigma is employed, or a more ad hocapproach to improving performance is taken. Clinical analytics can:
- Help identify the top problem areas where there are high degrees of variation in outcomes involving serious impacts on patient health and/or excessive costs.
- Help in designing a solution to problem and measuring the effects of changes to make sure the problem really is solved.
Many clinical process improvement projects focus on reducing the frequency of serious complications of care.
- For hospital-acquired infections that progress into severe sepsis, analyses of mortality rates and costs can quantify the potential savings in both lives and dollars. Clinical analysis can also provide clues as to root causes of the problem by using detailed information about the care process captured in the EHR, helping focus the improvement team’s investigations into issues on the nursing floor and in the operating suite. Investigations will lead to recommendations for process changes, such as implementation of a rapid response protocol triggered by specific measures of patient status.
After changes in a care process are made, clinical performance reports can provide rapid, accurate feedback on what is working and not working to achieve the desired outcome: fewer cases of severe sepsis. When significant investments are needed to broadly implement an improvement, cost savings analysis can often help make the business case for "doing the right thing" to improve quality.
- The selection of specific brands and models of implantable devices to be used in surgery. In one surgical collaborative, analysis of surgical outcomes pointed to a particular type of mesh being associated with infections in the repaired tissues. Avoiding the use of that product resulted in lower infection rates.
- Analyses of orthopedic surgeries showed good outcomes for the several different brands of hip and knee implants that were, due to varying physician preferences, in use. Information showing uniform outcomes helped the surgical staff standardize on a smaller list of brands, allowing for cost savings due to volume discounts. In these cases, clinical analytics overlaps with supply chain analytics in the management of physician preference items.
Back to top >
Clinical benchmarking relies on many of the same measures providers use to manage their own performance, while aggregating information from multiple organizations to identify and quantify the highest levels of performance from among a diverse, representative sample of providers. Benchmarking helps hospitals and other providers avoid the trap of standardizing on mediocre performance levels and helps identify opportunities for improvement that may not be obvious from internal data alone.
Clinical benchmark reports can rank providers into quartiles or quintiles – or even deciles – based on their relative performance.
Rankings may be based on:
- An individual measure
- A composite measure that combines related measures within a domain
- A "balanced scorecard" formula that combines metrics from multiple domains.
In addition to ranking, benchmark reports can also help quantify the opportunity for improvement by showing what the results for a provider would be if performance were improved to the level of the best performers.
For example, an opportunity report could show the potential cost savings for a hospital that reduced its surgical complication rates to the level of the best group of surgeons in the comparative database.
Standardization of measurement to achieve comparability of data is essential for meaningful benchmarking. The clinical codes and classifications used for Medicare payment are often leveraged for benchmarking due to their ubiquity and familiarity. But some benchmarking organizations also employ their own unique classification schemes and provide tools to ensure that all organizations contributing data are using the same definitions. Patient registries, focused on narrow populations of patients with a specific condition or procedure, will often implement unique data definitions to allow for deeper clinical analysis.
Back to top >
Clinical Decision Support involves providing the right information at the right time to help guide a clinician towards providing the best care to the individual patient they are currently treating. The introduction of predictive clinical analytics into the workflow of clinicians is an effort that that many healthcare organizations are keenly interested in, and examples of projects with good results are becoming more numerous every day.
In the past, clinical decision support has often been limited to providing physicians with convenient access to static clinical reference information such as documentation of standard care paths or details of drug dosages and interactions. And some less than successful efforts at providing guidance during the order entry process have led to a backlash from physicians due to ‘alert fatigue’. But now we are starting to see examples of more intelligent clinical decision support solutions that use predictive analytics to tailor very specific care guidance based on details of the individual patient’s condition.
- CDS for physicians focused on VTE prophylaxis (i.e., avoiding the formation of blood clots in patients). A group of hospitalists documented clinical logic that was embedded into an order entry system to create a tailored care plan for each patient. The clinical logic included:
- A patient assessment
- Assignment to a risk category
- Specific recommended treatments
Clinical information systems were configured to trigger these rules at the right time so the physician could easily enter detailed orders consistent with best practices.
- CDS for nurses focused on ensuring ongoing compliance with several clinical protocols, including those tracked and reported as Meaningful Use Core Measures. The hospital put a graphical display at each nursing station showing each bed in the unit, along with critical information about the status of each patient. Software continually analyzed the status of each patient to identify needed interventions. It also used stoplight alerts, countdown timers and other graphical displays to remind nurses of required care, helping them prioritize activities in caring for their patients.
There is also considerable interest in using mathematical modeling tools for CDS. Some organizations are experimenting with mathematical models to predict a patient’s risk of readmission using clinical data elements from an EMR.
In addition to hospitals, several other types of organizations (e.g., insurers and public health departments) analyze clinical data for their own purposes. As health systems transform themselves into Accountable Care Organizations (ACOs) under health reform, these population-focused clinical analytics are also becoming more important to providers.
Insurance companies and other payers do not have patients – they have members. At any given time, some of those members will not be using any health services at all. It is still important for them to be accounted for in performance measures. In place of the ‘per case’ statistics that are so prevalent in provider settings, payers very often use ‘per member’ statistics, where the denominator is the number of members enrolled during the time period (perhaps refined by additional criteria such age, sex, health status or location). Rates are typically presented as either per member per month (PMPM) or per member per year (PMPY). Population analyses have most often been based on healthcare claims data, limiting the depth of clinical detail available.
Over-utilization of health services is a concern from a financial perspective, but unnecessary procedures can also be harmful to patients. Payers routinely analyze claims and pre-authorization requests to compare them to standards of care, and may deny payment for services beyond their accepted standards of care. Over-utilization is also a concern from a public policy perspective. Using Medicare data, the Dartmouth Atlas Project has created numerous analyses and reports documenting wide geographical variations in the rates at which elective surgical procedures are used by patients and their doctors.
Under-utilization of health services is a potential hazard as payers try to control costs and is an important focus of regulatory reporting and accreditation requirements. The Healthcare Effectiveness Data and Information Set (HEDIS) is a tool used by most health plans to measure their performance. HEDIS reporting measures track:
- Immunization rates for children
- Cancer screening rates for adults
- Access to preventive care
- Control of chronic conditions such as diabetes, asthma and high blood pressure.
The website of the National Committee for Quality Assurance (NCQA) is a one good source of information about HEDIS measures.
Disease Management and Wellness Programs also make extensive use of clinical analytics. Analysis of claims data can help insurers and at-risk providers identify the sub-populations of patients with chronic diseases who would benefit from case management services intended to help them proactively manage their health and avoid expensive acute episodes. Clinical analytics can trigger alerts to either the case manager or the patient to remind them of necessary actions, such as refilling a prescription, making an appointment with a doctor or sticking to an exercise routine.
Where home monitoring is in place, analytics can trigger actions based on patterns in blood pressure levels or other clinical measurements. Clinical analytics also plays an important role in evaluating the overall effectiveness of programs in reducing costs and improving health.
Population risk adjustment is an example of predictive analytics being used from a payer perspective. Risk adjustment methodologies look at member characteristics (e.g., age, sex and health history) and assign the member to a risk group that is predictive of future health costs. A variety of methodologies have been used by private insurers to manage their enrolled populations.
With the implementation of the Medicare Advantage program, which essentially subcontracts the insurance risk for Medicare members to private companies, CMS needed to have its own method of setting per-member payment levels that reflect the varying disease burden of different member populations. The CMS Hierarchical Condition Categories model (CMS-HCC) is used to adjust capitation payments to Medicare Advantage plans. CMS-HCC and the other proprietary models can also be used to define more homogenous populations for clinical analysis and for risk-adjustment of quality measures such as mortality.
Like DRGs in the acute care environment, the financially focused CMS-HCC and similar models are not ideal for clinical performance measurement, but may be the most practical alternative in some situations.
Back to top >
Every constituency in the healthcare business has an interest in learning more about the comparative effectiveness of drugs, devices, procedures and treatments. In the past, some comparative research projects have been accomplished using Medicare claims data. Other projects have involved the laborious collection of more detailed clinical data. The ongoing rapid adoption of EHRs brings with it the potential of aggregating much broader and deeper clinical data that will allow for comparing the effectiveness and costs of alternative patterns of care. Clinical analyses using these data can:
- Assesses outcomes across large populations of patients
- Adjust for risk factors
- Determine what differences are attributable to specific interventions
For some high-volume procedures, hospitals can meaningfully compare outcomes and costs of different treatments currently being provided at their site. For lower volume procedures and for new procedures, patient registries or other aggregated databases are required to achieve the level of statistical rigor needed to justify changes in medical practice. Registries for orthopedic surgery have been analyzed extensively in evaluating the effectiveness of different types of joint implants and highlighted problems with "metal-on-metal" joints.
Many government agencies and private non-profit organizations are involved in clinical effectiveness research (CER). Good sources of information about CER include The Agency for Healthcare Research and Quality (AHRQ), The Center for Medical Technology Policy (CMTP), and The ECRI Institute.
Back to top >
Public health departments at the state and federal level are interested in monitoring and responding to outbreaks of disease and in analyzing other trends in the health of their citizens. Legal mandates for reporting cases of certain important diseases are a critical foundation for their efforts. State health department-mandated "trauma registries" are another important source of data, focusing on major injures treated in hospital emergency departments. One of the key objectives of the Meaningful Use financial incentives for adopting EHRs is to ease the transmission of data from hospitals to public health agencies. Once fully implemented, this ongoing national effort should significantly expand the timeliness, depth and breadth of clinical data available for analysis by public health officials.
Analysis of public health databases can help guide government responses to disease outbreaks and make them more effective, by distributing vaccines where they are most needed. Analyses of trauma registries have been important determining where emergency services such as ambulances are needed and in shaping public policy on controversial issues such as motorcycle helmets and gun control.
Surveillance can also reveal unexpected patterns of adverse events associated with a drug, device procedure. For example, Kaiser Permanente’s large patient database flagged adverse events associated with Vioxx, leading ultimately to the withdrawal of that medication from the market. Broader deployment of EHRs and aggregation of clinical data for surveillance should allow for issues like this to be recognized more quickly in the future.
Recently, there has been excitement in the press about projects that have analyzed social media messages to track the spread of flu symptoms across the country. While lacking the clinical rigor of medical diagnoses based on physical examinations and lab tests, real-time analysis of activity on web sites may very well mature into a useful supplement to traditional epidemiological methods.