A data code is a standardized alphanumeric term (string of numbers, letters) that uniquely designates the meaning or referent of the data. That is, the code relates to the data element—a discrete piece of information. Data codes identify the type of data (drug, diagnosis, medical equipment, etc.) as well as the information in the data element (e.g., a specific formulation of penicillin, a specific diagnosis, a specific piece of medical equipment).
This section complements data sources (types) and discusses how the data sources display their information in terms of standardized coding systems.
With incentives and other drivers in health information technology, more providers are implementing electronic medical record (EMR) systems. Besides improving legibility and accessibility, EMRs encode much of the data stored in them—even in what appear to be narrative notes. Diagnoses may be encoded as ICD codes or drugs as NCDs (see below), but characteristic of EMRs is that they also encode information about the encounter content, diagnoses, procedures, medications, procedures, and tests as SNOMED (Systematized Nomenclature of Medicine) codes, which can therefore be retrieved in queries and used for analysis.
Administrative data is often used to identify good candidates for disease management programs and interventions, as well as to evaluate the effects of population health management interventions. The source documents for administrative data relate to the payer’s administration of health care insurance and delivery. Coding systems used for administrative data include:
- ICD-CM (International Classification of Diseases Clinical Modification): The International Classification of Diseases, Clinical Modification is used to code and classify morbidity data from the inpatient and outpatient records, physician offices, and most National Center for Health Statistics (NCHS) surveys. Most health care organizations in the United States continue to use the 9th edition. In the U.S., use of the 10th edition is mandated by 2013. ICD-9-CM codes may be expressed as the three-digit root (e.g., 250 – diabetes mellitus) or may be carried to the fourth (250.6 – diabetes with neurological manifestations) or fifth (250.61 – Type 1 diabetes with neurological manifestations) digit to express more detail about a diagnosis or finding. In addition to expressing diagnoses and findings, ICD-9-CM codes may be used to express procedures performed in an inpatient setting, injury or poisoning (E codes), or health status (V codes). ICD-10-CM significantly expands the size and number of codes available. In addition, it contains structural changes to both the diagnosis and surgical procedures codes including the E and V codes.
- DRG (Diagnosis-Related Groups): A method for classifying hospital stays into homogeneous categories of disease states and resource utilization. The initial idea behind DRGs was to create clinically meaningful groups of patient diagnoses and procedures that may receive the same treatment and therefore be expected to cost about the same amount to treat. Thus, Medicare could pay a single fee to a hospital based on the DRG linked to the principal reason for the hospital stay. The assignment of an inpatient stay to a DRG is based on principal and significant co-morbid diagnoses, principal and secondary procedures if performed, presence of complications, type of treatment (medical or surgical), age, and discharge status. Currently, there are more than 500 DRGs. Clinically-related DRGs roll up to 25.
- CPT-4 (Current Procedural Terminology, 4th edition): A listing of procedures (including evaluation and management, medical, surgical, imaging, laboratory, and pathology) performed by health care providers. The CPT listing is maintained by the American Medical Association and is updated yearly; the current version contains well over 8,000 codes and descriptors. Note that hospitals use ICD procedure codes for reporting surgical procedures and CPT/HCPCS for line level reporting; CPT and HCPCS (see below) codes are used by providers (e.g., physicians, labs, ancillary providers) actually performing the procedures.
- LOINC (Logical Observation Identifiers Names and Codes): This code set names the test and gives considerable additional detail including the test result.
- HCPCS (Healthcare Common Procedure Coding System): A method for providers and suppliers to report services, procedures, and supplies. HCPCS is considered to have three levels: Level I, CPT; Level II, HCPCS National Codes; Level III, Local Codes. Generally, the designation “HCPCS” means Level II, which contains codes for transportation, drugs (usually given in an office or facility), supplies, and certain services such as rehabilitation.
- NDC (National Dug Code): An FDA-assigned code for each unique combination of manufacturer, generic drug, dose, and package size. A 10-digit three-part code identifies the labeler/vendor, drug, and package size; however, manufacturers add a digit to create an 11-digit code to allow computers to determine the manufacturer, drug, dose, and packaging. While the FDA has established a hierarchy of 21 major and 139 minor drug classes, private companies have established more clinically useful classifications.
- NUBC (National Uniform Billing Committee): These codes are established by a committee of the American Hospital Association, to ensure uniformity of billing by hospitals (the UB-04 data set).
Note: Coding manuals are often published by more than one entity. When an official organization such as the World Health Organization maintains a code set, and commercial organizations also publish code listings (often augmented with helpful contextual and explanatory information), we give only the official maintaining organization’s contact information.
Centers for Medicare and Medicaid Services. Washington, DC: US Department of Health and Human Services. Retrieved from http://www.cms.hhs.gov/
“CPT (Current Procedural Terminology).” American Medical Association. Retrieved from http://www.ama-assn.org/ama/pub/category/3113.html
“Classification of Diseases, Functioning and Disability.” National Center for Health Statistics. Retrieved from http://www.cdc.gov/nchs/icd.htm
“Logical Observation Identifiers Names and Codes (LOINC®).” LOINC. 2010. Regenstrief Institute. Retrieved from http://loinc.org/
Data mining is the exploration and analysis by automatic or semiautomatic means of large quantities of data in order to discover meaningful patterns that could be developed into rules.
Data mining involves utilizing a set of techniques in an automated approach to exhaustively explore and help identify complex relationships in very large data sets. Data mining helps discover hidden knowledge, unexpected patterns, and new rules within these large databases. The nature of data mining is exploratory and typically used with databases that are too large for traditional statistical techniques. A significant distinction between data mining and other analytical tools is in the approach used in exploring the data interrelationships.
Many analytical tools and data modeling processes support a verification-based approach, where a user creates a theory about specific data interrelationships and then uses tools to verify or refute those hypotheses. This approach relies on the expertise of the analyst to pose the original question and refine the analysis based on the results of potentially complex queries against a database.
The effectiveness of this type of analysis is limited by the ability of the analyst to pose appropriate questions, understand the results, and manage the complexity of the analytical process. Data mining, in contrast to other analytical tools such as query and reporting tools (SQL, multidimensional and relational tools, etc.), uses discovery-based approaches in which pattern matching and other algorithms are employed to help determine key relationships within the data. Data mining algorithms can look at numerous data relationships concurrently, highlighting the elements that are dominant.
Key steps in the data mining process are as follows:
The data must be obtained, stored, cleaned, and reduced to the most critical elements. Although intuitively obvious, it is important that the data available be useful to address the issue in question. The data must be assessed and explored statistically and graphically for validation. Data elements may need to be transformed to process correctly, including imputing missing values where appropriate. Modeling techniques can then be applied to the data to discover patterns and rules.
PROCESSING AND TRANSFORMATION
Very large databases must be stored and quickly accessed, utilizing intensive computational methodologies such as decision tree induction, rule induction, nearest neighbor (case-based reasoning), clustering methods (data segmentation), association rules (market-based analysis), etc. Each data mining application is supported by a set of algorithmic approaches used to extract relevant relationships in the data. The approaches are different based on the class of problems they are able to solve.
SOME COMMON DATA MINING MODELS (see also predictive models)
- Clustering is often the initial step in data mining. Clustering identifies groups of related records that can be used as a starting point for identifying further relationships. The process assigns records with a large number of attributes into a relatively small set of groups or segments. The assignment process is automated, typically using computer software. The characteristics of the data set are isolated and then partitioned and clustered. There is no need to identify the groupings desired or the attributes that should be used to segment the data set.
- Association rules classify a set of issues using a market-basket-type approach. The goal is to find trends across a large number of transactions that can be used to understand natural patterns such as purchasing behavior, etc. The results of association approaches are typically expressed in terms of confidence-rated rules (e.g., 80% of the people who purchase low-deductible health plans also purchase disability products).
- Sequence-based analysis is a typical market-based approach that looks at a collection of items as part of a point-in-time purchase or transaction. A sequence-based approach ties together a series of actions in a time sequence. With this approach, both the coexistence of the items and the sequence of their purchase are important. In health care, this method can be used to potentially identify the optimum course of treatment, such as which procedure in what sequence produces the optimum results.
- Classification is potentially the most commonly applied data mining technique. This approach typically employs tools like decision trees, neural networks, similar algorithms. Classification algorithms use a training set of example transactions in a predictive manner to model records into pre-defined classes. An example of an application of this method would be modeling the relative risk of an individual’s health status using neural networks or decision tree analysis.
- Decision trees are built based on the input of a training data set, which is based on historical data. Each record of the model set is run through the branches of the tree (rule) until the record reaches a leaf (result).
- The nearest-neighbor technique classifies a record by calculating the distances between the record criteria and the training data set. The record is then assigned to the class that is most similar to its nearest neighbor.
- Rule induction describes the data and allows the user to visualize what is going on in the data through a generalization from the information in the data.
- Neural networks use nonlinear models that try to learn based on pattern recognition, similar to the way our brains reason. These very complex models may be prone to overfitting based on the input of a training set.
After completion of the statistical components, it is important to understand and draw conclusions from the results. It is also important to determine whether the results are meaningful, significant, and new. Users must be careful in analyzing the results from data mining so as not to draw spurious conclusions.
- Data modeling results are susceptible to a number of important considerations:
- They are prone to be affected by dirty or incomplete data;
- They are prone to sampling bias;
- They don’t always create results that are intuitive or easily explainable by if-then reasoning; or
- They don’t always provide statistically valid results.
USE OF DATA MINING IN PREDICTIVE MODELING
Data modeling forms the core of one potential approach to predictive modeling. Certainly there are standard statistical methods (regression-based analysis, etc.) that could be used in predictive modeling utilizing a more verification-based approach. A data mining process for predictive modeling might apply tools like decision trees or neural networks to create an exploratory approach to identifying risk characteristics for individuals.
Brown, J. (2002). Data mining – a practical look at data preparation.
Burbidge, R. (2000). Data mining: Staying ahead in the information age, a tutorial in data mining. Cambridge: Cambridge University Press.
Duncan, I., and Robb, A. (2003). “Population risk management: Reducing costs and managing risks.” Intelligent and other computational techniques in insurance. Hackensack, NJ: World Scientific.
EPCC Home Page. Retrieved from http://www.epcc.ed.ac.uk/
Friedman, J. (1997). Data mining and statistics: What's the connection? Department of Statistics, Stanford University.
Moxon, B. (1996). Defining data mining. DBMS Data Warehouse Supplement.
Patterson, D. (1998). Artificial neural networks: Theory and applications. Singapore: Prentice Hall.
Thearling, Kurt. (2006). Data mining and analytic technologies. Retrieved from http://www.Thearling.com
Data Quality – Data Integrity
Data quality refers to the accuracy, consistency, and completeness of data. Before any work is done with administrative data, it is necessary to establish the validity or quality of the data (see reconciliation).
There are several types (sources) of data from which we can infer something about the past and current state of a person’s (or population’s) health and about what occurred during health care encounters. These data sources include the medical record, survey data, health risk appraisals, test (laboratory, imaging, pathology) results, administrative data, and data related to health care operations.
Briefly, medical records may be written or electronic and represent the health care provider’s documentation of the encounter (plus associated encounter records, test results, and communications); survey data (including health risk assessments) reported by the patient or caregiver; administrative data obtained from documents and processes related to coverage and payment for health care services (such as insurance eligibility; claims for payment for health care encounters, procedures, and devices/equipment; and prescription fills); and operations data derived from health plan or disease management operations, contacts with health plan personnel, disease management candidate or participant calls, mailings, or measures of retention.
Each type or source of data has advantages and disadvantages when used for the activities of population health management, such as identifying or enrolling candidates or measuring the results of program interventions.
The data sources section describes and contrasts the major sources of data used in population health management. Many data types (especially administrative) are coded in standardized ways, so they can be machine-read and can promote a common nomenclature; codes related to data sources are discussed under data codes.
Following are descriptions of the sources of population health management-relevant sources of data and their advantages and pitfalls in relation to the tasks of population health management.
- Medical records: This is considered gold standard data, to which the accuracy of other data sources is compared. However, medical records are costly and sometimes difficult to obtain. To provide a complete picture of a person’s health, medical records must be obtained from all sources of health care. Data in medical records may not always be accurate, because they may contain provisional diagnoses, or diagnoses that are later corrected after further study or observation. Such data may not be complete or legible if handwritten.
- Administrative data: The source documents for administrative data relate to the payer’s administration of health care insurance and delivery. While data does not directly relate to the content of health care delivery, some administrative data is indirectly related to health states and processes as claims for payment of health care services. For example, a claim for treatment for asthma could have an ICD-CM (International Classification of Diseases, Clinical Modification) codes for asthma and CPT (Current Procedural Terminology) codes for an outpatient visit and administration of a nebulized bronchodilator (treatment). In the same day we might find a claim with a National Drug Codes (NDC) (drug) for payment for a corticosteroid inhaler (an asthma controller medication). It is therefore possible to build a picture of the conditions, how the person was treated, and the sequencing of diagnostic and treatment events. Data-related codes are covered in more depth under data codes. We note here the following major sources of administrative data.
- Hospital claims with hospital inpatient diagnoses and procedures are claims from inpatient stays include principal diagnosis (the diagnosis believed to be responsible for the hospital admission after evaluation, not at the time of admission); secondary diagnoses (diagnoses significantly affecting management of the primary diagnosis, generally those believed to increase the length of stay, such as certain co-morbidities and complications), and principal/secondary procedures (the principal procedure is that which is most resource intensive). Additional coding includes whether the diagnosis was present on admission.
- Outpatient claims are performed in outpatient facilities, providers’ offices, and laboratories. Again, at least in theory, each claim for a specific service is linked to a diagnosis. For example, a chest x-ray obtained in an imaging center might link to the diagnosis of pneumonia.
- Claims for medications may be provided in a facility or office or by prescription to be filled at a pharmacy by the patient. There is no diagnostic code linked to this source of data.
- Medical equipment and supplies (examples include syringes, oxygen, wheelchairs, and dialysis supplies) are usually linked to diagnoses.
- Enrollment and demographic data: Health plans maintain files containing relevant information about each insured member—e.g. unique identifier, relationship-to-subscriber, age, zip code, and enrollment status.
- Operations data are generated in the course of health plan and population health management operations and document calls with health plan personnel, population health management candidate or participant calls, communications (such as mailings), and measures of program retention.
POPULATION HEALTH MANAGEMENT-RELATED ISSUES
- Administrative data: While very commonly used, administrative data is only one source for information about health states and health care encounters. Administrative data is relatively inexpensive and readily available and can be used to update a patient’s state of health or treatment changes, resulting in a health care encounter and reflected in a claim. This discussion focuses on administrative data, but other sources of such information may be used to supplement, corroborate, and help validate administrative data.
- Survey data/Assessment data: Intermediate in cost and difficulty to obtain compared with other administrative data, survey data or assessment data (HRA) provides information about someone’s health early in their health plan enrollment (as opposed to a waiting period for claims), and information about a person’s health that cannot be obtained from administrative data (e.g., smoking status, exercise, functional status, and quality of life). However, survey data can rapidly become out of date and are subject to inaccuracy. Surveys are completed by patients or caregivers, some of whom may not give accurate information because of language or comprehension barriers or may not be knowledgeable about their conditions or test results.
- Test (lab, imaging, pathology) results: These may be difficult to obtain in electronic fashion but when available are accurate (and more reliable than self-reported data) and can contain very useful information for population health management.
Administrative data have many advantages: From enrollment status, a disease management provider can determine whether a member is currently eligible for health care (including population health management) services and whether a person is eligible to be in a denominator that requires a threshold number of months’ enrollment (see example). For a population-based analysis, the total number of months’ enrollment during a specified period is the denominator in a per member per month calculation. Administrative data are readily available, becoming increasingly standardized in content and format, and, if used properly, are reasonably accurate in identifying individuals or groups who have specified diagnoses or who have undergone certain procedures.
However, it is important to recognize that administrative data have certain inherent disadvantages.
- Lack of sensitivity: if not coded, administrative data cannot be used to identify someone who might really have a disease, be taking a drug, or has had a procedure. For example, if being diabetic requires at least two occurrences of ICD-9-CM codes for diabetes during a specified period, an individual with zero or one occurrence will not be identified as diabetic. Sensitivity can be increased by requiring fewer occurrences of codes or not corroborating with drugs or procedures but at the expense of specificity (false positives).
- Lack of specificity: ICD-9-CM codes are sometimes used when the physician is only considering (ruling out) the diagnosis; queries requiring only a few occurrences of such a code will often misidentify someone as having the disease. This is often seen with diabetes, heart disease, hypertension, and heart failure. Specificity can be increased by requiring more ICD-9-CM codes occurrences or corroboration with drugs (e.g., having heart failure requires two occurrences of ICD-9-CM codes and two prescription fills for diuretics and ACE inhibitors) but at the expense of missing some people who really have the disease. It is anticipated that ICD-10-CM will enable more specificity in administrative data.
- Inability to indicate impactibility: Population health management programs work to the extent that they identify people whose condition is impactible. (Single codes rarely do that, though combinations of codes—e.g., diabetics with proteinuria not on an ACE inhibitor—may do so.) However, administrative data cannot identify people who are ready to change, who are nonadherent because of changeable health beliefs, or who have certain risk factors that are rarely coded (e.g., smoking, exercise) because they do not affect reimbursement or are simply not available as codes.
- Data validation: In population health management, administrative data are often used for program purposes (e.g. identification of chronic members). For such purposes, speed and ease of program implementation are often the most important criteria and will shape the data processes implemented in the population health management company. Measurement, however has other requirements (for example, speed is no longer critical, but accuracy and consistency with published financial data are). A population health management company that wants to ensure data integrity will implement data processes that reconcile incoming data, check consistency with norms, and track trends in key indicators. Without rigorous data quality controls, outcomes measures would be suspect.
- Use of codes to identify suitable candidates for population health management: Suppose a population health management program wants to find people who have heart failure, are (or will become) relatively sick (so that intervention would make a difference), and are impactible. If two occurrences of ICD-9-CM codes for heart failure in the past year are used as a criterion, almost everyone with clinically significant heart failure will be identified, but many people who do not have heart failure (because the physician used the codes as rule-outs or was improperly coded) will be included. In addition, the population health management program will have to contact far too many people to find the group who is sick enough, impactible enough, and willing to engage (high sensitivity, low specificity). Accordingly, one of the following criteria is required:
- At least two occurrences of ICD-9-CM codes for heart failure in 12 months;
- At least two prescription fills for a “loop” diuretic (a diuretic usually used for heart failure) in 12 months;
- At least two prescription fills for either digitalis, beta-blocker, or an ACE inhibitor in 12 months; or
- At least one hospital admission with heart failure automatically qualifies.
This approach will probably miss some people with mild heart failure but spend much less time contacting people who do not have heart failure. From here, it is possible to add qualifiers for impactibility (e.g., no ACE inhibitor or beta-blocker).
- Use of codes for measuring clinical outcomes: The population health management program wants to determine (using claims) how many members with heart failure are taking a beta-blocker. For this you need a denominator, which you decide is all members who:
- Meet the minimum claims definition for heart failure (ICD-9-CM codes and National Drug Codes (NDC));
- Have at least 6 months’ insurance coverage during the measurement year (enrollment data) and were insured during the last month of the measurement period; and
- Have at least one prescription fill for a beta-blocker (NDCs) in the 6 months preceding the end of the measurement period.
(Note: The above example is an illustration of how administrative data may be used and not a recommendation for specifying a clinical indicator.)
Chute, C., Cohn, S., Campbell, K., et al. (1996). The content coverage of clinical classifications. Journal of American Medical Informatics Association, 3(3), 224-233.
Duncan, I., and Robb, A. (2003). “Population risk management: Reducing costs and managing risk in health insurance.” Intelligent and other computational techniques in insurance. Hackensack, NJ: World Scientific.
Fowles, J., Fowler, E., and Craft, C. (1998). Validation of claims diagnoses and self-reported conditions compared with medical records for selected chronic diseases. Journal of Ambulatory Care Management, 21(1), 24-34.
International Society for Pharmacoeconomics and Outcomes Research (ISPOR). (2003). Health care cost, quality, and outcomes. ISPOR Book of Terms.
Decision Support (DS)
Decision support comprises processes and tools to assist health care providers, patients, and their care supporters to make prevention, screening, or treatment decisions that are aligned with specified values for outcomes (such as health states, quality of life, function, and cost or cost-effectiveness).
At many points during the course of prevention, screening, or management of a health condition, health care providers, patients, and care supporters must make a variety of decisions. Examples of such decisions include the following.
- Prevention: whether to obtain a flu shot; what diet to follow to reduce risk of heart disease.
- Screening: whether (or how often) to have a pap test (cervical cancer) or mammogram (breast cancer); whether to have a colonoscopy (colorectal cancer); whether to have a fasting blood glucose test (diabetes).
- Management: whether to take a statin drug for a person with diabetes (to prevent cardiovascular disease); whether to take adjuvant chemotherapy for a woman who just completed surgery and radiotherapy for primary treatment of breast cancer without evidence of spread to lymph nodes.
The process of decision making must occur because people have differing points of view about what outcomes are important(e.g., cost versus the possibility of 6 months longer life with a new chemotherapy regimen that increases the risk for serious infection) and because decisions must be made under conditions of uncertainty (e.g., surgery might give a 50% cure rate over a population, but the outcome for an individual is uncertain). In addition, applying the results of evidence-based medicine is often not straightforward because the patient might not be exactly like those in the studies that provided the evidence. (Note: Decision analysis is often used by payers—especially centralized payers such as governments—to prioritize health care expenditures. This use of decision support will not be explored here).
What decision is made depends on the decision maker’s point of view, the relative values the decision maker places on the outcomes of the options, and the knowledge available at the time the decision is made. Decision analysis is a formalized method of decision making which involves clearly identifying the decision to be made (the alternative actions, including no action); the point of view (who the decision maker is); the possible consequences of the alternative actions; the probability of these outcomes; and how the outcomes will be valued (e.g., cost, mortality, quality of life, function).
While physicians and patients usually make decisions informally—weighing their view of the evidence about the likelihood of various outcomes in light of what is important to the patient—there are tools that formalize the decision-making process. One such tool is the decision tree, which the Society for Medical Decision Making defines as: “A graphical representation of a decision, incorporating all clinically important choices, uncertain events (and their probabilities), and outcomes of the decision.” For example, suppose a woman who was just treated for localized breast cancer is working with her physician to decide whether to undergo adjuvant (after primary treatment) chemotherapy. Her choice is chemo or no chemo. On a population basis, chemo will give her a 90% chance of a long-term cure, but will entail many unpleasant side effects (reduced quality of life, risk for serious infection) in the short run. Not having chemo will give her an 85% chance of a long-term cure but no further impact on her quality of life (though the anxiety of not choosing the other alternative has to be taken into account). The patient must now consider outcomes beyond survival at 10 years. Formal decision analysis can assign values (utilities) to the patient’s preferences for years of life versus the effects of chemotherapy. The decision tree uses the probabilities of successful outcomes of each option and the utilities (value to the decision maker) of those outcomes, and calculates an overall utility (such as quality-adjusted years of life or cost per quality-adjusted year gained) for each option. The decision maker then chooses the option with the greatest utility.
Decision support tools (aids) are being used more frequently in helping patients make informed decisions and to increase their satisfaction with the decision-making process. These tools are usually used in an encounter between the patient and a provider (which could include a disease management program). The Cochrane Review defines decision aids as:
Interventions designed to help people make specific and deliberative choices among options (including the status quo) by providing (at the minimum) information on the options and outcomes relevant to a person’s health status. The aid also may have included information on the disease/condition, costs associated with options, probabilities of outcomes tailored to personal health risk factors, an explicit values clarification exercise, information on others’ opinions, a personalized recommendation on the basis of clinical characteristics and expressed preferences, and guidance or coaching in the steps of decision making and in communicating with others.
Decision support proactively involves patients (and their physicians) in becoming better-informed users of health care. Their use implies knowing how to access evidence about how well (and under what circumstances) health care prevention, screening, and treatment works and explicitly incorporates the patient’s values in the decision process. Studies of decision support show that patients who use decision support are more likely to choose options that are more in line with evidence-based recommended practices and are less likely to experience conflict about making decisions. Some studies have shown reduced utilization of costly procedures in settings where evidence is lacking that such procedures have a beneficial clinical impact.
Decision support may be as straightforward as informing the patient of the existence of an evidence-based treatment that is recommended for most people like them and describing the potential benefits and harms. For example, a U.S. Preventive Services Task Force guideline on primary prevention of coronary heart disease recommends using low-dose aspirin when the 5-year risk for heart attack or coronary death exceeds 3% (this is determined from the Framingham equations). However, this recommendation is based on balancing the risks (bleeding) and benefits in a large population. Decision support for the individual might include using one of the many heart disease risk assessment tools and, if the risk exceeds 3%, comparing this risk with that of bleeding and one’s own viewpoint on taking medication.
A more complex decision that benefits from decision support might be for a 65-year-old man with apparently localized prostate cancer. He can choose among watchful waiting, radiotherapy, prostatectomy, brachytherapy, and alternative medicine. Each option has its likelihood of 10-year disease-free survival, profile of risks, costs, and impact on quality of life during and after treatment. In this situation, the patient’s values play a large role in choosing the best option—the one that balances risk and benefit for the individual.
Detsky, A., et al. (1997). Primer on medical decision analysis: Part 2 – building a tree. Medical Decis Making, 17(2), 126-135.
International Society for Pharmacoeconomics and Outcomes Research. (2005). Retrieved from http://www.ispor.org/
Naglie, G., et al. (1997). Primer on medical decision analysis: Part 3 – Estimating probabilities and utilities. Medical Decis Making, 17(2), 136-141.
O’Connor, A., et al. (2004). Decision aids for people facing health treatment or screening decisions. The Cochrane Database of Systematic Review.
Patient Decision Aids. (2005). Ottawa Health Research Institute. Retrieved from http://decisionaid.ohri.ca/index.html
Society for Medical Decision Making. Retrieved from http://www.smdm.org/
Sox, H., et al. (1988). Medical decision making. Boston: Butterworth-Heinemann. This is the classic text, still used in medical and informatics schools.
Demand management is “the use of self-management and decision support systems to enable, educate, and encourage people to improve their health and make appropriate use of medical care.” Demand management is typified by health call centers (also called telephone triage or telephone health information).
More specifically, demand management services are typically offered to assist callers in evaluating medical symptoms and to offer decision support. Typically, these phone lines (also known as health or “nurse” call lines) are staffed on a 24/7 basis by nurses, health educators, or others who have particular expertise in a particular area relating to the caller’s situation and/or condition. To ensure that advice provided adheres to evidence-based medicine, the decision-making process of demand management personnel is typically supported with algorithms or care guidelines that have been adopted by the host organization
As demand management includes teaching subscribers to become more skillful users of medical care, it is essentially a demand side effort. For example, the service is designed with the objective of directing the callers to appropriate but less expensive care management or treatment options.
Peterson, K., and Kane, D.(2001). Disease management – asystems approach to improving patient outcomes. Chicago: AHA Press.
URAC. “URAC Expands Scope of Health Call Center Accreditation Program.” Press release. September 15, 1999.
Vickery, D. and Lynch, W. (May 1995). Demand management: Enabling patients to use medical care appropriately. Journal of Occupational and Environmental Medicine, 37(5), 1-7.
The denominator, which is the lower part of a fraction used to calculate a rate, proportion, or ratio, is extremely important to the evaluations of population health management outcomes because it depicts the population (group with a specific diagnosis or overall population) of interest. The denominator must be carefully defined so that the evaluation actually measures what the study aims to evaluate.
For example, if the 1-year incidence of cancer in a given population is 1/10, this means that 1 person out of 10 contracts cancer in the year; 10 is the denominator.
The Care Continuum Alliance Selection Criteria Workgroup has developed a fundamental approach to guide denominator specifications for comparison of chronic care management programs for the five chronic conditions.
In assessing changes in statin use due to a population health management intervention, the denominator may consist of patients only with a principal cardiac diagnosis, members with other diagnoses indicating heart disease, or the total covered population.
See eligibility, enrollment, exposure, incident, and prevalent.
Care Continuum Alliance. (2010). Outcomes guidelines report, vol. 5. Washington, DC: Care Continuum Alliance.
The dependent variable in a population health management study is that variable (measured as an outcome or statistic) that is affected by the independent or causal variables.
Diagnosis-related groups (DRGs)
Diagnosis-related groups are a prospective payment system used to classify illnesses and based primarily on the patient’s diagnosis. Treatment procedures performed, age, gender, and discharge status are also taken into account in the classification. Hospitals use DRGs to group all charges for inpatient services into a single “bundle” for payment purposes.
DRGs were initially developed to provide data to hospital administrators or practice patterns and to enable the ability to influence physician behavior. In 1983, DRGs became the basis for Medicare’s prospective payment system for hospitals and other Medicare providers and suppliers. Several variants of DRGs have since been developed for non-Medicare populations.
“Diagnosis-related groups.” (2002). McGraw-Hill concise dictionary of modern medicine. The McGraw-Hill Companies, Inc.
“Diagnosis-related groups.” (2008). Mosby’s dental dictionary (2nd ed.). Philadelphia: Elsevier, Inc.
Generally, disability refers to any long- or short-term reduction of a person’s capacity for activity as a result of an acute, chronic, physical, mental, or emotional condition. A strict definition of disability may be found in the Social Security law and is based on a combination of physical condition and inability to work.
The term disability (as defined in Section 223 of the Social Security Act) means –
- “Inability to engage in any substantial gainful activity by reason of any medically determinable physical or mental impairment which can be expected to result in death or which has lasted or can be expected to last for a continuous period of not less than 12 months; or
- In the case of an individual who has attained the age of 55 and is blind (within the meaning of “blindness” as defined in section 216(i)(1)), inability by reason of such blindness to engage in substantial gainful activity requiring skills or abilities comparable to those of any gainful activity in which he has previously engaged with some regularity and over a substantial period of time.”
An individual is disabled under Social Security rules (and eligible for benefits) if the individual cannot do work that the individual did before and is unable to adjust to other work because of a medical condition(s). Disability must also last or be expected to last for at least 1 year or to result in death. Employers, workers’ compensation programs, and private insurers may apply different standards of incapacity in determining disability. For example, some programs may apply the concept of partial disability when an individual is impaired but able to perform some (but not all) duties, and a partial benefit may be payable.
See employee productivity, terms associated with.
“Disabilities/Limitations.” National Center for Health Statistics. U.S. Department of Health and Human Services. Retrieved from http://www.cdc.gov/nchs/fastats/disable.htm
Kraus, L., Stoddard, S. and, Gilmartin, D. (1996). Chartbook on disability in the U.S. access to disability data. U.S. National Institute on Disability and Rehabilitation Research. Retrieved from http://www.infouse.com/disabilitydata/disability/appendices_surveys.php#disability
Social Security Online. U.S. Social Security Administration. Retrieved from http://www.ssa.gov/
Disease management is a system of coordinated health care interventions and communications for populations with conditions in which patient self-care efforts are significant.
- Supports the physician or practitioner/patient relationship and plan of care;
- Emphasizes prevention of exacerbations and complications utilizing evidence-based practice guidelines and patient empowerment strategies; and
- Evaluates clinical, humanistic, and economic outcomes on an on-going basis with the goal of improving overall health.
Disease management components include:
- Population identification processes;
- Evidence-based practice guidelines;
- Collaborative practice models to include physician and support-service providers;
- Patient self-management education (may include primary prevention, behavior modification programs, compliance/surveillance);
- Process and outcomes measurement, evaluation, and management; and
- Routine reporting/feedback loop (may include communication with patient, physician, health plan and ancillary providers, and practice profiling).
Full-service disease management programs must include all six components. Programs consisting of fewer components are disease management support services.
The focus of disease management is on chronic conditions with certain characteristics that make them suitable for clinical intervention:
- Once contracted, the disease remains with the patient for the rest of the patient’s life;
- The disease is often manageable with a combination of pharmaceutical therapy and lifestyle change; and
- The average cost to some chronic patients is sufficiently high to warrant the expenditure of resources by the health plan or employer to manage the condition.
Traditionally, disease management has focused on the “big five” chronic diseases: Ischemic heart disease, diabetes, chronic obstructive pulmonary disease, asthma, and heart failure. Disease management programs are generally offered telephonically, involving interaction with a trained nursing professional, and require an extended series of interactions, including a strong educational element. Patients are expected to play an active role in managing the disease.
Because of the presence of co-morbidities or multiple conditions in most high-risk patients, this approach may become operationally difficult to execute, with patients being cared for by more than one program. Over time, the industry has moved more toward a whole person model in which all the diseases a patient has are managed by a single population health management program.
Care Continuum Alliance. (2010). Outcomes guidelines report, vol. 5. Washington, DC: Care Continuum Alliance.
A study in which at least two separate groups receive the experimental medication or procedure at different times, with neither group being made aware of when the experimental treatment or procedure has been given. The purpose of a double-blind study is to eliminate the risk of prejudgment by the participants, which could distort the results. Double-blind studies are often chosen when a treatment shows particular promise and the illness involved is serious. It can be hard to recruit human subjects for a blinded study of a promising treatment when one group will receive only a placebo or an existing medicine. A double-blind study may be augmented by a cross-over experiment, in which experimental subjects unknowingly become control subjects, and vice versa, at some point in the study.
Drug-drug interactions are changes in a drug's effects caused by another drug taken during the same time period. Types of drug-drug interactions include duplication, antagonism, alteration of how the body metabolizes one or both drugs, and magnification of the drugs’ effects because of impaired metabolism due to impaired function of the organ system responsible for drug elimination.
See also patient safety.
The term drug interaction refers to the potential for one drug to alter the intensity of pharmacological effects of another drug being taken concurrently. The net result may be enhanced or diminished effects of one or both of the drugs or the appearance of a new effect that is not seen with either drug alone. The drugs involved can be prescription medications, over-the-counter medicines, and even vitamins and herbal products.
Drug-drug interactions occur when a drug interacts, or interferes, with another drug.
This can alter the way one or both of the drugs act in the body or cause unexpected side effects.
- One drug might reduce or increase the effects of another drug;
- Two drugs taken together may produce a new and dangerous interaction; or
- Two similar drugs taken together may produce an effect that is greater than would be expected from taking just one drug.
Prescription drugs also can interact with each other.
- Mixing antidiabetic medication (e.g., oral hypoglycemics) and beta-blockers (e.g., Inderal) can result in an increased response of the antidiabetic drug and increased frequency and severity of low blood sugar episodes. The presence of liver disease can aggravate this response even further.
- Mixing antidiarrheal medication (e.g., Lomotil) and tranquilizers (e.g., Transxene, Valium), sedatives (e.g., Dalmane, Quaalude), or sleeping pills (e.g., Amytal, Nembutal, Seconal) can result in an increased effect of tranquilizers, sedatives, or sleeping pills.
- Mixing antihypertensive medication (e.g., Reserpine, Aldoril, Combipres) and digitalis (e.g., Lanoxin) can result in abnormal heart rhythms.
- Mixing anticoagulants (e.g., Coumadin, Warfarin) and sleeping pills (e.g., Nembutal, Amytal, Seconal) can result in decreased effectiveness of the anticoagulant medication.
In addition to prescription medications, over-the-counter medications can interact with each other. For example, taking a cough medication with alcohol at the same time as an antihistamine medication can increase drowsiness and decrease alertness; mineral oil taken with fat-soluble vitamins (A, D, E, K) can decrease the absorption of the vitamins.
In addition to interacting with each other, over-the-counter medications can also interact with prescription medications.
- Aspirin can significantly increase the effect of blood-thinning drugs (anticoagulants), thus increasing the risk of excessive bleeding.
- Antacids can cause blood-thinning drugs (anticoagulants) to be absorbed too slowly.
- Antacids can interfere with drug absorption of antibiotics (i.e., tetracycline), thereby reducing the effectiveness of the drug in fighting infection.
- Antihistamines, often used for allergies and colds, can increase the sedative effects of barbiturates, tranquilizers, and some prescription pain relievers.
- Decongestants in cold and cough medications can interact with diuretics or “water” pills to aggravate high blood pressure.
- Iron supplements taken with antibiotics can reduce or stop the ability of the antibiotics to fight infection. (The chemicals in the supplement and the antibiotic bind together in the stomach, instead of being absorbed into the bloodstream.)
- Salt substitutes provide significant exogenous potassium which can result in elevated blood potassium levels. This hyperkalemia may occur because of interaction with “water” pills or blood pressure medication or may occur in the face of chronic kidney disease. This can result in nausea, vomiting, muscle cramp, diarrhea, muscle weakness, or even cardiac arrest.
“Customer Communications.” Customer Support. First data bank. Retrieved from http://www.firstdatabank.com/customer_support/customer_communications/
“Drug Interations.” The Merck manuals online medical library. Retrieved from http://www.merck.com/mmhe/sec02/ch013/ch013c.html
NDDF manual. San Bruno, CA: First Data Bank.
Nies, A. (2001). “Principles of therapeutics.” The pharmacological basis of therapeutics. New York: McGraw-Hill.
Pharmacy and You. (2005). American Pharmacists Association. Retrieved from http://www.pharmacyandyou.org/
Dually Eligible (Dual Eligible)
This term references the population of individuals who are entitled to Medicare (Part A and/or Part B) and who are also eligible for Medicaid because of their income level. Some dually eligible individuals are now included in the Centers for Medicare and Medicaid Services Medicare Health Support Pilots that extend chronic care management services to this population. This population tends to be older and have a greater number of co-morbidities than does the Medicaid population alone.
“Dual eligibility.” Centers for Medicare and Medicaid Services. Washington, DC: U.S. Department of Health and Human Services. Retrieved from http://www.cms.gov/dualeligible/