Tracking the Uptake of Evidence:  Two Decades of Hospital Practice Trends for Diagnosing Deep Vein Thrombosis and Pulmonary Embolism FREE

ADVANCES IN clinical trial methods have led to a progression of clinical trials evaluating diagnostic¹ and treatment modalities^2- 3 that have profoundly affected the clinical management of patients with venous thromboembolism. Large volumes of evidence have been generated during the past 2 decades by researchers in the clinical field of venous thromboembolism. In particular, rigorous prospective studies (level 1)^1,4- 7 have confirmed or refuted dogmatic approaches to the diagnosis of venous thromboembolism by evaluating a broad spectrum of randomized consecutive patients, using a definitive reference test, and interpreting the results of each test or outcome without knowledge of the other test findings. The result has been a paradigm shift in the literature toward more rigorous approaches to the diagnosis of deep vein thrombosis (DVT) and pulmonary embolism (PE). Outcome studies, with long-term follow-up to determine the safety of withholding treatment in patients with negative results, have allowed refinement of the diagnostic pathway.^1,8

Although we have a general sense that diagnostic approaches have changed during the past 2 decades, the evolution of diagnostic testing for DVT and PE has not been clearly documented. Furthermore, the association of key new evidence in the peer-reviewed literature with changes in the use of diagnostic tests for PE and DVT has not been documented. This question is of considerable relevance because previous research in other areas has demonstrated marked delays between the appearance of definitive new evidence and its application by practicing clinicians.⁹ Definitive level 1 studies of DVT have demonstrated the importance of objective diagnostic tests.¹ The first test to be evaluated was ascending contrast venography.^10- 11 Testing progressed to noninvasive diagnosis with impedance plethysmography and subsequently Doppler ultrasonography, including B-mode imaging.^12- 17 The literature on the diagnosis of PE evolved in a more complex manner.^18- 55 Initial enthusiasm^{29,43- 44,48- 49} for the merits of ventilation-perfusion (V˙/Q˙) lung scans was subsequently dampened by level 1 studies showing that (V˙/Q˙) lung scanning was often nondiagnostic.^47,50- 51 The applicability of noninvasive leg testing for DVT as a surrogate for the diagnosis of PE was subsequently shown.^56- 58

The National Hospital Discharge Survey (NHDS)⁵⁹ has provided valuable information on trends in hospital care across the United States for a variety of medical conditions. This data resource was used to derive a historical profile of the use of diagnostic tests for DVT and PE with a specific focus on trends over 21 years in (1) case volumes for both of these conditions, and (2) specific diagnostic tests performed. The timing of level 1 studies of specific diagnostic tests for DVT and PE was related to these trends. A distinction was made between the evolving literature and the eventual definitive level 1 evidence for specific diagnostic approaches.¹ These rigorous level 1 studies^{11- 17,47,50- 52,56- 58,60- 72} provided the key to understanding the role of each diagnostic test or diagnostic pathway. These studies were meticulously designed to avoid bias.

Data from the NHDS showing 21-year trends from 1979-1999 for the diagnosis of patients with venous thromboembolism were analyzed. The purpose of this article is to show how advances in the literature during the past 2 decades correlated with clinical practice.

Data from the NHDS⁵⁹ were used for this study. The NHDS database includes a broad mix of teaching and community hospitals. Data from this study are available on CD-ROM,⁷³ and much of it is available in hard copy.^74- 91 The NHDS is based on data abstracted from a national probability sample of discharges from short-stay, noninstitutional hospitals, exclusive of federal, military, and Veterans Administration, in the 50 states and the District of Columbia. Only hospitals with an average length of stay for all patients of less than 30 days or those whose specialty is general (medical or surgical) or children's general, regardless of length stay, are included in the survey. Hospitals with an average length of stay of 30 days or more were excluded from the NHDS before 1988. Hospitals must have 6 or more staffed beds to be considered for the survey.

The number of responding hospitals and sampled patient abstracts in the NHDS for 1979 through 1999 ranged from 400 to 480⁷³ and from 181 000 to 307 000, respectively, representing approximately 8% of all hospitals and 1% of all discharges. The annual hospital response rate for the NHDS generally exceeds 90%. The survey includes all hospital discharges, including newborn infants and patients who died in the hospital.

The survey design, sampling, and estimation procedures were planned to produce calendar-year estimates. Trained medical personnel coded diagnoses and procedures using the International Classification of Diseases, Ninth Edition, Clinical Modification (ICD-9-CM). A minimum of 1 and a maximum of 7 diagnostic codes were assigned for each sample abstract. If an abstract included surgical or diagnostic procedures, a maximum of 4 procedure codes was assigned.

The NHDS is based on a national probability sample of discharges from noninstitutional hospitals exclusive of federal, military, and Department of Veterans Affairs hospitals, located in the 50 United States and the District of Columbia. A 3-stage sampling plan was introduced in 1988 and replaced an earlier, similarly designed 2-stage sampling plan.⁵⁹ For both designs, there is a probability of sample hospitals and a systematic sampling procedure to select discharges within hospitals. The changes introduced with the 1988 redesign do not compromise the ability to conduct trend analysis.⁵⁹

There were 112 primary sampling units composed of counties, groups of counties, county equivalents (such as parishes and independent cities), or towns and townships.

Hospitals in the primary sampling unit that had 1000 or more beds were always selected for inclusion in the survey and were termed "certainty hospitals."⁷³ All other hospitals were selected using systematic random sampling. These were selected from the primary sampling units with a probability proportional to their annual number of discharges.

A sample of discharges from each hospital was selected using a systematic random sampling technique.

Estimates of total numbers of patients with DVT, patients with PE, and diagnostic tests performed in the entire United States were derived from the number of sampled patients with DVT, the number of sampled patients with PE, and the number of diagnostic tests performed in sampled patients. This estimating was done using a multistage estimation procedure that produces essentially unbiased national estimates and has 3 basic components: inflation by reciprocals of the probabilities of sample selection, adjustment for nonresponding hospitals and for missing discharges within hospitals, and population weighting ratio adjustments.⁷³ As the statistics from the survey are based on a sample, they may be different from the figures that would have been obtained if a complete census had been taken.

All available diagnostic code fields were screened for specific codes to identify patients with DVT or PE. Since 1979, the ICD-9-CM has been used for classifying diagnoses and procedures in the NHDS. Although the ICD-9-CM has been modified annually, the diagnostic codes for PE and infarction and phlebitis and thrombophlebitis of deep vessels of the lower extremities have changed little. The specific ICD-9-CM codes used for identification of patients with PE are 415.1, 634.6, 635.6, 636.6, 637.6, 638.6, and 673.2. The ICD-9-CM codes used for identification of patients with DVT are 451.1, 451.2, 451.8, 451.9, 453.2, 453.8, 453.9, 671.3, 671.4, and 671.9. Five-digit codes such as 415.11 (included under the code 415.1) were not listed separately, as they were included under the corresponding 4-digit codes.

Patients who underwent diagnostic procedures for DVT or PE were identified by screening the following procedure codes: 88.66—phlebography of femoral and other lower extremity veins using contrast material, 88.43—arteriography of pulmonary arteries, 92.15—pulmonary (radioisotopic) scan, and 88.77—DVT ultrasonic scanning.

Descriptive statistics were used to graphically display trends over time in case volumes and number of diagnostic tests performed. These graphs showed estimated numbers of cases or diagnostic tests performed in the United States rather than numbers of sampled cases or procedures. It is recognized, however, that hospital discharge data incompletely capture many diagnoses and procedures. Therefore, trends in the relative number of patients with DVT or PE and trends in the relative use of specific diagnostic tests are the focus. Methodological issues, including the phenomenon of diagnosis related group (DRG) "creep,"⁹² change in the ICD-9-CM system, and the issue of sensitivity and specificity⁹³ for capturing specific diagnoses and procedures, are discussed.

Linear regression analyses were used to calculate the slopes of selected linear segments of the curves describing the data.⁹⁴ Pearson correlation analyses were performed for the same linear segments to assess the extent of dispersion of points around the regression lines.⁹⁴ More complex equations to describe the curves were explored, but the fit of these curves never surpassed that of the linear analyses of selected segments. Differences in the number of diagnostic tests performed over time were assessed using t tests and analysis of variance.⁹⁴

The range of publication years of level 1 studies^{4- 7,11- 17,47,50- 52,56- 58,60- 72} that describe the validity of diagnostic tests and confirm their clinical outcomes was illustrated as timelines on the graphs describing the trends in the use of these tests.

MEDLINE and EMBASE literature searches were performed. The American Thoracic Society guidelines,¹ which categorized diagnostic studies into level 1 or not, were also used as an external reference source. Rigorous prospective level 1 studies are key to understanding the role of each diagnostic test, as these prospective studies are meticulously designed to avoid bias.

The 21-year trends in the number of patients diagnosed as having DVT or PE are shown in Figure 1. For DVT, the trend was one of relative stability between 1979 and 1989, followed by an increase between 1989 and 1999. For this latter period, a significant upward linear trend was shown (slope, +15 817 patients per year; r = 0.978; P<.001). The trends for PE were somewhat different, with an initial decrease in cases between 1979 and 1989, followed by an increase between 1989 and 1999. Again, a significant linear trend for this latter period was shown (slope, +3446 patients per year; r = 0.804; P<.003), although the increasing slope was less than that for DVT.

The 21-year trends in the use of ascending contrast venography and Doppler ultrasonography for the diagnosis of DVT are shown in Figure 2. The use of venography increased linearly between 1979 and 1988 (slope, +8601 procedures per year; r = 0.981; P<.001) and then declined linearly and equally sharply until 1996 (slope, −9900 procedures per year; r = −0.986; P<.001). After 1996, the number of recorded venograms remained low and relatively constant.

The number of venous ultrasound examinations increased linearly between 1979 and 1991 (slope, +8752 procedures per year; r = 0.981; P<.001). Between 1991 and 1997, the number of ultrasound examinations remained relatively constant at a high level. Reflecting these divergent trends, there was a strong negative correlation between the frequency of venous ultrasound examinations and the frequency of venograms between 1988 and 1999 (r = −0.768; P = .004).

The corresponding chronology of evidence for the diagnosis of DVT is also shown in Figure 2. Initial level 1 studies that used venography as a reference standard contributed to the open timeline in Figure 2. These studies, published between 1976 and 1982, showed the need for an objective diagnosis and the potential use of noninvasive leg testing.^11,60- 64 Noninvasive leg testing was not initially widely available. Indeed, impedance plethysmography remained largely a research test.^60- 64 These observations correlate with the rapid ascent of venography as a diagnostic test for DVT. The second key contribution in the diagnosis of DVT (depicted by the shaded timeline in Figure 2) was the appearance of level 1 studies documenting the validity of noninvasive testing^{12- 17,65- 72} against the reference standard of venography. These latter studies correlate in time with the rapid ascent of Doppler ultrasonography and the corresponding decline in the use of venography use.

The 21-year trends in the use of diagnostic tests for PE, namely, V˙/Q˙ lung scanning, pulmonary angiography, and Doppler ultrasonography, are shown in Figure 3. The use of V˙/Q˙ lung scanning between 1979 and 1982 was relatively constant. Between 1982 and 1986, there was a sharp increase in the use of V˙/Q˙ lung scans (slope, +24 430 procedures per year; r = 0.986; P = .002). A sharp linear decrease occurred between 1986 and 1999 (slope, −7893 procedures per year; r = −0.986; P<.001). This coincided temporally with a rapid rise in use of Doppler ultrasound.

The use of pulmonary angiography increased gradually and linearly for the entire 21 years (slope, +570 procedures per year; r = 0.876; P<.001). In the latter years studied (1986-1999), there was a negative correlation between the frequency of V˙/Q˙ scanning and pulmonary angiography (r = −0.567; P = .04).

The corresponding chronology of evidence for the diagnosis of PE is depicted by the open and shaded timelines in Figure 3. Initial level 1 studies that contributed to the open timeline used pulmonary angiography as a definitive reference standard test. These studies confirmed the need for objective testing of the lung for PE and the lower extremity for DVT.^47,50 These level 1 studies showed that V˙/Q˙ lung scanning alone did not give a definitive diagnosis in many patients,^47,50- 51 that diagnostic leg testing was a valuable adjunct,^47,50 and that pulmonary angiography was necessary in selected patients.^47,50- 51

Rigorous level 1 clinical trials assessing clinical outcomes using long-term follow-up confirmed the clinical value of serial noninvasive leg testing in patients with DVT^65- 72 (Figure 2) or PE^56- 58 (Figure 3). The findings of these studies correlate with the continued increased and ultimately stable use of Doppler ultrasonography.

Our 21-year analysis indicates that the sequential introduction of improved diagnostic approaches based on rigorous clinical research is reflected in profound and appropriate changes in hospital-based clinical practice for patients with suspected venous thromboembolism in the United States. Although marked delays have been observed in the uptake of evidence in other clinical areas,⁹ and the phenomenon of "clinical inertia"⁹⁵ is well recognized, our findings point to a clear response by clinicians to the appearance of definitive evidence.

Changes in the frequency of use of venography, Doppler ultrasonography, and V˙/Q˙ lung scanning are not attributable to changes in the frequency of diagnosis of PE and DVT, as trend curves for procedure use (Figure 2 and Figure 3) do not correlate with trend curves depicting disease frequency (Figure 1).

Contrast-enhanced spiral computed tomography emerged as a diagnostic test for PE in the 1990s.^96- 98 Because this test became widely available only in the late 1990s, it is unlikely that its use has been indirectly captured by our trend analysis, although an effect in the late 1990s cannot be excluded.

Some methodological issues require consideration. Important issues include sensitivity and specificity⁹³ of the NHDS summary sheet for capturing diagnostic test use, the phenomenon of DRG creep,⁹² and changes over time in the ICD-9-CM coding system. The specificity of ICD-9-CM coding is high. Thus, most procedures that were coded in discharge abstracts actually occurred. The frequencies reported, however, are underreported because of the imperfect sensitivity of coding for capturing diagnostic procedure occurrence.⁹³ The consequence of lower sensitivity is that the absolute values shown on the trend curves in Figure 1, Figure 2, and Figure 3 are an underestimate of the actual values. In contrast, the directional trends and relative positions of the curves described herein are likely to be correct. The potential for variation in sensitivity of coding over time represents a possible threat to the validity of our findings, and, in particular, 2 phenomena may have introduced variability into the sensitivity of coding: DRG creep and changes over time to the ICD-9-CM coding system. In the early 1980s, the Health Care Financing Administration introduced DRGs as a mechanism for reimbursing hospitals providing care to Medicare recipients. Hospital administrations rapidly recognized that reimbursement was directly linked to the extent of coding for individual patients. The phenomenon of DRG creep is thus an artifact of coding that might have increased the sensitivity of coding when this method of reimbursement was introduced. Despite the potential for DRG creep in our findings, we see that increases in the use of certain diagnostic tests in the mid-1980s coincided with the appearance of published evidence that likely explains these increases. Furthermore, the use of some diagnostic tests declined in the middle to late 1980s. These observations, therefore, suggest that DRG creep is unlikely to be a major contributor to the findings in Figure 2 and Figure 3. As for the issue of changes over time to the ICD-9-CM coding system, there were no confounding changes to the coding for PE, DVT, or associated diagnostic tests during the 21 years studied. Thus, it is unlikely that changes to the ICD-9-CM perturbed our findings. A minor issue is that a minimal number of diagnostic tests were obtained for atypical reasons, including V˙/Q˙ lung scanning before thoracic surgery and pulmonary angiography for chronic pulmonary hypertension.

Other factors contributing to the overall trends or their relationships are the availability of new diagnostic technology, peer example, and the general literature. Although many of the changes in the diagnostic approach over time are generally useful, there remains considerable room for improving the diagnostic approach to DVT and PE.

In summary, there were dramatic increases and reductions in the use of diagnostic tests between 1979 and 1999 that reflect the evolving paradigms of care for venous thromboembolism. The changes observed correlate over time with the results of rigorous level 1 diagnostic studies. Generally speaking, there has been considerable skepticism⁹⁵ about the impact of evidence on clinical practice. Our 21-year perspective on diagnostic approaches to venous thromboembolism indicates that the apparently disappointing impact of various approaches to continuing medical education may reflect an inadequate interval of follow-up rather than failed application of the published evidence. Our observations, based on a US-wide hospital discharge survey, indicate that changes in clinical practice occur over many years and long-term follow-up is required to capture these changes.

Corresponding author and reprints: Paul D. Stein, MD, St Joseph Mercy Oakland Hospital, 44555 Woodward Ave, Suite 107, Pontiac, MI 48341-2985 (e-mail: steinp@trinity-health.org).

Accepted for publication August 16, 2002.

We thank Robert Pokras, MA, and Jennifer R. Popovic, MA, Centers for Disease Control and Prevention, National Center for Health Statistics, Hyattsville, Md, for their assistance in the preparation of this manuscript and Vinay Bharadia, BSc, MSc, Rita Biel, BSc, Natasha Burke, BSc, Jill Ho-You, BA, and Jeanne Sheldon, BA, for their assistance.