Frequency distribution of international normalized ratio (INR) values, using results of prothrombin time test as the unit of analysis. Practice sites are described in the "Practices" subsection of the "Subjects and Methods" section. Percentages have been rounded and may not sum 100.
Frequency distribution of time in target range of international normalized ratios, using patient-day as the unit of analysis. Practice sites are described in the "Practices" subsection of the "Subjects and Methods" section. Percentages have been rounded and may not sum 100.
Time until next visit when last international normalized ratio (INR) was 2.00 to 3.00 (target range; A), last INR of at least 3.50 (high out-of-range; B), and INR of less than 1.50 (low out-of-range; C). Practice sites are described in the "Practices" subsection of the "Subjects and Methods" section. Percentages have been rounded and may not sum 100.
Samsa GP, Matchar DB, Goldstein LB, Bonito AJ, Lux LJ, Witter DM, Bian J. Quality of Anticoagulation Management Among Patients With Atrial FibrillationResults of a Review of Medical Records From 2 Communities. Arch Intern Med. 2000;160(7):967-973. doi:10.1001/archinte.160.7.967
Most treatment of patients at risk for stroke is provided in the ambulatory setting. Although many studies have addressed the proportion of eligible patients with atrial fibrillation (AF) receiving warfarin sodium, few have addressed the quality of their anticoagulation management.
As a comprehensive assessment of quality, we analyzed the proportion of eligible patients receiving warfarin, the proportion of time their international normalized ratios (INRs) were within the target range, and, when an out-of-target range INR value occurred, the time until the next INR measurement was made.
Retrospective review of the medical records of 660 patients with AF managed by general internists and family practitioners in Rochester, NY, and the Research Triangle area of North Carolina.
Only 34.7% of eligible patients with AF received warfarin. The INR values were out of the target range approximately half the time, and the response to these values was not always timely. For all the measures considered, both Rochester practices with access to an anticoagulation service had higher (albeit not ideal) quality of warfarin management than the remaining practices.
We found significant deficiencies in the practice of warfarin management and suggestive evidence that anticoagulation services can partially ameliorate these deficiencies. More research is needed to describe the quality of anticoagulation management in typical practice and how this management can be improved.
ATRIAL FIBRILLATION (AF) is the second strongest risk factor for stroke and can increase the risk for stroke 5-fold.1 Randomized trials have shown that treatment with warfarin sodium decreases the risk for stroke by 60% to 70% in comparison with a decrease of 20% to 25% with aspirin.2 Various guidelines and consensus statements support the use of warfarin among patients with AF.3,4
Because of its narrow therapeutic index, warfarin therapy must be monitored carefully. Levels below the target international normalized ratio (INR) range of 2.00 to 3.00 afford less protection against stroke, whereas levels above the target range increase the risk for bleeding.3
Arranged by increasing effort required for their collection, 3 measures of the quality of anticoagulation management among patients with AF are the proportion of eligible patients receiving warfarin, the proportion of time these patients' INRs are within the target range, and, when an INR value outside the target range (hereafter referred to as out of range) occurs, the proportion of time an appropriate response is made. Most studies have addressed the first measure of quality, and some studies have addressed the second, but few studies have addressed the third.5
Anticoagulation management is by no means trivial. For example, although most physicians agree with the recommendations to prescribe warfarin to patients with AF,6,7 in the absence of an adequate reminder system, it is easy to forget to order the medication. For patients receiving warfarin, laboratory test results may not be available until after the patient has left the clinic (thus complicating the process of dosage adjustment); inadequate record-keeping systems can result in the physician being unaware of laboratory results; and inadequate communication systems can lead to dosage changes being communicated to the patient late or not at all and to delays in rescheduling missed appointments. Many primary care physicians have relatively few patients undergoing anticoagulation therapy, and thus have little incentive to develop more efficient management systems.
These considerations suggest that the process and outcomes of anticoagulation management may be less than optimal. Anticoagulation services (ACSs) have been cited as a mechanism to improve the provision of anticoagulation.5,8 The rationale for ACSs is that the day-to-day details of anticoagulation management are delegated from the busy physician to an ACS manager (typically a pharmacist, physician's assistant, nurse, or nurse practitioner), who is responsible for dosing changes, scheduling, patient education, and other aspects of anticoagulation management. Because the ACS manager specializes in these tasks, economies of scale can be obtained, and the quality of care potentially can be improved.
Recognizing that most care of patients at risk for stroke is provided by internists and noninternist primary care physicians,9 our purpose was to examine comprehensively the quality of anticoagulation management by primary care physicians for their ambulatory patients with AF. Outcome variables address all 3 of the quality measures listed above. We also examined whether patterns of care differed in practices that had vs those that did not have access to an ACS.
We sought to identify practices with relatively large numbers of general internists and/or family practitioners, since these practices were likely to have many patients at high risk for stroke. For convenience, we focused on the Rochester, NY, area and the Research Triangle area of North Carolina (ie, Durham, Raleigh, and Chapel Hill). In the Research Triangle area, we identified practices with 5 or more general internists and/or family practitioners. Of 15 potentially eligible practices, 8 agreed to participate in the study. We worked directly with the participating practices.
In Rochester, we collaborated with a managed care organization that included most medical practices in the city. Eligible practices had 3 or more general internists and/or family practitioners participating in the managed care organization and at least 200 patients enrolled in the plan. There were 29 such practices, 13 with at least 4 physicians and 16 with 3 physicians. We selected all 13 practices with at least 4 physicians and randomly selected 6 of the practices with 3 physicians. All 19 practices agreed to participate. Two of the selected practices in Rochester had access to an ACS.
This analysis of patients with AF is part of a larger project assessing practice patterns for primary care physicians treating patients with AF, transient ischemic attack (TIA), and/or ischemic stroke (IS). An analysis of practice patterns for patients with TIA and/or IS but not AF will be the subject of a separate report.
We searched computerized billing records from August 1, 1992, through July 31, 1994, to identify patients with diagnostic codes suggesting AF, TIA, or IS. From this search, 2481 records were identified (1411 in the Research Triangle area and 1070 in Rochester). All 1070 records from Rochester were selected. Because of resource constraints, we could select only 1119 of the 1411 records from the Research Triangle. Given these constraints, we selected all possible records from the family practices and the smaller internal medicine practices, as well as a random sample of records from the larger internal medicine practices. Within the larger internal medicine practices, an effort was made to select equal numbers of records per physician.
We sought to identify patients aged 18 years and older with diagnoses of AF, TIA, or IS. For the purposes of this study, we considered the diagnosis of AF to be verified by results of a new electrocardiogram, chart documentation of a diagnosis of AF, or a diagnostic code of AF in the administrative files in conjunction with administration of warfarin. Atrial fibrillation could be chronic or intermittent, but patients with transient AF secondary to a temporary condition or medical procedure were excluded.
The time for which we searched billing records constituted the basic study period, and this is the period for which the inclusion criteria were applied. We defined the index visit as the first visit that occurred during the study period at which AF, TIA, or IS was mentioned. We abstracted forward for 2 years (even if before August 1992) from the index visit to determine various items of the medical history (eg, comorbidities, contraindications to warfarin therapy). Follow-up of anticoagulation management extended for 9 months beyond the index visit (even if after July 1994).
A data collection instrument was developed, tested using actual medical records, and translated into a computerized form for direct abstraction using laptop computers. The instrument included sociodemographic information, medical history, and a synopsis of office visits, laboratory reports, and telephone contacts, as well as information regarding hospital stays (eg, discharge summaries) and referral reports from specialists.
The 12 abstractors in the Rochester area were all experienced nurses, employed by the insurer and involved in its quality assurance program. In the Research Triangle area, abstraction was performed by a group of 8 experienced nurses and certified medical records technicians under direction of the Research Triangle Institute, Research Triangle Park, NC.
Abstractor training consisted of a 3-day intensive session and a subsequent 2-day follow-up session. Training concluded with independent review of medical records from 6 patients. To encourage consistency, regular telephone conference calls were held to discuss abstractor questions. Supervisors also visited practices to review progress. Approximately 5 records per practice were reabstracted (ie, by a different abstractor) as a qualitative check of interobserver agreement.
Patients were classified according to risk status and contraindications for anticoagulation therapy with warfarin. These contraindications were consistent with the package insert of warfarin, including impaired mental state, high risk for falls, active bleeding disorder, hematologic disorder, history of intracranial bleeding, hemorrhagic stroke before the index visit, thrombocytopenia, anemia, cerebral aneurysm, pregnancy, allergy or intolerance to warfarin, unstable anticoagulant control, and previously documented patient refusal of warfarin. Because relatively few contraindications were recorded within the medical record, we did not separate contraindications into those considered absolute vs those considered relative.
Stroke risk status was categorized as low, medium, or high. Low-risk patients had none of the following risk factors: aged 70 years or older or history of stroke, TIA, hypertension, diabetes, or heart disease (heart failure or coronary artery disease). Medium-risk patients had 1 of the risk factors, whereas high-risk patients had 2 or more.2
Process and outcome variables, all measured for at least 6 months after the index visit, included any warfarin use, use of the INR to report prothrombin time (PT) test results, time in target INR range, and response to out-of-range values. Warfarin use during the follow-up period was defined on a per-patient basis (ie, warfarin present or absent) and aggregated as the percentage of patients receiving therapy. The proportion of PT tests reported as INRs was defined on a per-test basis (ie, number of nonmissing PT tests reported as an INR divided by the number of nonmissing PT tests).
The assessment of time in target INR range was limited to PT tests reported using the INRs. Because physicians tend to test more frequently in response to an out-of-range INR value, we report time in range not only per test, but also per patient-day. For this latter purpose, INR values for each day of follow-up were estimated using linear interpolation between existing values.10 When assessing time in the target range, we excluded follow-up during the first 3 months after initiation of therapy, since INR values are expected to be less stable immediately after starting therapy. We allowed therapy to be discontinued and then reinstituted (eg, for a minor surgical procedure); for this purpose, we assumed that therapy was discontinued if warfarin was not given for 30 consecutive days.
The primary process measure was the response to out-of-range values, with a high out-of-range value defined as an INR of at least 3.50 and a low out-of-range value defined as an INR below 1.50. (These cutoffs were, in part, chosen to have a sufficiently large number of out-of-range values to analyze.) It is recommended that when faced with an out-of-range INR value, the physician should determine the probable cause, typically (but not always) modify the warfarin dose, and order a repeated PT test within the near future.3 Because we were unable to collect sufficient information to determine whether the warfarin dose should have been modified (eg, if a patient with historically stable INRs reported a temporary change in diet and/or alcohol intake, then the physician reasonably might assume that the INR will return to its target range even in the absence of a dose change), our assessment of process was limited to analyzing the time until the next PT test.
The analysis of time in target range and process of anticoagulation management included all INR values (ie, regardless of the patient's risk and contraindication status).
Data typically were reported using frequencies, percentages, means, and SDs. Calculation of P values was limited to comparisons between practice groups. For categorical variables (eg, PT test in target range or not), these P values were calculated using χ2 tests, the denominator being the PT test or the patient-day, as appropriate. For time until next PT test (ie, a continuous variable), these P values were calculated using an analysis of variance model including site (ie, Research Triangle, Rochester, with access to ACS, and Rochester without access to ACS) and INR category.
Of 2189 potential subjects, 1030 had an administrative code indicating probable AF. Of their records, 874 were located and abstracted (the remainder typically pertained to archived records of patients who were deceased, records of patients transferred to other practices, or records that were inexplicably missing). Of these 874 abstracted records, 612 had the diagnosis of AF confirmed by medical record review. An additional 48 patients with AF were located in the TIA-IS samples, bringing the total number of AF records reviewed to 660.
Table 1 presents demographic and medical history data for patients with AF. The mean age was 68.8 years; 52.1% were male; and 92.6% were white. More than 80% had at least 1 additional risk factor for stroke (eg, aged ≥70 years or history of stroke, TIA, hypertension, diabetes, or heart disease). Approximately 18% had a contraindication to warfarin therapy that could be discerned from chart review.
Table 2 presents warfarin prescription by clinical status. Overall, 34.7% of patients with AF received warfarin at some point during the observation period. The likelihood of being prescribed warfarin was only weakly related to stroke risk (ie, 26.7% of patients in the lowest-risk category received warfarin, compared with 31.6% of patients in the middle-risk category and 39.3% of patients in the highest-risk category) and not at all related to the documentation of contraindications in the medical record (ie, 33.8% when such contraindications were present and 35.0% when such contraindications were absent). These relationships were consistent across practice groups. Warfarin use was more common in the Rochester practices with access to an ACS compared with all other practices (44.3% vs 33.0%; P=.03).
For the 229 patients with AF receiving warfarin (ie, regardless of contraindication status), results of 1727 PT tests were recorded. Most test results (83.4%) were reported as INRs. This percentage varied by site (P<.001), with 99.8% of values from the Rochester practices with access to an ACS being reported as INRs, compared with 95.5% for the other Rochester practices and 73.4% from the Research Triangle practices. Subsequent analyses are limited to the 1441 values (from 174 patients) reported as INRs.
Figure 1 and Figure 2 summarize the results for these 1441 INR values. Treating the unit of analysis as the PT test (Figure 1), only 43.7% of measurements were within target range, this figure being higher (54.9%) in the Rochester practices with access to an ACS and lower (43.6% and 34.2%) elsewhere (P<.001). For patients in Rochester practices with access to an ACS, departures from target range were more or less equally distributed between low and high values, and a similar pattern of results was also observed for the Rochester practices without access to an ACS. In contrast, INRs for patients in the Research Triangle practices tended to be below the target range, with 51.0% of INRs below 2.00. Despite the tendency to underdose warfarin in the Research Triangle practices, these patients also had a higher percentage of tests with an INR of at least 5.00 (ie, 4.1% vs 1.7%; P=.07).
Figure 2 presents similar data using patient-day as the unit of analysis. The INR was within target range 60.3% of the time for the Rochester practices with access to an ACS, in comparison with 46.9% for Rochester practices without access to an ACS and 35.6% for the Research Triangle practices (P<.001).
Figure 3 presents time until next PT test after results that were within target range (ie, 2.00-3.00), high out of range (ie, ≥3.50), and low out of range (ie, <1.50). Within their respective sites, physicians scheduled follow-up visits sooner after an out-of-range test than after a test in the target range (P<.001). This phenomenon was particularly apparent for the Rochester practices with access to an ACS. For high out-of-range values, 51.0% of subsequent visits occurred within 7 days, compared with 34.1% of other Rochester practices and 32.1% of Research Triangle practices; for low out-of-range values, these figures were 41.7%, 29.3%, and 29.8%, respectively (P<.001).
Despite strong evidence that warfarin is effective in reducing stroke in patients with AF, we found that a substantial proportion of patients with AF who are eligible for treatment (ie, without documented contraindications) are not receiving warfarin, which is consistent with most other reports.11,12 We also found that INR values were out of the target range approximately half the time, and that the response to out-of-range values was not always timely. For all the measures considered here (ie, frequency of use, reporting by INRs, time in target range, and response to out-of-range values), the Rochester practices with access to an ACS had higher (albeit not ideal) quality of warfarin management compared with the remaining practices. Patterns of care in Rochester practices without access to an ACS tended, on balance, to be more closely aligned with the Research Triangle practices than with the Rochester practices having access to an ACS.
Our study has a number of limitations. Not all patients with AF can be identified successfully from diagnoses that are recorded on administrative files, and we cannot speculate how the practice patterns for such patients would differ (if at all) from those observed herein. Also, information in the medical record can be incomplete, particularly as applied to the recording of contraindications to warfarin therapy, and also to the reasons for why dose changes might or might not have been made.
Another limitation involved the choice of process measure. Ideally, for patients with out-of-range values, we would have preferred to study not only the time until the next visit, but also whether a dose change was indicated and made. We believe it to be virtually impossible to make a definitive analysis of process of anticoagulation management based on retrospective medical record review. Instead, such an assessment must be made prospectively or, if made retrospectively, be limited to sites using computerized anticoagulation management–tracking software, since such software prompts the user to record the elements required to assess process of care.
Apart from any considerations about how best to operationalize process measures, we note that any analysis that is limited to processes of care rather than clinical outcomes, such as thromboembolism and/or major bleeding, implicitly assumes that process can be linked with outcome. With only 30 to 40 patient-years of follow-up for each study group, it is unrealistic to anticipate that this link could be demonstrated in a data set of this size. The relationship between anticoagulation intensity and clinical outcomes has been well established elsewhere.8 Even without direct observation, the implications of differences in proportion of time within target range might be explicated using modeling. For example, if patients treated by an ACS are within their target range approximately 20% more of the time (in absolute magnitude) than those who are not, and if we assume for purposes of illustration that the annual clinical event rates during the time patients are out of range are increased by 2% (in absolute magnitude) compared with annual event rates for the time patients are within target range, then we would expect that patients treated by an ACS would have, on average, 0.4% (ie, 0.02 × 0.20) fewer major clinical events per year. These events could then be assigned an economic cost, allowing a formal cost-effectiveness analysis to be performed.
As another limitation, although we observed that the quality of anticoagulation management tended to be better in the practices with access to an ACS than in the other practices, it is important to recognize that our study was not designed to demonstrate a causal link between the two. In fact, such a study would require more sites than used herein as well as randomized allocation.8 The Managing Anticoagulation Services Trial that is currently underway is addressing these issues.13,14
A final limitation is that these data describe practice patterns in 1992-1994, and practice patterns may be changing over time. Prescription of warfarin for patients with AF appears to be rising gradually,11 presumably in response to various large trials demonstrating its efficacy. Much less is known about the quality of anticoagulation management among patients receiving warfarin, and we know of no reason (ie, analogous to the effect of the large warfarin trials on increasing the prescription of warfarin) that would suggest that the quality of anticoagulation management is likely to be improving in typical practice. On the contrary, many of the barriers to improved warfarin management are systemic (eg, lack of effective tracking systems, inadequate coordination with laboratories) and thus require system-level solutions.13 The ACSs are such a system-level solution, and these results contribute to the growing literature supporting this form of organization.5
Because of resource constraints, quality improvement initiatives often concentrate on measures that can be obtained from administrative databases and/or cursory record reviews. In anticoagulation management, this typically involves linking files containing diagnoses (eg, AF and various contraindications to warfarin) with pharmacy files containing the use of warfarin, to determine the proportion of probably eligible patients with AF who are receiving warfarin. The implicit assumption behind this approach is that a practice that performs well on a single quality indicator should also perform well on other, unmeasured indicators. Although such a trend was noted herein (ie, with access to an ACS evidencing the highest quality care for all indicators), this need not be true in general, especially if the quality indicators address very different aspects of care.
We found, consistent with intuition and the experiences of others, that retrospective review of the medical records of ambulatory patients was a difficult and time-consuming task, and one that is ultimately constrained by the quality of the documentation in those records. This is one of the most persuasive arguments for an integrated, computerized medical record. In particular, the computerized tracking software used by the ACS contained all the information necessary to assess time in target range and response to out-of-range values, thus mostly or entirely eliminating the necessity for a traditional chart-based review. Such an approach is consistent with the principles of continuous quality improvement, which suggest that measurements of a process should be taken as soon as possible, thus facilitating feedback of information to responsible parties before serious quality problems develop. The application of such software makes self-evident sense in an ACS, which is entirely focused on the single task of anticoagulation management. A much greater challenge is how to implement this same basic idea in the offices of internists and other nonspecialist providers who regularly encounter patients with highly varied medical conditions. In any event, what is in question is not the general principle, namely, that quality-related information is best captured prospectively as part of the usual processes of care, but only its best implementation.
We found significant deficiencies in the practice of warfarin management and suggestive evidence that ACSs can partially ameliorate these deficiencies. Considerably more research is needed to describe the quality of anticoagulation management in typical practice and how this management can be improved.
Accepted for publication August 3, 1999.
This research was supported by a grant (contract 282-81-0028, Stroke Prevention Patient Outcomes Research Team) from the Agency for Healthcare Research and Quality, Rockville, Md.
Reprints: Gregory P. Samsa, PhD, Center for Clinical Health Policy Research, Suite 230, First Union Bldg, 2200 W Main St, Durham, NC 27705 (e-mail: firstname.lastname@example.org).