 Critical pathways are healthcare management plans that specify patient goals and the sequence and timing of actions necessary to achieve these goals with optimal efficiency. More than 80% of hospitals in the United States use critical pathways for at least some of their patients. Unfortunately, critical pathway effectiveness in improving clinical efficiency is unclear.

 To assess whether critical pathways have been successful in reducing patient length of stay (LOS) and resource utilization in our tertiary-care academic medical center.

 A before-and-after observational study using multivariate linear regression analyses.

 We identified all critical pathways initiated in our medical center between 1993 and 1996 in which at least 50 adult patients would be evaluated in the year preceding and succeedingpathway implementation; 13 pathways satisfied these inclusion criteria. Using a before-and-after design, multivariate linear regression was used to evaluate each pathwayâ€™s effect on average monthly LOS and resource utilization after adjusting for case mix and secular trends.

 Three of the 13 pathways were associated with a statistically significant immediate decrease in inpatient LOS: acute myocardial infarction (20.7% decrease; P = .001), cesarean section (14.6% decrease; P = .03), and kidney transplantation (24.5% decrease; P = .003). Only 1 pathway, percutaneous transluminal coronary angioplasty (PTCA), produced a statistically significant decrease in LOS slope (a decrease of 5.2% per month; P = .001). Two pathways were accompanied by a statistically significant immediate reduction in ancillary resource utilization: kidney transplantation (26.4% decrease; P = .001) and community-acquired pneumonia (21.8% decrease; P = .002). Only the PTCA pathway produced a statistically significant decrease in resource utilization slope during the 12- month follow-up period (a decrease of 8.4% per month; P < .001).

 Although some pathways did reduce LOS or resource utilization or both, most pathways reduced neither. Because substantial resources must be expended on pathway development, implementation, and maintenance, future efforts should be placed on further evaluating the effectiveness of critical pathways and understanding the reasons behind their success or failure before additional resources are consumed for this management strategy.

More than 80% of hospitals in the United States use critical pathways as care strategies for at least some patients.

 Critical pathways are management plans that specify goals for patients and the corresponding ideal sequence and timing of staff actions to achieve those goals with optimal efficiency.

 Implementation of critical pathways, developed for myriad diseases and admitting diagnoses, usually requires a substantial time commitment by a multidisciplinary development team.

 Such endeavors are largely regarded as worthwhile by healthcare managers who have "embraced critical pathways as a method to reduce variation in care, decrease resource utilization, and potentially improve healthcare quality."

The major goal of a critical pathway is to maximize clinical efficiency of care by reducing patient length of stay (LOS) and resource utilization. By closely coordinating care by both nursing and medical staff from the time a patient is admitted until hospital discharge, pathways are theoretically well suited to minimize waste and inefficient care. Unfortunately, evaluation of the effectiveness of critical pathways has been inadequate.

 The University of Michigan Medical Center, a tertiary-care academic medical center, has been using critical pathways for more than a decade. To evaluate the effectiveness of critical pathways at that center, we assessed whether these clinical management plans have been successful in reducing patient LOS and resource utilization. 

The University of Michigan Medical Center, an 872-bed hospital with a level-1 trauma center, is a primary referral hospital in Southeastern Michigan. Critical pathways have been developed and widely implemented for various disorders and admitting diagnoses at the University of Michigan Medical Center since 1991. 

Pathway development at the University of Michigan Medical Center entailed a multidisciplinary process, usually led by either a physician or nurse with special expertise or interest in the disease process focused on by the pathway. For each pathway, 4 steps usually occurred. (1) A series of meetings were held, often during the course of several months, in which evidence from both published literature and expert opinion were used to develop a final version of the pathway for use in patients with the diagnosis of interest. (2) The pathway was then disseminated to the nursing and medical staff providing clinical care to patients with the diagnosis under consideration. (3) In addition to educating the nursing and medical staff about the specifics of the pathway, copies of the pathway were available in each clinical area for future reference. (4) A person was appointed as pathway "leader" so that modifications could be made in a coordinated manner as needed in response to new information. For a few pathways, the clinical staff occasionally used the pathway as an orientation tool for new house officers or nurses. In their final versions pathways were generally formatted as Gantt charts, thereby outlining the suggested processes of care using a time-task matrix.

 Of note, no type of incentive-financial or otherwise-was offered to encourage pathway use. 

We identified all critical pathways initiated in our medical center between 1993 and 1996 in which at least 50 adult patients were evaluated in the year preceding and succeeding pathway implementation. Thirteen pathways satisfied inclusion criteria. The diagnoses (or procedures) covered by these pathways were the following: acute myocardial infarction, acute pancreatitis, asthma, breast surgery, cesarean section (C-section), community-acquired pneumonia (CAP), hip arthroplasty, kidney transplantation, knee arthroplasty, liver transplantation, lung lobectomy, percutaneous transluminal coronary angioplasty (PTCA), and radical nephrectomy. We used ICD-9 procedure or diagnosis codes to identify patients who would be considered for management by each pathway. As we were primarily interested in the effectiveness of pathways as a management method, we included patients whether or not they were actually managed on the particular pathway. (If this were a clinical intervention, our approach could be termed an intention-to-treat analysis.) 

We excluded patients transferred into our hospital for continued inpatient care, as most such patients likely would not have been managed by a pathway. We also excluded LOS outliers, defined as those patients with a LOS more than 3 standard deviations above the geometric mean for their corresponding diagnosis-related group (DRG). We excluded outliers for 2 reasons. First, we believe that such patients are fundamentally different from the population we wished to generalize about and are not part of the target population for the application of critical pathways. Specifically, outlier patients tend to have very severe or complex illness (or occasionally social issues) requiring a completely different management approach than that used in most routine cases. The pathways concept itself is not suited for addressing extreme situations; indeed, the strategy is based on clinical experience with routine cases. Second, even if we thought extreme outliers in this study were part of the same population as the other cases, there are statistical reasons to assess them differently. In situations in which extreme skew exists, a variety of techniques are commonly used. We believe that outright deletion did not provide a substantially different result from the down-weighting that would have occurred with the other statistical techniques that we would have had to use if we had left the DRG outliers in the sample (e.g., median or quartile regression). 

Using a before-and-after design, we evaluated each pathway's effect on patient LOS and resource utilization (as determined by University of Michigan-derived standardized relative value units

). The unit of observation was the average monthly LOS or resource utilization, adjusted for case mix. We constructed separate linear regression models for each pathway using each outcome (LOS and resource utilization) and included patient data 1 year before and 1 year after pathway implementation. 

To account for secular trends, 5% of patients with other diagnoses but in the same general discipline as pathway patients (e.g., surgery, cardiology) were randomly selected to serve as controls. In this manner, we were able to assess both secular trends occurring specifically among pathway patients as well as overall ecological effects occurring in the hospital. We used regression models to evaluate whether any change in LOS or resource utilization could be detected after pathway implementation. Specifically, we looked for both a 1-time immediate change and a change in the slope of utilization over time occurring after the pathway was implemented. 

Any changes observed were compared with changes in the non-pathway diagnoses as a control group, and only those changes above and beyond those seen in non-pathway diagnoses were considered as potentially attributable to the pathway. Thus our regression models included variables identifying whether an observation was pathway or non-pathway (PATH), the month of the observation (MONTH), and whether the observation was from before or after pathway implementation (PERIOD). The interaction between pathway and period tested the hypothesis that a 1-time decrease (or increase) in utilization occurred after implementation. The 3-way interaction between PATH, MONTH, and PERIOD tested the hypothesis that there was a change in the slope of utilization over time after pathway implementation that was above and beyond any background ecologic changes occurring in non-pathway diagnoses. The full model included all possible 2-way interactions and the 3-way interaction. However, if the 3-way interaction was not significant it was dropped from the model. Thereafter, any non-significant 2-way interactions were dropped. The 2-way interaction testing the hypothesis of the 1-time change in utilization was retained in the models regardless of its statistical significance. 

To simplify interpretation of the results of the regression model, 

represent the final model developed using the above strategy for each diagnosis. We used a natural log transformation so that the distribution of the dependent variables would more closely approximate the normal distribution. Although dramatic changes in patient case mix are atypical within a 24-month period at our facility, we nevertheless adjusted for such changes using the DRG case-mix index. Our analyses therefore account for both secular trends and case-mix changes within groups. 

A pathway was considered successful if a statistically significant decrease, immediate or gradual, was noted in 1 of the outcome measures. Statistical significance for all analyses was based on a 2-sided 

< .05. Analyses were conducted with SAS version 6.08 (SAS Institute Inc, Cary, NC). RESULTS Overall, 5 of the 13 pathways (38%) were associated with a statistically significant decrease in LOS, resource utilization, or both, after controlling for secular trends and case-mix changes (

). Evaluating pathway effects stratified by outcome measure, however, provided additional important information. 

No statistically significant increase in LOS was noted after pathway implementation. Three of the 13 pathways were associated with a statistically significant immediate decrease in LOS: acute myocardial infarction (20.7% decrease; 

= .001; Figure 1a), C-section (14.6% decrease; 

= .03; Figure 1e), and kidney transplantation (24.5% decrease; 

= .003; Figure 1h). Only 1 pathway, PTCA (Figure 1l), produced a statistically significant decrease in LOS slope (5.2% decrease per month; 

= .001). Of note, the PTCA pathway was not associated with an immediate decrease in LOS. For most of the 13 pathways (69%), no significant decrease in LOS was noted either immediately or during the course of the 12-month follow-up period. 

Pathway Effects on Resource Utilization  

Two pathways were accompanied by a statistically significant immediate reduction in ancillary resource utilization: kidney transplantation (26.4% decrease; 

= .001; Figure 2h) and CAP (21.8% decrease; 

= .002; Figure 2f). Again, only the PTCA pathway (Figure 2l) produced a statistically significant decrease in resource utilization slope during the 12-month follow-up period (8.4% decrease per month; 

< .001). In the asthma pathway (Figure 2c), a statistically significant increase was noted in resource utilization immediately after pathway implementation (17.7% increase; 

= .04). Similar to the findings with LOS, most of the pathways (77%) did not significantly decrease resource utilization either immediately or during the 12-month follow-up period. 

Importantly, if we had not accounted for secular trends, the pathway for C-section would have appeared to reduce resource utilization by 45.2% (

< .001). However, after including ecological changes within our regression models, the effect of the pathway was no longer statistically significant (18% reduction in resource utilization; 

Figures 1 and 2 depict how each of the 13 pathways affects LOS and resource utilization, respectively. In general, if a pathway was associated with a significant decrease in LOS it was also associated with a significant decrease in resource utilization. As can be seen, however, most of the time pathway implementation was statistically unrelated to either LOS or resource utilization. Figures 1 and 2 do reveal a marked heterogeneity of pathway effectiveness across the different diagnoses and procedure. Describing a few of the pathway results in detail clarifies the varied effects of the pathways. 

For the LOS evaluation, the pathways for acute myocardial infarction (Figure 1a), C-section (Figure 1e), and kidney transplantation (Figure 1h), were associated with an immediate decrease in LOS following the intervention but no gradual effect (i.e., the intervention changed the intercept but not the slope). The PTCA pathway (Figure 1l) reveals no immediate effect of the intervention but a substantial gradual effect (i.e., the intervention changed the slope but not the intercept). This example highlights the need to allow for gradual effects in the evaluation process and for evaluating the effect of a pathway several months after implementation. The plots in Figure 2 evaluating resource utilization are roughly similar to the plots for the corresponding pathways in Figure 1. 

We evaluated the effectiveness of critical pathways to improve clinical efficiency in 13 clinical areas in our institution. Our study revealed the following important finding: although some pathways did reduce LOS, resource utilization, or both, most pathways reduced neither. Our findings add to the growing, but still limited, number of studies that have evaluated critical pathways while also considering the secular trends that have been reducing LOS and resource utilization independent of pathway use. Whereas several earlier uncontrolled studies7 that did not consider secular trends have found critical pathways to be effective in increasing efficiency, 2 recent studies have not found critical pathways to be effective once ecological trends and other quality improvement strategies are considered.

 Given the decrease in LOS observed during the past decade in US hospitals, it is important that investigators evaluating pathway effectiveness adjust for this downward trend. Indeed, when we estimated national ecological trends using the national Healthcare Cost and Utilization Project database ( http://www.hcup.ahrq.gov/HCUPNet.asp ), which was funded by the Agency for Healthcare Research and Quality, we found substantial similarity between our data and the secular trends found in the population as a whole. 

 performed a before-and-after evaluation of 5 surgical pathways implemented between 1993 and 1997 in their tertiary-care hospital. Although each pathway significantly decreased postoperative LOS (from 3% to 21% depending on the pathway), similar reductions in LOS were observed for 3 of the procedures at surrounding hospitals not using pathways.

 Based on their findings, Pearson and colleagues

 "dropped" plans to develop pathways for additional conditions. In their study of Medicare beneficiaries at 32 Connecticut hospitals, Holmboe and colleagues

 evaluated the effectiveness of a critical pathway for patients hospitalized with acute myocardial infarction. Compared with patients in hospitals without an acute myocardial infarction critical pathway but using other quality improvement strategies (e.g., physician opinion leaders and peer review), patients hospitalized at the 10 pathway hospitals had similar lengths of stay.

 Additionally, patients in hospitals using critical pathways did not have increased use of proven medical therapies (e.g., aspirin, beta-blockers, reperfusion) or an improved survival compared with patients hospitalized at medical centers not using an acute myocardial infarction pathway.

 However, a recent study by Marrie and colleagues

 from Canada focusing on a critical pathway for CAP found that pathway to be substantially more efficient than standard care. Specifically, this 19-center randomized trial found that hospitals randomized to the critical pathway had significantly reduced average LOS (8.2 days vs. 9.6 days) and duration of intravenous antibiotics (4.6 days vs. 6.3 days), without worsening quality of life or increasing the rate of adverse events.

What are the implications of our findings in the context of previous work performed in this area? Because substantial resources must be expended on pathway development and implementation,

 pathway effectiveness should be clearly demonstrated before additional resources are consumed for this management strategy. The confidence placed in critical pathways as a management tool to coordinate a large number of diagnoses and procedures to enhance clinical efficiency should perhaps be re-appraised. However, although the majority of pathways we evaluated were not effective, some clearly were. In fact, the PTCA pathway, in particular, was markedly successful in decreasing both LOS and resource utilization. Thus, increased efforts should be made to focus on those diagnoses or procedures that are most likely to benefit from the application of a critical pathway. 

Evaluating the reasons behind the success or failure of an individual pathway is complex. When considering the results of our evaluation, we identified some possible explanations for why some pathways were successful while most were not. Importantly, these explanations were not based on a priori hypotheses and thus serve primarily for hypothesis generation. Pathways for a few surgical procedures were successful. Specifically, the pathways for C-section and kidney transplantation reduced LOS or resource utilization significantly more than secular trends would have predicted at our hospital. We were unable to determine why these pathways were successful while pathways for the other surgical procedures were not. One hypothesis is that kidney transplantation and C-section are relatively routine procedures, thus lending themselves to closely coordinated care. Additionally, perhaps if we had larger sample sizes, some of the other surgical pathways would have led to statistically significant decreases in LOS (e.g., knee arthroplasty) or resource utilization (e.g., hip arthroplasty). Most of the remaining unsuccessful surgical pathways, however, had very little effect on LOS or resource utilization, indicating that lack of power would be an unlikely explanation for our findings. Yet even if some of these pathway effects were statistically significant, often the modest effect sizes would not justify their use. Pathways are, after all, practical management tools, and their purpose is to obtain clinically meaningful effects, not merely statistically significant ones. 

Pathways for cardiology-related domains also appeared to be successful. We have a potential explanation for these findings. During the past several years at our institution, much effort has been expended in improving the quality of care patients with cardiovascular disorders receive. Indeed, the cardiology leaders of these quality improvement efforts at our institution have published their successful efforts widely.

 Thus, having well-respected opinion leaders spearhead quality improvement initiatives in a focused clinical area likely accounts for the success of the cardiology-related pathways in our hospital. However, pathways intended for 2 medical diagnoses (asthma and pancreatitis) were not effective, perhaps due to their notably variable clinical courses and high degree of patient-specific management strategies. 

Given our results and those of others who have evaluated critical pathways, how should we approach the important issue of rational resource use in a hospital environment? We think that a "one-size-fits-all" management approach should be treated skeptically, regardless of its theoretical appeal. To choose the appropriate management technique, decision makers must understand the underlying processes and key points of care that are to be managed, and these appear to differ by medical procedure and even by hospital. For example, if LOS and ancillary resource consumption are the management targets, what is the "potential for control" of these targets for a given class of patients? What is the distribution of ancillary service costs (e.g., operating room, recovery room, radiology, and laboratory) for a class of patients? How many of these ancillary costs are potentially controllable? Who determines the utilization of these ancillary services? Are there national benchmarks that can be used to inform these decisions? If opportunities for controlling these ancillary service components are available, what are the key steps in the care process that are either "rate-limiting" or clinically discretionary? These questions highlight the complexity of decision making and imply that the widespread adoption of a particular management technique is unlikely to achieve universal success. Healthcare managers may thus be better served by choosing among a selection of management techniques for the approach most likely to be successful. 

Our study should be considered in the context of several important limitations. First, this was an observational study using a before-and-after design. It was not an experiment with prospective evaluation, although we did adjust for secular trends and case mix. Nevertheless, other possible confounding factors could theoretically affect our findings. Second, we limited our analyses to only those pathways with enough patients to provide a stable estimate of effect. Perhaps pathways with smaller numbers of patients may be systematically more effective. We have no reason to think this assumption is true, however. Third, our analyses were performed at a single site, namely a tertiary-care referral hospital. Thus, our findings may not necessarily be indicative of pathway effectiveness in different clinical settings. Fourth, in an attempt to adjust for secular trends, we may have underestimated the effect of a pathway on either LOS or resource utilization. Despite our belief that adjustment for secular changes is the appropriate way to measure pathway effectiveness given a before-and-after study design, this method of analysis does bias our findings toward the null hypothesis (i.e., no difference between pathway and non-pathway patients). Finally, because we included patients whether or not they were actually managed by the particular pathway, the method of analysis is again biased toward the null. We were primarily interested in the effectiveness-not efficacy-of pathways as a management tool, as the utility of pathways to enhance efficiency in a carefully selected minority of patients within a certain clinical area is not their purpose. 

Despite these limitations, however, our study reveals that it would be prudent to proceed cautiously with the development and implementation of critical pathways. Despite the theoretical benefit of such management tools, pathway effectiveness is still largely unproven. Although earlier studies found these tools to be effective, more recent studies that have considered secular trends have been much more cautionary. Our evaluation reveals that although some pathways significantly reduced LOS and resource utilization, most did not. Future efforts should therefore be placed on critically evaluating the effectiveness of critical pathways and understanding the reasons behind successful and unsuccessful strategies. 

We thank Maureen Thompson, RN, for providing additional information about several of the specific pathways evaluated in this study.

 Et tu, critical pathways? [letter; comment]. 

2. Pearson SD, Goulart-Fisher D, Lee TH. 

Critical pathways as a strategy for improving care: problems and potential. 

3. Every NR, Hochman J, Becker R, Kopecky S, Cannon CP. 

Critical pathways: a review. Committee on Acute Cardiac Care, Council on Clinical Cardiology, American Heart Association. 

Critical pathways: what do you do when they do not seem to work? [letter; comment]. 

Use and evaluation of critical pathways in hospitals. 

6. McMahon LF Jr, Creighton FA, Bernard AM, Pittinger WB, Kelley WN. 

The integrated inpatient management model. A new approach to clinical practice. 

7. Campbell H, Hotchkiss R, Bradshaw N, Porteous M. 

8. Pearson SD, Kleefield SF, Soukop JR, Cook EF, Lee TH. 

Critical pathways intervention to reduce length of hospital stay. 

9. Holmboe ES, Meehan TP, Radford MJ, Wang Y, Marciniak TA, Krumholz HM. 

Use of critical pathways to improve the care of patients with acute myocardial infarction. 

10. Marrie TJ, Lau CY, Wheeler SL, Wong CJ, Vandervoort MK, Feagan BG. 

A controlled trial of a critical pathway for treatment of community-acquired pneumonia. CAPITAL Study Investigators. Community-Acquired Pneumonia Intervention Trial Assessing Levofloxacin. 

11. Mehta RH, Montoye CK, Gallogly M, et al. 

Improving quality of care for acute myocardial infarction: the Guidelines Applied in Practice (GAP) Initiative. 

12. Eagle KA, Gallogly M, Mehta RH, et al. 

Taking the national guideline for care of acute myocardial infarction to the bedside: developing the guideline applied in practice (GAP) initiative in Southeast Michigan. 

Articles

MANAGERIAL

Use of Critical Pathways to Improve Efficiency: A Cautionary Tale

LETTERS TO THE EDITORS

Correspondence Regarding, "Validation of Data Collection for the HEDIS Performance Measure on Chlamydia Screening in a MCO"

To profile the pattern of cyclo-oxygenase 2 inhibitor (COX-2) use, including length of therapy, medical conditions treated, and gastrointestinal (GI) risk profile of users.

Descriptive retrospective analysis of medical and prescription claims data from a large preferred provider organization in the Midwest.

During an index period of January through May 31, 2000, patients new to COX-2 therapy were evaluated 365 days before and after their first prescription. Among the inclusion criteria, patients had to have no previous use of COX-2 therapy, be at least 18 years of age, and be continuously eligible during the entire study period.

Of the more than 300 000 members with at least 1 day of coverage in the index window, 1312 members met the inclusion criteria. The average age of COX-2 users was 49.5 years (SD = 11.4) and 60% were female. The number of daysâ€™ supply of COX-2 agent obtained by members was highly skewed, with a mean of 116 days (SD = 119.5) and a median of 60 days. The medical conditions associated with COX-2 use included a variety of musculoskeletal conditions, the most common being low back pain (22%) and osteoarthritis (18%). Approximately 19% of members did not have a diagnosis associated with COX-2 use. Sixty-five percent of those new to COX-2 therapy did not have an indication of being at risk for GI events, and 68% had no indication for trying a lower-cost nonselective nonsteroidal anti-inflammatory drug (NSAID) prescription prior to beginning COX-2 therapy. Taken together, 45% did not have a GI risk factor or prior use of nonselective NSAID prescription therapy.

These findings suggest that opportunities exist to encourage the cost-effective prescribing of COX-2 therapy. Possible methods include implementation of step therapy, academic detailing, and physician education programs, among others.

From 1999 to 2000, the per member per year costs for the drug class analgesic anti-inflammatories grew by 37.9%, the highest rate of growth among the top 25 therapy categories, as measured by Express Scripts' 

 Contributing to this growth was the introduction of the newer, more expensive cyclo-oxygenase-2 (COX-2) specific inhibitors, whose combined market share grew by approximately 40% from 1999 to 2000.

 The first COX-2 inhibitor approved in the United States was celecoxib (Celebrex) in December 1998 and the second, rofecoxib (Vioxx), entered the market in May 1999. Valdecoxib (Bextra) was approved by the US Food and Drug Administration (FDA) in November 2001. 

    All three COX-2 therapies are approved for relief of the signs and symptoms of osteoarthritis (OA), rheumatoid arthritis (RA), and for the treatment of primary dysmenorrhea. Celebrex and Vioxx are also approved for the management of acute pain in adults. Additionally, Celebrex is approved as adjuvant treatment of familial adenomatous polyposis. The COX-2 specific inhibitors and nonselective nonsteroidal anti-inflammatory drugs (NSAIDs; e.g., ibuprofen, naproxen) have shown similar clinical efficacy at equipotent doses for the management of acute pain and other conditions associated with pain.

    Pharmacologically, COX-2 inhibitors are distinct from nonselective NSAIDs because of their ability to selectively inhibit the COX-2 pathway.

 Nonselective NSAIDs inhibit both COX-1 and COX-2. Inhibition of COX-1 is associated with gastrointestinal (GI) toxicity,

 whereas COX-2 inhibition leads to relief of pain and inflammation. The GI toxicity associated with COX-1 inhibiting agents ranges from symptoms such as nausea, dyspepsia, anorexia, abdominal pain, flatulence, and diarrhea, to potentially life-threatening complications such as upper GI bleeding, perforation, and gastric outlet obstruction.

    It was thought that newly developed drugs designed to block COX-2 but not COX-1 would have anti-inflammatory properties but would avoid the ulcers and other GI events associated with COX-1 inhibition. Two large trials designed to test this theory have been conducted: the Vioxx Gastrointestinal Outcomes Research (VIGOR) trial

 and the Celecoxib Long-Term Arthritis Safety Study (CLASS) trial.

 In the VIGOR trial, the rate of complicated confirmed upper GI events (i.e., perforation, gastric or duodenal ulcer, obstruction, and bleeding) was significantly lower in the rofecoxib group compared with the naproxen group (0.52% vs. 1.22%; 

 However, the CLASS trial, in which celecoxib was compared with traditional NSAIDs (i.e., ibuprofen and diclofenac), did not show a statistically significant difference in the rate of ulcer perforation, gastric outlet obstruction, or upper GI bleeding between celecoxib and nonselective NSAIDs (0.76% vs. 1.45%; 

 Although the rate of serious GI complications observed in these studies was low (< 2%), the absolute number of patients affected is large considering the number of individuals currently taking NSAIDs. With an estimated 13 million Americans taking NSAIDs on a regular basis, annual estimates of hospitalization costs due to serious GI complications exceed $1 billion.

Both rofecoxib and celecoxib have been found to be associated with significantly reduced mild-to-moderate GI adverse events (event rates for rofecoxib compared with naproxen

= .02; celecoxib compared with nonselective NSAID18 = 31.4% vs. 36.8%, 

    One clearly distinguishing feature between COX-2s inhibitors and nonselective NSAIDs is their cost. In 2001, the average wholesale price per prescription was approximately $98 for Celebrex and $86 for Vioxx, compared with $29 for generic nonselective NSAIDs.

 Due in part to the cost difference, many researchers question the widespread use of COX-2s based on symptom benefit alone,

 particularly given that mild symptoms do not correlate well with serious adverse events (perforation, obstruction, ulceration, bleeding) and are often managed clinically with histamine-2 receptor antagonists (H2s) and proton pump inhibitors (PPIs). As a result, some researchers recommend that COX-2s be reserved for those patients at high risk for GI adverse events,

 and formal guidelines have been developed to reflect this advice.

    The rapid increase in the use of the more expensive COX-2 therapies, coupled with their comparable clinical efficacy to traditional NSAIDs, led to a series of research questions surrounding the use of COX-2s. The goal of our study was to answer these questions: 

    - What is the pattern of COX-2 use with respect to patient demographics and length of therapy? 

    - What conditions are COX-2s being used to treat? 

    - Are these agents being reserved for patients at risk for GI-related events? 

    Medical and prescription claims data from a large preferred provider organization in the Midwest were used to evaluate diagnostic conditions and patterns of use among new COX-2 users. The prescription copay structure varied across members, however. More than 90% of sample members had a 3-tier benefit structure with first-tier co-pays varying from $8 to $10, second tier from $15 to $20, and third tier from $25 to $30. COX-2s were on the third tier, and no other restrictions regarding their use were in place during the time of evaluation (i.e., step therapy or prior approval). 

    New COX-2 users were defined as patients with a prescription claim for a COX-2 during the index window January 1, 2000, to May 31, 2000, and with no COX-2 claim during the 12 months prior to the first claim in the index window. Patient's prescription and medical claims were evaluated 365 days before and 365 days after the index prescription (e.g., the first prescription in the index window). Other study inclusion criteria were that patients had to be at least 18 years of age and continuously eligible during the study period. 

    The COX-2 prescription claims were identified using the generic product identifier (GPI; First Databank [Medispan], Indianapolis, Ind.) code beginning "661005." The GPI is a 14-character code consisting of a hierarchy of 7 subsets, each providing more specific information about a drug product. The total days' supply of COX-2 therapy was calculated as the sum of the days' supply field for all COX-2 claims in the 365 days after and including the index prescription. 

Medical conditions were evaluated using the primary, secondary, and tertiary International Classification of Diseases (ICD-9) fields. 

outlines the diagnoses considered to be associated with COX-2 use, including how they were defined and measured. Medical conditions were rank ordered with FDA-approved indications ranked highest, followed by non-FDA approved, but possible indications for use. Patients were assigned to a diagnosis hierarchically. For example, patients with more than one of the diagnoses listed were grouped in the diagnosis category higher on the list. 

Prescription and medical claims data were used to identify members with risk factors for NSAID-related GI events. These criteria were adapted from the literature of established and possible risk factors of NSAID associated GI ulcer.

 The risk factors used in this study and how they were defined and measured are presented in 

Risk for GI-related events was defined as any one of the following factors: history of a GI event; age 60 years or older; recent use of anticoagulant or corticosteroid agents; and higher doses of COX-2 therapy. History of a GI event included diagnosis of gastric, duodenal, or peptic ulcers; 

infection; GI hemorrhage; and evidence of an endoscopy. Prior use of gastroprotective agents (i.e., PPIs and H2 blockers) was also included as a GI risk factor given recent evidence in the literature suggesting that use of these agents may serve as a proxy for GI risk.

 Although not established risk factors, we also included acid-related diagnoses such as gastroesophageal reflux disease, heartburn, and esophagitis, including ulceration or perforation of the esophagus. These factors were included to be as generous as possible in identifying individuals as possible at risk for GI events, recognizing the limitations of claims data and the time over which we evaluated GI risk. As a sensitivity analysis, and because of the lack of specificity around the age criteria, we also included estimates of GI risk using age Â¡Ã 65 years. 

    Of the 319,263 members with at least 1 day of enrollment during the index window, 5,561 had a COX-2 prescription claim. After screening for eligibility and age, 2,450 members remained. An additional 1138 were excluded due to prior use of a COX-2, resulting in a final sample size of 1,312. 

Sample Demographics and Patterns of Use  

Approximately 60% of those new to COX-2 therapy were female; the mean age was 49.5 years (SD = 11.4). The days' supply among new COX-2 users during the 365-day follow-up was highly skewed (

). The mean days' supply was 116 days (SD = 119.5) and the median was 60 days. 

Approximately 19% (n = 245) of members had no medical condition that would indicate the need for a COX-2 inhibitor (Table 1). The diagnosis representing the largest proportion of new COX-2 users was low back pain (22%). Next were diagnoses related to joint pain and inflammation including OA (18%) and joint derangements and other unspecified disorders of the joint (20%). Other conditions related to COX-2 use included various disorders of the muscle tissue and soft tissue, including sprains and strains (3%), disorders of the muscles and tendons and their attachments (6%), and RA (3%). 

 profiles the percent of members meeting each of the GI risk factor criteria and the percent with 1 or more risk factor. Among individuals new to COX- 2 therapy, 18.8% met the age criterion of being at least 60 years. Increasing the age criterion to a minimum of 65 years decreased the absolute percentage meeting the criteria by 12 percentage points to 6.8%. Approximately 17% had a history of GI problems, the vast majority of which represented prior use of gastroprotective agents. Approximately 6% and 3%, respectively, had recent use of corticosteroid agents or anticoagulant agents. 

    Overall, using the age criterion of a minimum of 60 years, 65% of persons new to COX-2 therapy had no evidence of being at risk for NSAID-related GI problems. Using the age criterion of a minimum of 65 years, 73% of those new to COX-2 therapy had no evidence of being at risk for NSAID-related GI problems. 

Prior Use of Nonselective Nonsteroidal Anti-inflammatory Drug Therapy  

Approximately 68% of new COX-2 users did not have a claim for a nonselective NSAID during the 1 year prior to beginning their COX-2 therapy. The likelihood of using a nonselective NSAID prior to COX-2 use was not significantly different among those with and without a GI risk factor (Â¦Ã–2 = 1.07, 

= .302). Taken together, approximately 45% of new COX-2 users did not have a GI risk factor or prior use of a nonselective NSAID agent. 

    The profile of COX-2 use within our sample was for predominantly short-term treatment of a variety of musculoskeletal conditions. Such individuals are often not at risk for GI events, or have not tried a lower-cost nonselective NSAID agent prior to beginning COX-2 therapy. These findings suggest the need for more cost-effective prescribing of COX-2 medications among those new to therapy. This conclusion is based on findings of the sensitivity of COX-2 cost-effectiveness models to assumptions regarding the underlying GI risk of the modeled population,

and no evidence to suggest that COX-2s are cost effective in the short-term treatment of pain. When cost-effectiveness models evaluate the sensitivity of results to assumptions regarding GI risk, incremental costs per event avoided go beyond that considered economically acceptable

 or differences in total expected costs between COX-2s and nonselective NSAIDs are practically indistinguishable.

Several study limitations may limit generalizability of the study findings. The data represent only one plan's experience, and COX-2 prescribing patterns could vary depending on numerous factors including physician incentives and plan design. Limitations of medical claims data, particularly to identify specific diagnoses, are also recognized.

 Diagnoses associated with COX-2 use or GI risk may have been missed because of undercoding. We attempted to overcome these limitations through inclusion of gastroprotective agent therapy identified using prescription claims data, which has been shown to be a reliable and valid source of data compared with medical claims data.

 The time during which we evaluated individual history of GI events was limited to 365 days. We recognize that a GI event could have occurred prior to this look-back period. However, we were generous in our definition of GI risk (i.e., meeting any one criteria), possibly overstating the percent with GI risk. Further research, using a longer look-back period or medical record review, would need to be conducted to determine the extent of false-negative or false-positive identification of risk. 

    The conditions for which COX-2s appeared to be used included a variety of musculoskeletal conditions. Although COX-2s are indicated for acute pain, the fact that close to 1 in 4 new COX-2 users appeared to be prescribed such therapy for the treatment of low back pain was a surprising finding. Etiology of low back pain is often difficult to determine, and the manifestations can include both inflammatory and non-inflammatory pain. Some proportion of COX-2 therapy is likely for low back pain associated with OA; however, further analysis would be needed to determine the extent of OA within this diagnosis category. Approximately 20% of members did not have a diagnosis that would be an indication for COX-2 use, which could be related to the limitations of medical claims data discussed earlier, or use of COX-2 inhibitors in dental practice, data not captured in our medical claims database. 

    Forty-five percent of new COX-2 users-individuals with no GI risk or prior traditional NSAID use-were given a COX-2 inhibitor as first line-therapy when lower-cost nonselective NSAIDs might have been the most cost-effective approach. The premise of this conclusion rests on the wealth of research evaluating predictors of NSAID-associated GI complications, no evidence of superior clinical efficacy of COX-2s, and no evidence of the cost effectiveness of COX-2s in patients not at risk for GI events. The research on predictors of NSAID-related GI events have resulted in a generally accepted list of factors thought to significantly increase a patient's risk of developing an NSAID-related GI ulcer.

 While the presence of any 1 risk factor is an important determinant in predicting overall GI risk, the presence of more than 1 factor has also been shown to increase a patient's risk for GI toxicity. Silverstein and colleagues noted that among the 4 risk factors found to be significantly related to the risk of serious upper GI complications, patients with no risk factor have a 0.4% risk for having a serious GI complication, those with any 1 risk factor have ~1% risk, and those with all 4 risk factors a 9% risk.

    The evidence of GI risk should also take into consideration the fact that more than half of the patients who were new to COX-2 therapy obtained at most a 60 days' supply of therapy during the 1-year follow-up. This pattern of use is not unlike that observed in studies of OA patients prescribed nonselective NSAIDs-half of the patients on aspirin, ibuprofen, naproxen, or piroxicam had discontinued therapy after 66, 53, 51, and 120 days, respectively.

    With regard to duration of therapy and the risk of nonselective NSAID-related GI adverse events, whereas the relative risk of adverse GI events is greatest within the first month of therapy,40 findings from the Arthritis, Rheumatism, and Aging Medical Information System database (ARAMIS) suggest that the risk of GI bleed remains constant during a 10-year follow-up.

 The ARAMIS findings indicated that patients taking NSAIDs for 5 years have a 5-times greater risk of bleeding than do patients taking NSAIDs for 1 year. The bleeding risk for individuals undergoing NSAID therapy for 3 months was one quarter that for individuals taking NSAIDs for 1 year. The degree of GI risk associated with short-term NSAID use is an important issue for further research, given the large amount of short-term COX-2 use. 

    The high proportion of COX-2 use by persons with no evidence of GI risk or trial of nonselective NSAID, together with the substantial amount of short-term use, suggest that steps could be taken to encourage more cost-effective prescribing of COX-2 agents. Ways in which plans can encourage the cost-effective prescribing of COX-2s include programs such as step therapy, prior approval, and academic detailing. Step therapy is a program designed to manage the use of select or first-line therapies (i.e., nonselective NSAIDs) before alternative or second-line therapies (i.e., COX-2s) are covered. These programs are typically automated at the point of service. The limitation of step therapy for NSAIDs is that this strategy does not take into consideration the GI risk profile of the patient. 

    Prior authorization requires that approval for the product be given prior to prescribing the product. The criteria for determining approval can vary, but in the case of COX-2s, could include the generally agreed upon list of established risk factors.

 Prior approval has the advantage of ensuring that these high-cost products are reserved for patients at risk for NSAID-related GI events; however, plans would have to balance program costs carefully against the potential savings given the high volume of calls that such a program could generate. To overcome the limitations of both step therapy and prior approval for this therapy class, some plans have used a combination of step therapy and prior approval using an automated process at the point of service. 

    An example of a combined step therapyÂ¡ÃŒprior approval program was described by Tucker and colleagues. 

Their step therapy component included prior use of nonselective NSAIDs and the prior approval component included checks for GI risk (i.e., prior use of a gastroprotective agent or other ulcer treatment, or anticoagulant or antiplatelet therapy).

 Other criteria could also be automated, including patient age and use of corticosteroids. When implementing prior approval programs for COX-2s, plan administrators should be aware that the age selected as the criterion for increased GI risk is important in determining the overall percent of members who would be affected by these programs. Our results varied by approximately 8 percentage points when the age criterion was 60 versus 65 years. There is no definitive age at which a person suddenly is at risk for an NSAID-related GI event; rather, findings suggest that the relationship between GI risk and age is linear, increasing steadily by ~4% per year increase in age.

    Evidence that plans are implementing academic detailing or physician education programs for COX-2s has also appeared in the literature.

 Treatment guidelines based on The Standardized Calculator of Risk for Events (SCORE) criteria were disseminated among physicians of a northern California HMO and reinforced through academic detailing by drug education pharmacists. SCORE is an instrument designed to estimate an individual's risk for serious NSAID-induced GI toxicity (e.g., stomach ulcer or bleeding ulcer).

 It is based on 6 criteria: age, self-reported health status, diagnosis of RA, use of oral corticosteroids, hospitalization for a stomach or intestinal problem or prior history of stomach ulcer, and history of GI side effects (heartburn, stomach pain, nausea, vomiting) when taking NSAIDs. SCORE was developed by researchers at Stanford University using data from ARAMIS.

    These examples are just some of the steps plan sponsors are taking to encourage the cost-effective use of COX-2 therapy. As more health plans move toward adoption of these programs, research should be conducted to ensure that the risk identification tools avoid false-negative results (i.e., those with risk are identified as not having risk) and false-positive results (i.e., those without risk are identified as having risk). Such research is critical to the provision of quality, cost-effective healthcare.

1. Teitelbaum F, Martinez R, Parker A, Kolling B, Svirnovskiy Y

Maryland Heights, Mo: Express Scripts, Inc; June 2001.

Celebrex [package insert]. Skokie, Ill: G.D. Searle & Co; April 24, 2000.

Vioxx [package insert]. Whitehouse Station, NJ: Merck & Co; July 2000.

4. Bensen WG, Fiechtner JJ, McMillen JI, et al. 

Treatment of osteoarthritis with celecoxib, a cyclo-oxygenase-2 inhibitor: a randomized controlled trial. 

5. Simon LS, Weaver AL, Graham DY, et al. 

Anti-inflammatory and upper gastrointestinal effects of celecoxib in rheumatoid arthritis: a randomized controlled trial. 

Celecoxib versus diclofenac in long-term management of rheumatoid arthritis: randomized double-blind comparison. 

7. Dougados M, Behier MJ, Jolchine I, et al. 

Efficacy of celecoxib, a cyclo-oxygenase 2-specific inhibitor, in the treatment of ankylosing spondylitis. 

8. Acevedo E, Castaneda O, Ugaz M, et al. 

Tolerability profiles of rofecoxib (Vioxx) and Arthrotec. 

A randomized trial of the efficacy and tolerability of the COX-2 inhibitor rofecoxib vs ibuprofen in patients with osteoarthritis. 

10. Saag K, Van Der Heijde D, Fisher C, et al, for the Osteoarthritis Studies Group. 

Rofecoxib, a new cyclo-oxygenase 2 inhibitor, shows sustained efficacy, comparable with other nonsteroidal anti-inflammatory drugs. 

11. Cannon GW, Caldwell JR, Holt P, et al, for the rofecoxib phase III protocol 035 study group. 

Rofecoxib, a specific inhibitor of cyclo-oxygenase 2, with clinical efficacy comparable with that of diclofenac sodium. 

12. Malmstrom K, Daniels S, Kotey P, et al. 

Comparison of rofecoxib and celecoxib, two cyclo-oxygenase-2 inhibitors, in postoperative dental pain: a randomized, placebo- and active-comparator-controlled clinical trial. 

13. Morrison BW, Christensen S, Yuan W, et al. 

Analgesic efficacy of the cyclo-oxygenase-2-specific inhibitor rofecoxib in post-dental surgery pain: a randomized, controlled trial. 

14. Morrison BW, Daniels SE, Kotey P, et al. 

Rofecoxib, a specific cyclo-oxygenase-2 inhibitor, in primary dysmenorrhea: a randomized controlled trial. 

Efficacy of single-dose and multidose rofecoxib in the treatment of post-orthopedic surgery pain. 

The coxibs, selective inhibitors of cyclo-oxygenase-2. 

17. Bombadier C, Laine L, Reicin A, et al. 

Comparison of upper gastrointestinal toxicity of rofecoxib and naproxen in patients with rheumatoid arthritis. 

18. Silverstein FE, Faich G, Goldstein JL, et al. 

Gastrointestinal toxicity with celecoxib vs nonsteroidal anti-inflammatory drugs for osteoarthritis and rheumatoid arthritis: the CLASS study: a randomized controlled trial. 

CLINICAL

Prescribing COX-2s for Patients New to Cyclo-oxygenase Inhibition Therapy

This activity is designed for primary care physicians and others who care for patients with chronic obstructive pulmonary disease (COPD), care managers, and health plan leaders.

To review the latest research into the economic and social burden of the escalating numbers of COPD cases and to present information on how to apply practical guidelines that can positively impact the prevention and progression of the disease by early detection and intervention.

Upon successful completion of this educational program, the reader should be able to:

1. Assess the economic and social impact of COPD in the US.

3. Recognize the importance of smoking cessation in altering the course of COPD.

Program reviewed and released November 15, 2003; program expires November 15, 2004.

The University of Pennsylvania School of Medicine is accredited by the Accreditation Council for Continuing Medical Education to provide continuing medical education for physicians.

The University of Pennsylvania School of Medicine designates this educational activity for a maximum of 1 category 1 credit toward the AMA Physicianâ€™s Recognition Award. Each physician should claim only those credits that he/she actually spent in the activity. To receive credit for your completion of this educational activity, you should read and review the material contained in this article, correctly answer the posttest questions, and complete the evaluation and request for credit information. All requests for credit must be submitted to the University of Pennsylvania School of Medicine Office of Continuing Medical Education within the term of approval for this activity and a score of at least 80% correct achieved on the posttest in order to receive CME credit. The estimated time to complete this activity is 1 hour. This CME activity was produced under the supervision of Katrina Armstrong, MD, MSc, Assistant Professor of Medicine and Clinical Epidemiology and Senior Scholar, Center for Clinical Epidemiology and Biostatistics, University of Pennsylvania School of Medicine.

This research was funded in part by a grant from Boehringer Ingelheim Pharmaceutical, Inc. Danbury, Conn. The authors have disclosed no significant financial interests or other relationships with commercial entities related directly or indirectly to this educational activity.

The authors have indicated that this article does not reference unlabeled/unapproved uses of drugs or devices. Affiliation and Contact Information From National Jewish Medical and Research Center, Denver, Colo. Address correspondence to: David Tinkelman, MD, National Jewish Medical and Research Center, 1400 Jackson Street, Denver, CO 80206.

To determine the socioeconomic, demographic, and clinical utilization characteristics of the â€œtypicalâ€ patient with chronic obstructive pulmonary disease (COPD).

Retrospective analysis of a large dataset of patients enrolled in a disease management program for COPD. 

Patients were 2129 individuals with a diagnosis of COPD. Information was collected at the time these individuals entered the National Jewish Medical and Research Center disease management program for COPD.

The data show that the typical stereotype of a COPD patient (ie, elderly and unemployed) may not apply to a large proportion of patients already diagnosed with COPD. Almost half (49.7%) were less than 65 years old. In addition, 46.1% were employed and missed an average of 4.6 days of work in the previous 6 months. 

The cost of COPD to individuals and to society is increasing. Yet it is a potentially preventable and treatable disease. Early detection and intervention are essential.

Chronic obstructive pulmonary disease (COPD) is a progressive, insidious disease that begins long before the diagnosis is made. Because symptoms may be minimal or attributed to aging, most people do not seek help until they have lost enough lung function to make them short of breath when they perform activities of daily living. For sedentary individuals, this may not occur until they have lost more than 50% of their lung function. Before their presentation, however, most patients already will have experienced a decline in the performance of work and leisure activities,

November 2003

Use of Critical Pathways to Improve Efficiency: A Cautionary Tale

Correspondence Regarding, "Validation of Data Collection for the HEDIS Performance Measure on Chlamydia Screening in a MCO"

Prescribing COX-2s for Patients New to Cyclo-oxygenase Inhibition Therapy

Chronic Obstructive Pulmonary Disease: The Impact Occurs Earlier Than We Think

Mechanisms to Improve Pharmaceutical Use in Managed Care: To Study Controls, Control the Studies [Please]

Changing Medication Use in Managed Care: A Critical Review of the Available Evidence

Implementation of Systems Strategies for Breast and Cervical Cancer Screening Services in Health Maintenance Organizations