-29- 3. ASSESSING THE FEASIBILITY OF MEASURING QUALITY WITH CLAIMS DATA Chapter 2 highlighted that claims data are widely available, relatively inexpensive, and amenable to analysis because they are readily available in an electronic format. However, standardized quality measurement efforts that rely primarily on claims data are based on just a handful of indicators. Chapters 3 and 4 characterize the dimensions of clinical quality that could be evaluated with claims data. With this knowledge, we can identify the situations where the additional costs of using medical records might be justified by sufficiently improved quality information and also identify some incremental changes that could be made to administrative databases that would allow for significant gains in quality measurement without medical record review. This chapter addresses two questions: (1) What dimensions of technical quality can be measured with claims data? (2) If incremental steps were taken to supplement the information in claims data, how would the capacity of electronic data for quality measurement increase? I address these questions by assessing the feasibility of constructing more than 550 quality of care indicators with claims data. In this context, feasibility is defined by whether all of the information required to construct an indicator is typically available, or could be reasonably well approximated, from claims data. Then, data elements that are not currently contained in claims data, but are found in other electronic clinical information systems are used to study how much the utility of claims data for quality measurement might increase if the additional data were available. The accuracy of quality measurement with claims data is addressed in Chapter 4. To provide background, the following are described: (1) the quality of care indicators that are used in this analysis, (2) how quality of care indicators are constructed with claims data, and (3) a -30- framework for characterizing the dimensions of quality that are more or less well suited for assessment with claims data. SELECTED INDICATOR SET: QA TOOLS Indicators from the RAND Quality Assessment Tools (QA Tools) system were used to evaluate the feasibility of quality measurement with claims data. QA Tools is a comprehensive and clinically detailed approach to quality measurement (McGlynn, Damberg et al. 2000).9 The QA Tools system includes 553 indicators that relate to the adult population and span 37 clinical areas (Table 3.1).10 The QA Tools system measures processes of care and has been used to assess quality in communities across the nation, as well as within individual HMOs and provider groups (Asch, Kerr et al. 2001). ___________ 9 Other indicator sets that measure processes include the Healthplan Employer Data and Information Set (HEDIS), the CMS’s Health Care Quality Improvement Program (HCQIP), and the Joint Commission on Accreditation of Healthcare Organization’s (JCAHO) ORYX/Performance Measurement System. However, I selected the QA Tools system over these alternatives because it is the most extensive and included indicators corresponding to all of those found in the alternate measurement systems. 10 QA Tools also includes quality indicators specific to children. However, to limit the scope of this analysis, these indicators were not considered. -31- Table 3.1 Adult Clinical Areas in the QA Tools System Adult screening and prevention Hormone replacement therapy Alcohol dependence Human immunodeficiency virus Asthma Hyperlipidemia Atrial fibrillation Hypertension Benign prostatic hyperplasia Immunizations Breast cancer Low back pain Cataracts Orthopedic conditions Cerebrovascular disease Osteoarthritis Cervical cancer Pain management for cancer Cesarean delivery Peptic ulcer disease and dyspepsia COPD Pneumonia Colorectal cancer Prenatal care and delivery Congestive heart failure Prostate cancer Coronary artery disease Tuberculosis Depression Upper respiratory tract infection Diabetes mellitus Urinary tract infections Family planning Uterine bleeding and hysterectomy Headache Vaginitis & STDs Hip fracture SOURCE: (Asch, Kerr et al. 2000; Kerr, Asch et al. 2000; Kerr, Asch et al. 2000) Indicators included in the QA Tools systems depended on (a) the presence of evidence indicating a specific intervention improves care, (b) the proximity of the intervention to improved health, and (c) the data that were believed to be available from medical record data. The types of health care processes assessed by the indicators was determined, in part, by the variation in the level of evidence to support causal relationships between health care processes and outcomes. The quantity and rigor of scientific evidence linking patient education to improved health outcomes, for example, is not as substantial as the literature that demonstrates links between specific medications and improved health outcomes.11 Without an established process-outcome link, a valid indicator cannot be developed. ___________ 11 The varying degrees and quality of evidence for the different modes of care is reflected by the contents of The Cochrane Library, an electronic publication that is published quarterly (www.updateusa.com/cochrane). The Library includes nearly 2500 abstracts of systematic reviews from around the world on the effects of health care interventions. The Library contains, for example, 772 reviews on “drug-therapy,” but only 52 reviews on “patient-education.” -32- It is the provision of specific services that are most closely linked to good outcomes. Therefore, relatively few QA Tools indicators assess the care modalities of admission and visit. What is actually done during an admission or visit is more important clinically than the encounter itself. The QA Tools indicators were written to be constructed with data abstracted from medical records (McGlynn, Damberg et al. 2000). If a proposed indicator required information that could not be obtained from medical records, then it was not accepted into the QA Tools system. For example, medical records are generally not considered to be a good data source to assess education because providers often fail to document the education and counseling delivered during the course of a health care encounter (Stange, Zyzanski et al. 1998).12 Thus, relatively few indicators assess whether education was provided. However, education indicators, such as advising smokers to quit or providing diabetics with dietary counseling, were included in the QA Tools system if there was professional consensus that failure to document the provision of the indicated care is consistent with poor quality. The fact that the availability of data from medical records framed the development of the indicators represents the principal value of using the QA Tools indicators for this analysis. That is, the QA Tools indicators allow me to consider a large set of indicators that represent what we would like to know about the delivery of care without the bias due to what information is typically considered to be unavailable from claims data. CONSTRUCTING QUALITY OF CARE INDICATORS Data elements are the building blocks for constructing quality of care indicators. As illustrated by the following example, one or more data elements are required to construct an indicator: ___________ 12 The dearth of good information on the provision of education from any type of data limits the ability of researchers to develop valid quality measures to assess education. -33- Men under age 75 with preexisting heart disease who are not on pharmacological therapy for hyperlipidemia should have total cholesterol, HDL, and LDL levels documented at least every five years. To construct the indicator, the criteria must be identified to define whether an individual is eligible for and has received the indicated care. The following types of data elements are needed to determine whether a patient satisfies the eligibility criteria: gender, age, disease history, and medication use. Among eligible patients, those who had their total cholesterol, HDL, and LDL tested in the past five years would pass the indicator. To assess whether it is feasible to use a specific data source to construct a quality of care indicator, we need to know whether each of the data elements involved in the eligibility and scoring criteria are either available or can be reasonably well approximated from the data source of interest. Data elements may be available from more than one type of data source (e.g., a patient’s gender might be determined through claims data, medical records or a patient survey), and some quality measurement efforts may be able to employ more than one data source. However, the goal of the current analysis is to determine the dimensions of clinical quality that could be assessed with claims data alone. CONCEPTUAL MODEL FOR CHARACTERIZING QUALITY MEASURES This analysis focuses on quality of care indicators that measure processes. Processes can be described by the following: • Setting: inpatient, ambulatory, home • Appropriateness of care: underuse, overuse, misuse • Specific conditions: diabetes, asthma, urinary tract infection, atrial fibrillation, etc. • Function of care: screening, diagnostic, treatment, follow-up • Type of care: preventive, acute, chronic -34- • Modality of care: admission, education, history, immunizations, laboratory services, medication, physical examination, surgery, visits to a health professional Although processes can be classified in many ways, I use the modality of care to organize and guide the analyses reported here. The function and type of care are used as the secondary means of classification to analyze and describe the type of indicators that can be constructed with claims data. A discussion of these classifications and how the QA Tools indicators are distributed across them is presented below. Primary Means of Indicator Classification: Modality The modality of care characterizes the intervention or service provided to a patient. I selected modality to characterize the dimensions of quality that can be assessed with claims data because the availability of information from claims data depends directly on the modality of care. Specifically, entries in claims records are triggered by claims for payment (to providers) or reimbursement (to consumers) for billable events. Therefore claims data include information about billable modes of care such as office visits, inpatient stays, laboratory tests, or filled prescriptions. In contrast, claims data do not have information about care modalities such as taking a patient history, providing health education and counseling, or performing the basic elements of a physical examination because these interventions generally do not qualify for separate payments and thus do not result in a claim for payment. Table 3.2 defines the common modes of care, specifies the number of QA Tools indicators that assess each of them, and indicates whether the mode is typically billable. The developers of the QA Tools systems classified the modes of care assessed by each indicator. Table 3.2 reflects the primary classification; of the 553 indicators, 36 had a secondary classification for mode. classifications were not referenced. In this analysis, secondary Modes of care were classified as typically billable if there are corresponding standardized codes (from -35- ICD-9-PCS, CPT, HCPCS, UB-92, NDC) that can be entered on claims forms. For example, claims for hospital admissions can be submitted for payment using procedure and revenue codes, but there are no procedure codes specific to education activities. There are ICD-9 V-codes that can be coded for education, if people are seeking consultation without complaint or sickness, however these codes are not used to determine the rate of payment. -36- Table 3.2 Modalities of Care: Definitions, Distribution in QA Tools, and Relationship to Billing Modality Definition Number of QA Tools Indicators 9 Billable? Yes Admission Patient admitted to a hospital for care in an inpatient setting. Education Counseling or information provided by a health care provider that informs patients of health behaviors (e.g., smoking cessation, blood sugar monitoring, dietary modifications) that contribute to improved health outcomes. 32 No History Provider documentation of chief complaint, associated symptoms, review of systems, medication use, allergies, and prior medical problems and treatments. 75 No Immunization Administration of a vaccine. 18 Yes Laboratory Services Performance of laboratory and radiology tests. 149 Yes Medication Medications prescribed by the physician. 120 Yes Physical examination An examination of specified system functions, documentation of vital signs and appearance of the patient. 88 No Surgery Performance of a surgical procedure. 36 Yes Visit A patient encounter with a health care provider in an ambulatory setting (i.e., doctor’s office, outpatient clinic, emergency room). 10 Yes Other intervention Delivery of health care services not defined by the modes of care listed above 16 Yes -37- (e.g., physical therapy, hearing evaluation). Since claims data exist primarily to pay for health care, I anticipate that modality of care will provide a useful framework for understanding the types of quality of care indicators that can be constructed with claims data. To better understand the capacity of claims data for quality assessment, I will also characterize the indicators that can be constructed with claims data by the function and type of care. Secondary Means of Indicator Classification Function. The function of care (screening, diagnosis, treatment, and follow-up) relates to why a service was performed. Figure 3.1 shows that among the QA Tools indicators, various modalities of care can be employed to screen, diagnose, treat or follow-up on a condition. Some modes of care are used primarily to implement a single function of care, while other modes are employed across the continuum of care. For example, medication is generally used to treat a condition, but is not a common intervention for screening, diagnosis, or following-up on a condition. In contrast, laboratory tests are performed to screen, diagnose, inform treatment and provide follow-up for a disease. -40- Type. Many quality of care reporting efforts evaluate preventive, acute and chronic care separately.13 Given this tendency, it is important to consider which types of care are more or less amenable to measurement with claims data. Figure 3.3 illustrates that multiple care modalities are used to assess each type of care. Again, if the modes of care are categorized as being billable or not, it suggests that claims data exist for the measurement of some indicators of preventive, acute and chronic care (Figure 3.4). ___________ 13 The Foundation for Accountability (FACCT) has developed a framework to communicate health care quality information to consumers (Lansky 1998). Many organizations, including NCQA, have adopted the FACCT framework for widespread quality reporting efforts. The FACCT model organizes comparative information about quality performance into five categories: The Basics, Staying Healthy, Getting Better, Living with Illness, and Changing Needs. The Staying Health, Getting Better, and Living with Illness categories involve information on the quality of care in preventive, acute and chronic care, respectively. -43- Summary The availability of information from claims data is directly related to the primary function of claims data, namely, payment for health care services. Although all modes of care can contribute to overall quality, services within only some modes typically affect payment. Given the relationship between the modality of care and the likelihood of information being available from claims data, I will use the modality of care to predict the dimensions of quality that can be assessed with claims data. I will also describe the dimensions of quality that can be measured with claims data in terms of why care should be delivered (i.e., the function of care) and for what kind of condition (i.e., type of care). METHODS Four steps were taken to characterize the dimensions of quality that can be assessed with claims data. First, I identified the data elements required to construct each of the 553 QA Tools indicators. Next, I classified each of the data elements into 22 general categories. I then assessed whether each data element is available from claims data. Finally, I determined whether it is feasible to construct the indicators with claims data based on the availability of data elements. These four steps are now detailed. Step 1: Identifying Data Elements To identify the data elements required to construct each of the QA Tools indicators, I used the analysis plans developed at RAND to construct the indicators with medical records data. The RAND analysis plans have a single eligibility and scoring statement for each indicator; these statements refer to specific variables that were defined through the medical records data abstraction process. I decomposed these eligibility and scoring statements into individual data elements. Consider, as an example, the following indicator: -44- A theophylline level should be obtained for patients on theophylline who present with an exacerbation of COPD. For this indicator, the data elements required to determine eligibility are: • • Date of COPD exacerbation Dates of prescriptions for theophylline Similarly, the data element for passing is: • Dates theophllyine level checked A total of 982 unique data elements were identified as being required to construct the 553 indicators from the QA Tools system. If a data element was required to construct more than one indicator, it was counted only once. For example, gender was required to construct 27 of the indicators, but represents just one of the total 982 data elements.14 For the current analysis, general descriptions of the data elements were listed. Detailed specifications such as the relevant ICD- 9-CM, CPT, or NDC codes were not identified for analysis in this chapter.15 The analysis described in Chapter 4 generates detailed specifications for a sub-set of the QA Tools indicators. Step 2: Classifying the Data Elements After the data elements required to construct each of the QA Tools indicators were identified, they were classified into 22 primary categories. These categories provide a mechanism for describing the types of information that can and cannot be obtained from claims data. Sub-categories were also assigned to some of the data elements to further detail the types of information needed to assess the quality of care. Table 3.3 lists the primary and secondary data element categories that were assigned. ___________ 14 If data elements were counted each time they were required to construct an indicator, the count of data elements would have been 2,463. 15 Specific code values would be required to actually construct the indicators with claims data. -45- Table 3.3 Categories Used to Classify Data Elements Primary Category Age Sub-Categories Counseling/discussion Diagnosis Current diagnosis, duration, history, new, ruleout, severity Documentation Contraindication, other Encounter Follow-up, new to provider, provider type, routine care, specific service, time Gender History Diagnosis, drug/alcohol use, family, functional status, immunization, lab result, medication, medication allergy, physical activity, procedure, sexual, smoking status, social, treatment, other Immunization Laboratory service performed Laboratory-results Laboratory-time Medical equipment Patient preference Physical exam Findings, vitals, other Prescription Daily dose, duration, filled, hospital administration, initial, ordered, route Procedure-findings Procedure-performed Procedure-time Race -46- Referral Setting Ambulatory, critical care bed, ER, ER/hospital, hospital, office, phone Signs & symptoms Table 3.4 summarizes the distribution of the types of data elements required to construct the 553 QA Tools indicators and highlights that the data elements required to define eligibility typically differ from those required to score a quality of care indicator. For example, data elements that characterize a patient, such as age, diagnosis and gender, are generally needed for eligibility, while data elements for health care interventions such as counseling or discussion, performing laboratory tests and procedures, and prescribing medications are more frequently required for scoring. -47- Table 3.4 Distribution of Data Element Types across Components of Indicators* Age Counseling/discussion Diagnosis Documentation Encounter Gender History Immunization Laboratory service performed Laboratory-results Laboratory-time Medical equipment Patient preference Physical exam Prescription Procedure-findings Procedure-performed Procedure-time Race Referral Setting Signs & symptoms Eligibility Scoring Indicator 64 5 406 2 27 27 90 0 3 0 41 8 12 35 0 60 19 68 64 44 409 14 62 27 140 19 70 64 6 0 2 23 78 26 57 2 1 0 99 95 9 1 3 1 76 127 4 141 6 0 1 16 30 64 7 3 3 98 168 30 166 8 1 1 108 120 * The columns in this table report the number of eligibility statements, scoring statements, and complete indicators that require the different types of data elements. Step 3: Assessing the Availability of Data Elements in Claims Data After the data elements required to construct the QA Tools indicators were identified and classified, I assessed whether they were available in claims data. To do so, I assumed claims data to include basic demographic information (e.g., gender, age), and claims for hospitalizations, ambulatory care, and outpatient pharmacy.16 Although ___________ 16 The availability of pharmacy claims data depends on the benefit package of the health plan. If a health plan does not offer a pharmacy benefit or if outpatient prescriptions are carved-out and paid by another entity, then pharmacy claims are not likely to be a component of the health plan’s claims data. However, for the purpose of this feasibility analysis, I have assumed outpatient pharmacy claims to be a component of a health plan’s claims data. -48- there are operational systems that electronically capture other types of information, including laboratory results, radiology findings, and medications administered during hospitalizations, they were not considered to be routinely available in claims data because these data are generally used and controlled by providers, not the health plan where the claims data reside. A data element was considered to be available if it could be found directly in claims data by referencing a specific type of code (e.g., ICD-9-CM, CPT, or NDC) or data field (e.g., gender), or if the data element could be approximated with claims data through reasonable assumptions. Information that can be directly obtained from claims data includes (a) hospital admissions or office visits for a specific condition, (b) the types of medications dispensed at outpatient pharmacies, and (c) various laboratory tests or other procedures. For many other types of information, claims data do not directly represent the information of interest. For example, there are no data fields in claims files for aspects of a patient’s diagnostic history. However, claims data can be used to gauge imperfectly whether a patient has a history of a certain condition by looking in the past for the diagnosis codes of interest. If the claims data did not include the diagnosis code of interest for the patient, it suggests the patient does not have a history of the condition. Step 4: Feasibility of Constructing Indicators with Claims Data Based on the availability of data elements, I determined whether each QA Tools indicator could be constructed with claims data. I concluded it is feasible to construct the indicator with claims data if all data elements required to determine both eligibility and scoring were available or could be approximated. The process of listing, classifying and coding the availability of data elements is illustrated below for two indicators and depicted in Figures 3.5 and 3.6. Example 1: An indicator that can be constructed with claims data Figure 3.5 expands upon the data element analysis for the following indicator discussed earlier in this chapter: -49- A theophylline level should be obtained for patients on theophylline who present with an exacerbation of COPD. The three data elements required to construct the indicator are date of COPD exacerbation, dates of prescriptions for theophylline and the dates on which the patient’s theophllyine level was checked. Row 1 of Figure 3.5 indicates that a visit for a COPD exacerbation is required to construct eligibility; the data element was classified as a diagnosis with a secondary classification of severity. Since there is a specific ICD-9-CM code (491.21) that indicates a COPD exacerbation, there is a “Y” in the column titled “Available” to indicate that the data element can be obtained from claims data. Only patients who were taking theophylline at the time of the COPD exacerbation are eligible for this indicator. Although the provider is likely to document in the medical record the medications that the patient is taking at the time of the exacerbation, this would not be entered in claims data. However, pharmacy claims can be used to approximate whether the patient was on theophylline at the time of the exacerbation (Row 2, Figure 3.5). To do this, we must assume that if a patient was on theophylline, that he filled the prescription within the health plan so that a claim was generated, and took the medication regularly. Then we can use the days supplied variable from the pharmacy claims to see whether a sufficient amount of medication was dispensed for the patient to be taking the theophylline at the time of the exacerbation. These assumptions are likely to exclude some patients who were taking theophylline and include others who were not taking the medication. Nevertheless, employing some assumptions, it is feasible to construct the eligible population with claims data (Row 3, Figure 3.5). The only data element required to score the indicator in this example is a list of the dates on which the patient had his theophylline level checked (row 4). Laboratory tests such as measurement of theophylline, can be identified with CPT or ICD-9 procedure codes; therefore, the data element is coded as being available from claims data. Since both the eligibility and scoring populations for this -50- indicator can be determined with claims data, it is feasible to construct this indicator with claims data. Figure 3.5 Illustrating the 4 Steps of the Feasibility Analysis: Example 1 Quality of Care Indicator: A theophylline level should be obtained for patients on theophylline who present with an exacerbation of COPD. Primary Row Data element classification ELIGIBILITY 1 Date of COPD exacerbation Diagnosis 2 Dates of prescriptions for theophylline Prescription 3 Feasible to determine eligibility? SCORING 4 Theophllyine level checked 6 Feasible to construct indicator? Severity Y Filled Y NA Y Yes Lab-performed 5 Feasible to determine scoring? Mode: Laboratory Function: Diagnosis Type: Acute Secondary Class Available Yes Yes Example 2: An indicator that cannot be constructed with claims data Consider the following indicator: Type 2 diabetics who have failed dietary therapy should receive oral hypoglycemic therapy. This indicator cannot be constructed with claims data. Eligibility is defined by people who have a diagnosis of Type 2 diabetes and continue to have uncontrolled blood sugar values after trying dietary therapy. The diagnosis of Type 2 diabetes (Figure 3.6, row 1) can be obtained from claims data. However, determining whether a patient has tried dietary therapy generally involves significant inaccuracy. If the -51- patient had a specific visit with a nutritionist or diabetic educator (row 2), there could be a procedure or revenue code for the visit that would indicate that the patient tried dietary therapy. On the other hand, if the patient’s physician provided dietary counseling during a regular office visit (row 3), the claims data are unlikely to capture that activity because education and counseling during office visits are rarely distinct billable services. An alternate way to discern whether a patient has tried dietary therapy is to assume that in the absence of pharmacotherapy (rows 4 and 5), dietary therapy is being used to treat the diabetes. With this asssumption, claims data can reasonably approximate whether the patient tried dietary therapy. If hemoglobin A1c (HbA1c) values are high, this suggests that the current therapy is not sufficiently controlling the patient’s blood sugar. Therefore, HbA1c values are needed to determine whether a patient fails dietary therapy. Although claims data can be used to determine whether HbA1c has been measured, the result of a laboratory test is not captured in claims data (row 6). Since claims data cannot be used to identify the diabetics on dietary therapy or those patients that are failing their current therapy, eligibility (Figure 3.6, row 7) cannot be determined with claims data. To score this indicator, we need to know whether an oral hypoglycemic agent was prescribed. is available in claims data. It is expected that this information When prescriptions are filled within the given health care system a claim is generated. This claim indicates the medication prescribed by the provider (row 8). However, no claim will exist if a patient fails to fill the prescription or obtains the medication outside the payment system. Nevertheless, the data elements required to score this indicator are expected to be in claims data because I assumed most filled prescriptions will result in a claim. -52- Figure 3.6 Illustrating the 4 Steps of the Feasibility Analysis: Example 2 Quality of Care Measure: Type 2 diabetics who have failed dietary therapy should receive oral hypoglycemic therapy. Primary Row Data element classification ELIGIBILITY 1 Diagnosis of Type 2 diabetes Diagnosis 2 Dietary counseling visit Encounter Mode: Medication Function: Treatment Type: Chronic Secondary Class Available NA Y Specific service Y NA N 3 Dietary/nutritional issues discussed Counseling/ discussion 4 Prescription for oral hypoglycemic agent Prescription ordered Y 5 Prescription for insulin Prescription ordered Y NA N ordered Y 6 HbA1C values Laboratory results 7 Feasible to determine eligibility? SCORING 8 Prescription for oral hypoglycemic agent No Prescription 9 Feasible to determine scoring? 10 Feasible to construct indicator? Yes No Summary of approach to feasibility analysis The data elements required to construct each of the 553 QA Tools indicators were identified, assigned to one of 22 categories, and assessed according to whether they could be determined or reasonably well approximated from claims data. The feasibility of constructing each indicator was then determined. Figures 3.5 and 3.6 illustrate this process. Some indicators can be constructed completely with claims data (Figure 3.5). For other indicators it is feasible to identify the -53- eligibility or scoring populations for the indicator, but not both (Figure 3.6). There are also QA Tools indicators for which claims data cannot be used to determine whether the eligibility or scoring criteria were satisfied. The distribution of the types of data elements available in claims data and the kinds of indicators that could be constructed with claims data are described below. FINDINGS: DATA ELEMENT ANALYSIS Almost 1000 unique data elements were identified as being required to construct the 553 QA Tools measures; 59% of these elements were categorized as being available from claims data. the data element analysis. Table 3.5 summarizes For each of the 22 primary data element classifications, the table includes (a) the number of data elements that were assigned to that category and (b) the proportion of the data elements in the category that were classified as being available in claims data. The categories are presented in descending order of proportion available from claims data. -54- Table 3.5 Number of Data Elements by Classification and Proportion Available in Claims Data Data element classification Age Gender Immunization Medical equipment Setting Laboratory service performed Procedure-performed Encounter Diagnosis Prescription Procedure-time Signs & symptoms History Physical exam Laboratory-results Counseling/discussion Documentation Laboratory-time Patient preference Procedure-findings Race Referral Total Count of Data Elements Proportion Available in Claims Data 1 1 7 3 27 61 1.00 1.00 1.00 1.00 1.00 0.95 84 23 219 174 4 71 117 68 42 30 16 2 3 26 1 1 982 0.92 0.91 0.90 0.88 0.25 0.23 0.15 0.03 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.59 As shown in Table 3.5, basic demographic information, namely age and gender, is readily available from claims data. Other information usually available in claims data includes diagnoses, the types of laboratory tests or procedures that were performed, the types of prescriptions dispensed, and the setting of the delivered care. The availability of these data elements is consistent with the types of information required to administer (e.g., identify health plan enrollees) and pay for health care services. At the other end of the spectrum, none of the data elements could be obtained from claims data in the counseling/discussion, documentation, laboratory-time, patient preference, procedure-findings, race, or referral classifications. -55- As described in Chapter 2, ICD-9-CM, CPT, HCPCS, UB-92 Revenue Codes and NDC are standard coding systems used in claims data. Although these coding systems convey information on diagnoses, delivered services, and dispensed medications, the availability of information within each of these domains is variable. The potential availability of this information from claims data is described below. Availability of information on laboratory tests Three primary classifications of data elements were assigned to laboratory services: performed, results and time. As shown in Table 3.5, information on whether laboratory tests were performed can almost always be determined with claims data (95% of the laboratory-performed data elements could be obtained from claims data).17 However, data elements related to the results or timing of laboratory services do not affect payment and are not a required component on submitted claims. Availability of information on procedures The data elements related to procedures were assigned to one of the following categories: performed, timing, and findings. The proportions of data elements for each of these classifications that could be obtained from claims data are 0.92, 0.25, and 0.00 respectively (see Table 3.5). Neither the findings of procedures (e.g., ejection fractions, mammogram suggestive of malignancy) nor the time18 at which procedures are performed are coded in claims data. The data elements in the procedure-performed category that could not be determined from ___________ 17 The three data elements from the laboratory-performed category that could not be obtained from claims data are whether the fluid from arthrocentesis of the knee was analyzed for (1) cell count, (2) gram stain, and (3) crystals. CPT code 20610 represents arthrocentesis of the knee, but there are no other codes to specify the types of analyses that were performed on the fluid. 18 The date on which a procedure is performed is available in claims data, but detailed information such as the hour and minute at which a procedure is performed is not available. Therefore, claims data are not sensitive to time-frames that are less than 24 hours. The one data element for time that was coded as being available from claims data was that an EKG be performed within 24 hours of admission. The other indicators required that procedures be performed in less than 24 hours, in these cases the “procedure-performed” data elements were coded as unavailable. -56- claims data were patient care services that generally are not paid separately (e.g., repositioning the patient) and therefore do not have a specific procedure code. Availability of information on medications Eighty-eight percent of the data elements in the prescription category could potentially be obtained from claims data (see Table 3.5). The sub-classifications assigned to data elements in the prescription category included daily dose, duration, filled, initial, hospital administration, ordered, and route (see Table 3.6). With the exception of the sub-category of hospital administration, all data elements in the prescription category were coded as being available from claims data. Although a hospital may have an information system that logs all medications administered to a patient during a hospitalization, the information tends not to become a part of health plans’ claims data because the administered medications typically do not affect payment to the hospital. Another issue related to inpatient medications is the ability to assess the timeliness with which they were administered. Even if health plans had access to inpatient pharmacy information, the time of administration relative to time of admission would remain unknown. -57- Table 3.6 Assessment of Data Elements in Claims Data: Prescription SubClassifications Prescription subclassification Count Daily dose Duration Filled Hospital administration Initial Ordered Route Total 11 6 47 12 8 72 10 174 Proportion Available from Claims Data 1.00 1.00 1.00 0.00 1.00 1.00 1.00 0.88 Availability of information on diagnoses The availability of the data elements in the diagnosis category also varies by sub-classification. As shown in Table 3.7, none of the data elements relevant to the duration of a diagnosis and one-quarter of the data elements that describe the severity of a diagnosis are potentially available in claims data. Disease severity can be determined with ICD-9-CM codes for only a few conditions. However, for some conditions a combination of pharmacy and encounter data can be used to characterize disease severity. For example, if a patient receives multiple asthma medications within a year, or has an emergency room visit or hospitalization with a primary diagnosis of asthma, the pattern of care warrants an inference of moderate to severe asthma (rather than mild or exercise induced asthma). -58- Table 3.7 Assessment of Data Elements in Claims Data: Diagnosis SubClassifications Diagnosis subclassification Count Current diagnosis Duration History New Rule-out Severity Total 141 2 28 31 1 16 219 Proportion Available from Claims Data 0.97 0.00 0.98 0.97 1.00 0.46 0.90 Almost all of the data elements relevant to whether a patient has a history of a diagnosis or a new diagnosis were classified as being available from claims data (see Table 3.7), even though ICD-9-CM codes typically do not specify new or preexisting conditions. Historical and new diagnoses were considered to be potentially available from claims data, assuming access to longitudinal claims data that could provide information that would with reasonable accuracy establish some of these clinical details. For example, if there is a diagnosis code for a specific condition, and no prior ICD-9-CM code for the condition of interest, then one could assume the diagnosis is new. Obviously, looking further back in claims or encounter data (e.g., 3 years rather than 6 months) increases the level of confidence associated with an inference concerning whether a diagnosis is new or preexisting. Summary of data element availability Almost 1000 data elements were identified as being required to construct the 553 QA Tools indicators. While 59% of these data elements were judged to be potentially available from claims data, availability varied widely by data element type (see Table 3.5). Although substantial information on diagnoses, laboratory tests, procedures, and medications is captured in claims data for payment, not all data elements related to these domains can be obtained from claims data. Clinically detailed information, such as the severity of a condition or the results and timing of laboratory services and procedures, is sometimes relevant to an indicator, but such information is not usually -59- available from claims data. Data elements related to history, physical exam, and signs and symptoms, which are frequently required for quality measurement, are also frequently unavailable from claims data. The implications of data availability for the types of indicators that are feasible to construct with claims data are described below. FINDINGS: INDICATOR ANALYSIS Table 3.8 reports the number of QA Tools indicators, by modality, that could be constructed with claims data. Of the 553 QA Tools indicators that were reviewed, 186 (34%) were characterized as being feasible to construct with claims data. This finding has strong practical implications because HEDIS, the most common set of measures used to measure performance, includes only 26 indicators of technical quality. This suggests that current measurement efforts fail to fully recognize the capacity of claims data for quality measurement. -60- Table 3.8 Number and Proportion of Indicators that can be Constructed with Claims Data – by Modality Modality Total # of Indicators Non-billable modes of care Education 32 History 75 Physical exam 88 Non-billable sub-total Billable modes of care Admission Immunization 195 9 18 Laboratory services 149 Medication 120 Other intervention 16 Surgery 36 Visit 10 Billable sub-total TOTAL 358 553 # of Indicators Feasible to Construct with Claims Data (proportion) Eligibility Score Indicator (Eligibility + Scoring) 11 (0.34) 63 (0.84) 67 (0.76) 141 (0.72) 3 (0.09) 3 (0.04) 20 (0.23) 26 (0.13) 1 (0.03) 1 (0.01) 11 (0.13) 13 (0.07) 3 (0.33) 12 (0.67) 99 (0.66) 56 (0.47) 9 (0.56) 10 (0.28) 10 (1.00) 199 (0.56) 340 (0.61) 9 (1.00) 18 (1.00) 124 (0.83) 91 (0.76) 10 (0.63) 31 (0.86) 10 (1.00) 293 (0.82) 319 (0.58) 3 (0.33) 12 (0.67) 87 (0.58) 48 (0.40) 5 (0.31) 8 (0.22) 10 (1.00) 173 (0.48) 186 (0.34) As depicted in Figure 3.7, only 7% of the indicators assessing nonbillable modes of care (education, history, and physical exam) could potentially be constructed with claims data.19 Among the indicators ___________ 19 I coded indicators that pertained to either physical reassessment or additional counseling as being feasible to construct with claims data based on the assumption that a follow-up encounter within a reasonably short time-interval could reveal whether the indicated care was delivered. Therefore, 14 of the indicators assessing -61- assessing modes that are typically billable, 48% could be constructed with claims data. non-billable modes of care were considered to be feasible to construct with claims data. -63- Inability to determine eligibility frequently limits the utility of claims data to construct indicators for billable modes of care. Although the scoring components of 82% (N=293) of the indicators for billable modes could potentially be constructed with claims data, there is sufficient information to determine the eligible population for only 56% (N=199) of these indicators. The proportion of indicators that can be constructed with claims data ranged from 0.27 to 0.55 by the function of indicated care (Table 3.9) and from 0.28 to 0.36 by the type of care being assessed (Table 3.10). This suggests that it is feasible to use claims data to measure the quality of clinical care for selected aspects of all types of conditions across the continuum of the care process. In contrast, the current set of HEDIS Effectiveness of Care measures, includes no measures for the quality of diagnostic care processes or for acute care. Table 3.9 Number and Proportion of Indicators that can be Constructed with Claims Data – by Function of Care Function of Care Total # of Indicators Screening 41 Diagnosis 208 Treatment 238 Follow-up 66 Total 553 # of Indicators Feasible to Construct with Claims Data (proportion) Eligibility Score Indicator (Eligibility + Scoring) 27 18 11 (0.66) (0.44) (0.27) 150 99 64 (0.72) (0.48) (0.31) 105 160 75 (0.44) (0.67) (0.32) 58 42 36 (0.88) (0.64) (0.55) 340 319 186 (0.61) (0.58) (0.34) -64- Table 3.10 Number and Proportion of Indicators that can be Constructed with Claims Data -- by Type of Care Type of Care Preventive Total # of Indicators 65 Acute 174 Chronic 314 Total 553 # of Indicators Feasible to Construct with Claims Data (proportion) Eligibility Score Complete Measure 31 40 18 (0.48) (0.62) (0.28) 119 96 63 (0.68) (0.55) (0.36) 190 183 105 (0.61) (0.58) (0.33) 340 319 186 (0.61) (0.58) (0.34) Summary of Feasibility Assessment Relative to current applications of claims data, broader measurements of quality with claims data appear to be feasible. The analysis identified 186 (34%) adult QA Tools indicators that could be constructed with claims data. Feasibility of constructing indicators is much more common when the relevant health care services are typically billable. However, lack of data elements required to determine eligibility frequently preclude construction of these indicators. Less than 60% of the eligibility statements associated with measures involving billable modes of care can be constructed with claims data. This is because clinically detailed information such as laboratory results, the findings of procedures, and signs and symptoms are frequently required to determine eligibility, but are not available from claims data. In sum, it is usually feasible to construct quality of care indicators with claims data if they (a) rely on diagnoses and demographic information to determine eligibility and (b) assess billable modes of care such as encounters, immunizations, laboratory tests, and outpatient prescriptions. The accuracy of using claims data to construct the feasible indicators is analyzed in the next chapter. The remainder of this -65- chapter investigates how the capacity for quality measurement could increase if additional information were available in claims data. INCREASING THE CAPACITY OF CLAIMS DATA FOR QUALITY MEASUREMENT The primary advantage of using claims data for quality measurement is that they are widely available in an electronic format for large numbers of patients. However, the assessment of data availability points to several types of information that generally cannot be obtained from claims data (see Table 3.5). Leading examples are examination findings, test results, timing of events, patients’ signs and symptoms, and the content of education and counseling activities. When these types of information are required to construct quality of care indicators, an alternate data source, such as medical records, is needed. Looking forward, standardized electronic medical records (EMRs) may one day be available to provide all information required for quality measurement in a standardized electronic format. In the interim however, I consider whether there are incremental steps that could be taken to supplement the information currently available in claims data. If so, how would the capacity for quality measurement with claims data increase? The analysis described presently addresses these questions. How Could Additional Information be Obtained? To analyze how supplemental information would affect quality measurement with claims data, I considered what health plans could do without computerizing medical records to incorporate additional electronic information with their claims data. Although other stakeholders—including providers, purchasers, and regulators—could take actions to improve quality measurement, the perspective of health plans is taken because they possess claims data and have the most capability to shape how they are used. For example, to increase their capacity for quality measurement with electronic data, health plans could: (a) link electronic data that already exist in operational systems20 to claims ___________ 20 Operational systems in laboratories, hospital pharmacies, and electrocardiography carts, for example, include most data electronically on laboratory results, medications administered during an inpatient -66- data files, (b) add data fields to existing claims forms, or (c) change the way the data currently submitted in claims are coded. If a health plan were to pursue one or more of these options, it would need to balance the potential gains for quality measurement and other claims functions against the costs of implementation. To frame this analysis, I identified four types of information— laboratory results, procedure findings, vital signs, and signs and symptoms—that are not in claims data, but are potentially amenable to quality measurement because they can be communicated through standardized codes. While claims data are typically limited to diagnosis (ICD-9-CM), procedure (CPT, ICD-PCS, HCPCS) and medication (NDC) codes, there are other standardized terminology systems, such as the Logical Observation Identifier Names and Codes (LOINC®) system21 and the Systemized Nomenclature of Medicine Reference Terminology (SNOMED®)22 that include codes for many more clinical concepts. How information about laboratory results, procedure findings, vital signs, stay, and electrocardiographic measurements (McDonald, Overhage et al. 1997). 21 LOINC was released publicly in 1996. The Regenstrief Institute at the University of Indiana developed and maintains the LOINC database. The database includes more than 25,000 observation concepts, including laboratory tests, vital signs, electrocardiographic measurements, intake and output measures, critical care, overall clinical impressions, and discharge summary. The American Clinical Laboratory Association (ACLA), an association of large referral laboratories whose members are responsible for more than 60% of U.S. outpatient laboratory test volume, has recommended LOINC for adoption by its members. The three largest commercial laboratories in the U.S. (representing approximately 30% of the market) have adopted LOINC as their code system for reportable test results, as have several health care systems (McDonald, Overhage, et al. 1997; Regenstrief Institute 2002). 22 SNOMED Clinical Terms (SNOMED CT®) was released in 2002 and is a concept-based reference terminology that includes more than 333,000 concepts related to pathology, diagnoses, findings, and symptoms. SNOMED CT is the combination of two established terminologies: SNOMED and Clinical Terms Version 3. SNOMED is a copyrighted work of the College of American Pathologists and was introduced in the 1970s. Clinical Terms Version 3 (formerly known as the Read Codes) is a copyrighted work of the National Health Service in the United Kingdom and was introduced in the 1980s. The National Health Service has mandated that electronic health records in the United Kingdom implement SNOMED CT in 2003; there is no similar mandate in the US. For more information about SNOMED, see www.snomed.org (accessed 2/18/2002). -67- and signs and symptoms might be obtained to supplement claims data is described presently. Laboratory results. Most, if not all, producer laboratories (i.e., where laboratory test results are determined) have electronic systems that identify the patient for whom the test was performed, the exact name of the test, and the result. Health plans, however, do not typically have access to this information. In order to obtain laboratory results, health plans could either (a) have a contractual agreement that source laboratories provide the data files that include laboratory test results, or (b) require that test results be submitted with claims for payment. Regardless of the mechanism for obtaining the laboratory results, a health plan needs to be able to understand the information it receives from the laboratories. Coding systems used to identify laboratory tests vary by laboratory, and health plans would need to receive information from multiple sources (e.g., commercial, hospital, and nursing home laboratories). If the source laboratories were all to use a single code system, such as LOINC, then the health plan would need to understand only the one system. However, if laboratories use alternate coding systems, including their own internal code values, the ability to link laboratory data with claims data becomes significantly more difficult. Although the laboratory results are likely to exist in an electronic form, the ability to use the information in conjunction with claims data will vary among health plans depending on the number of laboratories with which the health plan needs to coordinate and the degree of centralization among the producer laboratories about a common coding system. Procedure findings. In the absence of an electronic medical record (EMR), it is unlikely that providers have a centralized database that contains the findings from multiple types of procedures such as biopsies, ultrasounds, and electrocardiograms. To incorporate procedure findings with claims data, health plans would need to link across multiple operational systems, if they exist, or require providers to include results when they submit claims for payment. The ability of health plans to obtain data on procedure findings is likely to vary by -68- the providers with which they contract. Health plans with a relatively centralized system of providers, for example, will be in a better position to make these associations because the number of potential operational systems from which results would be obtained would be fewer. Among the providers with which health plans contract, those who have invested in information systems and have been cognizant of the ability to share information across multiple parties will be better able to share their information with health plans relative to providers who do not keep procedure findings in an electronic format, or keep them in an electronic format that does not use standardized terminologies and codes. Vital signs. Unless providers are using EMRs, it is unlikely that blood pressure values and heart rates are electronically reported in any database. Therefore, linking existing data or changing the way claims data are coded are not viable routes to obtain information about vital signs. For health plans to gain access to this information, they would need to modify their claims forms and require that health care providers submit this information on their claims for payment. The burden would be relatively low because only three numeric values (systolic blood pressure, diastolic blood pressure, and heart rate) would need to be recorded. Signs and symptoms. If all providers would maintain patient problem lists electronically and in a standard code system that could be understood by others, then health plans could link these problem lists to the existing claims data and have significantly richer information about patients’ conditions. lists is rare.23 However, electronic maintenance of problems Therefore, to obtain detailed information about patients’ presenting signs and symptoms the coding system currently used to communicate information about diagnoses would need to be altered. Currently, ICD-9-CM codes are used to describe the type of condition that is being addressed during the encounter, but they are often unable ___________ 23 Survey by medical records institute and SNOMED found that less than 1.5% of surveyed providers are using components of an EMR to support electronic problem list. http://www.medrecinst.com/resources/survey2002/overview.shtml (accessed 11/3/02). -69- to accurately represent the clinical problems being addressed. A more detailed coding system, such as SNOMED, has the potential to communicate a more clinically accurate description of patients’ presenting signs and symptoms and diagnoses in a standardized format. SNOMED has been shown to have more complete case identification for conditions relative to ICD-9-CM codes (Elkin, Ruggieri et al. 2001). Switching to an alternate system of diagnostic codes would be a significant change. To move from ICD-9 coding to SNOMED or another more clinically detailed system, health plans would need to provide significant support to providers. For example, a structured paper form or an electronic interface that allows providers to easily select the diagnostic observations and symptoms for a patient could be introduced or coding specialists could be hired to translate visit notes to SNOMED codes. Regardless of the approach, it would take significant resources to successfully change to a coding system that would provide meaningful information about signs and symptoms. Methods To analyze the potential gains of additional information to quality measurement, I examined the QA Tools indicators that cannot be constructed with claims data alone, but that could be constructed if the following types of data were available: • Laboratory results • Procedure findings • Vital signs • Signs and symptoms Of the 982 data elements required to construct the QA Tools indicators, 42 were for laboratory results, 26 for procedure findings, 11 for vital signs, and 71 for signs and symptoms. One or more of these data elements were required to construct 204 indicators.24 To analyze ___________ 24 Analysis earlier in this chapter found that of these 204 indicators, 184 (90%) cannot be constructed with claims data. Laboratory results were required for 64 indicators, procedure findings for 30 indicators, signs and symptoms for 120 indicators and vital signs were required for 31 indicators. Nineteen of the indicators that required signs and symptoms could be constructed with the current contents of claims data. -70- how quality measurement with electronic data would be enhanced if these types of data were available, I calculated the increases in the numbers and types of quality indicators that could be constructed by having each one of the additional data types available. The increase in capacity for quality measurement was also calculated assuming that all four types of data were available. The current analysis provides optimistic estimates of how much the capacity for quality measurement could be improved because I assumed that the additional information would meet all of the QA Tools requirements, and this is unlikely. For example, an indicator may require information about a symptom that is not coded in LOINC or SNOMED. Therefore, the results indicate the “upper bounds” for increases in the capacity for quality measurement with the additional data. Even if information about laboratory results, procedure findings, vital signs, and signs and symptoms required to construct the QA Tools indicators were available, there would still be indicators that could not be constructed because other types of data would remain unavailable (e.g., the content of patient education and counseling activities, components of patient history and provider documentation habits). Findings and Discussion Without additional information, the claims data for most health plans25 could support the construction of 186 of the 553 QA Tools indicators (Table 3.10). If in addition, a health plan was able to incorporate laboratory results, procedure findings, vital signs, and signs and symptoms with claims data, 260, or nearly half, of the QA Tools indicators could be constructed without use of medical records (Table 3.11). Information about signs and symptoms had the largest effect on the number of indicators that could be constructed and findings from procedures had the smallest effect. However, no single type of information increased the proportion of indicators that could be constructed by more than 6%. ___________ 25 In this context, the contents of claims data include claims for ambulatory encounters, outpatient ancillary services (e.g., laboratory and radiology tests), outpatient pharmacy, and inpatient encounters. -71- The increase in the number of indicators that could be constructed with the additional information is reported in Table 3.11 and depicted in Figure 3.9. Figures 3.9-3.11 show the results by mode, function and type of care; a table with the detailed numerical results is in Appendix A. Following Figures 3.8-3.11, there is a discussion of how each type of information could increase the capacity for quality measurement. Table 3.11 Potential Increase in Capacity for Quality Measurement – All Indicators Number of Indicators Feasible to Construct (%) Total # of Indicators 553 Original Lab Procedure Vitals + Signs & All Analysis Results + Findings + Original Symptoms additional Original Original + elements + 196 (0.35) Original 221 (0.40) Original 260 (0.47) 186 (0.34) 209 (0.38) 202 (0.37) -76- Description of effects of additional information on quality measurement. Laboratory results. If laboratory results were available and linked to claims data, then 23 additional QA Tools indicators could be constructed. The increased capacity for quality measurement comes from the ability to identify patients who satisfy eligibility criteria based on specific laboratory results (e.g., people with LDL > 130 mg/dl or HbA1c >= 9.5%). Since the QA Tools indicators measure processes, and not intermediate outcomes, laboratory results do not change the number of scoring statements that can be constructed. Procedure Findings. The availability of procedure findings, such as biopsy results, ejection fraction measurements, and ultrasound findings, increases the number of QA Tools indicators that can be constructed from 186 to 196. Of the four additional types of data elements that were examined, the availability of procedure findings had the smallest effect on increasing the capacity for quality measurement. Similar to the pattern with laboratory results, the availability of procedure findings increases the ability to determine who is eligible for an indicator, but does not help determine whether the indicated process was delivered. Blood Pressure Values and Heart Rates. If health plans had information on their enrollees’ blood pressures and heart rates, an additional 16 QA Tools indicators could be constructed. The availability of blood pressure values and heart rates would improve the ability to measure the quality of care delivered for cardiovascular conditions, especially hypertension. However, the ability to measure the quality of care for other types of conditions would remain unchanged. Signs and Symptoms. Of the four types of information that were analyzed, signs and symptoms would lead to the largest increase in the number of QA Tools indicators that could be constructed. If health plans were better able to identify their enrollees by the types of signs and symptoms for which they present for care, 35 additional QA Tools indicators could be constructed. Information about signs and symptoms would not increase the capacity to measure preventive care, but measurement of care for acute and chronic conditions would increase, -77- especially when assessing the diagnostic function of care (see Figure 3.11). Summary The purpose of this analysis was to determine whether incremental changes (i.e., changes short of implementing EMRs) could lead to significant gains in quality measurement. The results of this analysis suggest that the addition of any one type of information would not have a very large contribution to the number or type of indicators that could be constructed. However, if information about laboratory results, procedure findings, vital signs, and signs and symptoms were available to supplement claims data, the proportion of QA Tools indicators that could be constructed without medical record data would increase from about one-third to about one-half. Thirty-five additional indicators could be assessed if information on signs and symptoms were available, however results from laboratory tests and vital signs, which may be easier to obtain, would allow an additional 23 and 16 QA Tools indicators respectively to be constructed with claims data. As health plans consider what information to obtain for quality measurement, the gains to quality measurement must be compared to the costs associated with obtaining the data. Perceived benefits from quality measurement activities and the ability to obtain the different types of data elements are likely to vary considerably between health plans. For example, a health plan with a very young enrollment population would be much less interested in being able to measure quality about hypertension than a plan dominated by elderly enrollees, and therefore not be willing to invest the resources required to change their claims forms to include data fields for blood pressure values and to impose the reporting requirement on providers. Similarly, health plans with a more concentrated network of providers may be able to obtain additional data at a lower cost than a health plan with a very open network of providers. For example, a health plan that pays claims to two commercial laboratories rather than 15 is in a much better position to try to link laboratory results to claims data, other things equal. -78- Standard coding systems have been developed to communicate detailed clinical data electronically. However, minimal adoption of these standards limits a health plan’s ability to readily incorporate additional information with their claims data for quality measurement. Quality measurement will be significantly enhanced as additional pieces of clinical information are collected and stored in a standardized electronic fashion, but no one type of information would provide a dramatic improvement by itself.