• Nonsteroidal antiinflammatory drugs;
  • Coxibs;
  • Practice pattern variation


  1. Top of page
  2. Abstract


To examine the effects of physician specialty and comorbidities on cyclooxygenase 2–selective nonsteroidal antiinflammatory drugs (NSAIDs; coxibs) utilization.


Medical records of 452 patients from a regional managed care organization with ≥3 consecutive NSAID prescriptions from June 1998 to April 2001 were abstracted. Multivariable adjusted associations between coxib initiation and discontinuation and patient and provider characteristics were examined.


A total of 1,142 NSAID prescriptions were written over 9,398 total patient-months of followup. Compared with patients seeing family or general practitioners, patients seeing rheumatologists (odds ratio [OR] 3.4, 95% confidence interval [95% CI] 2.1–5.7) and internists (OR 2.3, 95% CI 1.5–3.6) were significantly more likely to receive a coxib, as well as patients with a history of osteoarthritis (OR 2.6, 95% CI 1.7–3.8), gastrointestinal disease (OR 2.3, 95% CI 1.2–4.5), and congestive heart failure (OR 4.1, 95% CI 1.0–16.4). Although specialists were more likely than generalists to prescribe coxibs, only family or general practitioners were significantly more likely to selectively use coxibs among their patients with a history of gastrointestinal disease. Fifty-four percent of NSAID prescriptions were discontinued, and coxibs were significantly less likely to be discontinued than were traditional NSAIDs (OR 0.6, 95% CI 0.5–0.8).


Our findings suggest significantly greater, but perhaps less selective use of coxibs among specialists, even after accounting for important covariates. The initiation and discontinuation of coxibs was influenced by physician specialty and by patient risk factors.


  1. Top of page
  2. Abstract

Nonsteroidal antiinflammatory drugs (NSAIDs) are among the most frequently prescribed medication class worldwide (1–8). Despite their popularity, patient safety remains a significant concern of chronic NSAID use (3, 9–19). It is unknown, however, whether factors leading to significant variation in safe physician practices observed for other musculoskeletal therapies (5, 20–22) may also influence use of the newer, potentially safer cyclooxygenase 2 (COX-2)–selective NSAIDs (or coxibs).

To identify associations with coxib use, we linked administrative, pharmacy, and medical record review data for a sample of chronic NSAID users who were enrollees in a regional managed health care organization. We hypothesized that risk factors for NSAID toxicity such as older age and comorbidity would be positively associated with increased coxib prescriptions, and that rheumatologists would be more likely than family or general practitioners to prescribe coxibs. In addition, we explored the reasons for discontinuation of NSAIDs.


  1. Top of page
  2. Abstract

Data sources and data collection.

After approval by the University of Alabama at Birmingham Institutional Review Board, we identified NSAID users from a sample of participants from a large, regional managed health care organization. National Drug Codes were used to identify all NSAID prescriptions from pharmacy claims. Study patients were restricted to chronic NSAID users, defined as persons receiving at least 3 consecutive nonaspirin NSAID prescriptions over any time interval from June 1998 to December 1999. To assess provider factors associated with practice pattern variation, NSAID users were stratified based on the specialty of their prescribing physician. We focused on providers who were more likely to care for chronic NSAID users (family or general practitioners, internists, and rheumatologists) and deliberately oversampled rheumatologists to address our hypotheses about the role of provider factors. From a total of 2,334 eligible patients using NSAIDs, 680 patients (29%) and their corresponding 136 providers (5 patients per provider) were randomly selected.

Pharmacy claims data were initially used to identify subjects. Sociodemographic factors, type of insurance coverage, as well as physician specialty and number of covered patients using NSAIDs, were collected from administrative claims and provider databases. Patients were categorized into 2 groups according to the presence or absence of a drug benefit associated with their insurance plan. All subsequent analyses were based on data derived from medical record review.

Medical record abstraction process.

Medical record review included all chart documentation between June 1998 and April 2001. Trained nurse abstractors used a customized version of the MedQuest software (developed by Fu et al under contract from the Health Care Financing Administration) for chart abstraction. Medical record abstractors achieved 97% interrater reliability.

Medical history, physical examination signs, and comorbidities (when recorded in the chart), with special attention to those that might predispose to NSAID use and discontinuation, were collected. Particular attention was devoted to NSAID use and medications that might adversely interact with NSAIDs or were markers for NSAID renal toxicity (diuretics, angiotensin-converting enzyme inhibitors) or were indicative of a predilection to NSAID gastrointestinal (GI) toxicity (corticosteroids, cytoprotective drugs, H2 blockers, and proton pump antagonists). NSAID use during the study period was defined as traditional (nonselective) NSAIDs only, or coxibs ever (including users of coxibs only and users of both types of medications). Intervals of use represented each uninterrupted use of any NSAID during the followup period. One NSAID could have more than one interval if the patient used the drug intermittently. Cumulative use of NSAIDs represented the sum of NSAID interval periods, subtracting the overlap periods (when a patient was taking more than one NSAID). Information on NSAID use was restricted to prescription medication. Over-the-counter medications, such as nonprescription NSAIDs and acetaminophen, were not considered in this study.

Statistical analysis.

Chi-square and one-way analysis of variance or Student's t-tests were used to describe categorical and continuous demographic, clinical, and NSAID utilization variables, respectively. To determine predictors of NSAID use and discontinuation, we performed multivariable logistic regression analysis using the model-building techniques described by Hosmer and Lemeshow (23). Generalized estimating equations (24) were used to adjust for artificial inflation of statistical significance resulting from patients being nested within physicians. Due to their clinical relevance, all multivariable analyses were adjusted for age and sex. For other variables, a bivariate P value ≤0.25 was required to enter the models. Data management, reduction, and analyses were conducted in Microsoft Access (Microsoft Corporation, Redmond, WA), SAS (SAS Institute, Cary, NC) and SPSS (SPSS, Chicago, IL).


  1. Top of page
  2. Abstract

We received a total of 452 records (66% record response rate) from 103 physicians (43 internists, 44 general and family practitioners, 16 rheumatologists) (76% physician response rate). There were no significant differences in sex and age between patients whose records were reviewed and those whose records were not provided.

A total of 1,142 NSAID prescriptions were written over 9,398 total patient-months of followup. Coxibs accounted for 435 (38%) of all NSAID prescriptions. Characteristics of the 452 NSAID users, stratified by type of NSAID used, are shown in Table 1. Compared with traditional NSAID users, coxib users were older, more likely to have history of GI disease and osteoarthritis (OA), and trended toward a higher prevalence of comorbidities and connective tissue diseases. Use of other prescription analgesics showed no association with use of coxibs. Patients using coxibs had a significantly longer cumulative drug use as well as a longer followup during the study period. Coxib ever users had more intervals of use than did traditional NSAID only users (2.9 versus 1.9; P < 0.001) (data not shown). There was no significant difference in the use of coxibs based on the type of drug benefit coverage; if anything, those with limited or no drug benefit trended toward more frequent coxib use.

Table 1. Characteristics of patients receiving traditional nonsteroidal antiinflammatory drugs versus coxibs*
CharacteristicTraditional NSAIDs only users (n = 157)Coxib ever users (n = 295)
  • *

    Except where indicated otherwise, values are the number (%). Coxib ever users were patients who received at least one prescription for rofecoxib or celecoxib. Gastrointestinal disease includes peptic ulcer, gastritis, or gastrointestinal bleeding. NSAID = nonsteroidal antiinflammatory drug.

  • P < 0.01.

  • P < 0.05.

  • §

    P < 0.001.

Age, mean ± SD, years59.8 ± 11.462.8 ± 11.4
Number of concomitant drugs, mean ± SD12.3 ± 7.014.5 ± 8.9
Cumulative duration of NSAID use, mean ± SD months10.4 ± 9.412.7 ± 8.5
Number of observation months, mean ± SD19.6 ± 8.921.4 ± 7.4
Number of physician visits per 12 observation months, mean ± SD6.4 ± 7.25.9 ± 3.5
Sex, women94 (59.9)194 (65.8)
Conditions associated with NSAID toxicity  
 Congestive heart failure2 (1.3)11 (3.7)
 Diabetes28 (17.8)50 (16.9)
 Gastrointestinal disease12 (7.6)46 (15.6)
 Hypertension89 (56.7)145 (49.2)
 Renal diseases4 (2.5)11 (3.7)
Arthritis type  
 Osteoarthritis86 (54.8)216 (73.2)§
 Rheumatoid arthritis14 (8.9)30 (10.2)
 Lupus or connective tissue disease type unknown41 (26.1)99 (33.6)
Possible concomitant medications associated with NSAID toxicity  
 Angiotensin-converting enzyme inhibitors40 (25.5)78 (26.4)
 Diuretics41 (26.1)98 (33.2)
 Other antihypertensives69 (43.9)113 (38.3)
 Corticosteroids37 (23.6)58 (19.7)
 Warfarin3 (1.9)6 (2.0)
 Other analgesics64 (40.8)125 (42.4)
 Any of the above concomitant medications124 (79.0)236 (80.0)

Patients seen by internists were older than patients seen by family or general practitioners and rheumatologists (see Table 2). Those seen by rheumatologists were less likely to have a history of diabetes or hypertension or to use antihypertensive medications.

Table 2. Characteristics of patients receiving traditional nonsteroidal antiinflammatory drugs and coxibs by prescribing physician specialty*
CharacteristicPhysician specialty
Family or general practitioner (n = 184 patients)Internist (n = 194 patients)Rheumatologist (n = 74 patients)
  • *

    Except where indicated otherwise, values are the number (%). Gastrointestinal disease includes peptic ulcer, gastritis, or gastrointestinal bleeding.

  • P < 0.001, by one-way analysis of variance or chi-square trend test.

  • P < 0.01, by one-way analysis of variance or chi-square trend test.

Age, mean ± SD, years57.8 ± 12.569.9 ± 9.460.8 ± 10.0
Comorbid conditions   
 Diabetes33 (17.9)43 (22.2)2 (2.7)
 Gastrointestinal disease24 (13.0)25 (12.9)9 (12.2)
 Hypertension101 (54.9)122 (62.9)11 (14.9)
 Congestive heart failure5 (2.7)8 (4.1)0 (0.0)
 Any of above comorbid conditions11 (60.3)137 (70.6)20 (27.0)
Possible concomitant medications   
 Angiotensin-converting enzyme inhibitors51 (27.7)56 (28.9)11 (14.9)
 Corticosteroids20 (10.9)52 (26.8)23 (31.1)
 Diuretics57 (31.0)67 (34.5)15 (20.3)
 Other antihypertensives68 (37.0)93 (47.9)21 (28.4)
 Any of above concomitant medications110 (59.8)145 (74.7)44 (59.5)

Patients seen by an internist or rheumatologist were more likely to initiate coxibs referent to those seen by a family or general practitioner (P < 0.001). This pattern persisted among patients with a history of GI disease (Figure 1A) or hypertension (Figure 1B). Among patients with either a history of GI disease or potential NSAID gastrointestinal risk factors (age ≥65 years, concomitant use of oral corticosteroids), those seen by an internist or rheumatologist were significantly more likely to receive coxibs than were those seen by a generalist (P = 0.03 and P = 0.003, respectively) (data not shown).

thumbnail image

Figure 1. Proportion of patients with gastrointestinal (GI) disease (A) or hypertension (B) receiving coxibs, stratified by prescribing physician specialty. GI disease includes peptic ulcer, gastritis, or GI bleeding. † = P = 0.04, comparison within family or general practitioners only. ‡ = P = 0.001, comparisons between all physician specialties.

Download figure to PowerPoint

When comparing the use of nonselective NSAID versus coxib use within each physician specialty, family and general practitioners but not internists or rheumatologists were more likely to selectively prescribe coxibs rather than traditional NSAIDs to their patients with a history of GI disease (odds ratio [OR] 2.6, 95% confidence interval [95% CI] 1.0–6.5) (Figure 1A). Only rheumatologists, however, trended toward a preference for coxib selection over traditional NSAIDs for their patients with hypertension (Figure 1B).

In Table 3, after multivariable adjustment for potential confounders, a history of OA, GI disease, or congestive heart failure remained significantly and positively associated with initiation of coxibs. Even after adjustment for these potential confounders, being seen by an internist or rheumatologist was still a significant predictor of coxib use.

Table 3. Patient and provider characteristics associated with ever use of a coxib
VariableUnadjusted odds ratio95% confidence intervalAdjusted odds ratio*95% confidence interval
  • *

    Includes age, sex, and variables with univariate P < 0.25 (osteoarthritis, comorbidities, provider type, insurance drug coverage).

  • Gastrointestinal disease includes peptic ulcer, gastritis, or gastrointestinal bleeding. c statistic = 0.72 (47); Hosmer and Lemeshow goodness of fit statistic P = 0.47 (48).

Age, years    
Sex (referent to men)1.30.9––1.9
Rheumatoid arthritis, systemic lupus erythematosus, or other connective tissue disease1.51.0––2.4
Comorbid conditions    
 Gastrointestinal disease2.21.1––4.5
 Congestive heart failure3.00.7––16.4
Provider type    
 General or family practitioner1.0Referent1.0Referent
Insurance drug coverage1.21.0––2.0

Figure 2 shows discontinuation of NSAIDs according to prescribing physician specialty. Patients seen by family or general practitioner were less likely to discontinue at least one NSAID drug compared with patients seen by internists (OR 0.5, 95% CI 0.2–0.9) and rheumatologists (OR 0.6, 95% CI 0.4–0.9). However, patients of rheumatologists were less likely to completely discontinue all of their NSAIDs compared with patients seen by internists (OR 0.4, 95% CI 0.2–0.9) or family and general practitioners (OR 0.6, 95% CI 0.3–1.3).

thumbnail image

Figure 2. Proportion of patients discontinuing nonsteroidal antiinflammatory drugs (NSAIDs), stratified by prescribing physician specialty. * = P = 0.05, comparisons between all physician specialties.

Download figure to PowerPoint

Patients who stopped at least one NSAID medication were more likely to have a history of GI disease (P = 0.004) and a history of OA (P = 0.045). Stopping all NSAIDs was not significantly associated with any of the patient characteristics examined (data not shown). Patients who had documented complaints of GI problems or whose medical record presented evidence of GI bleeding were more likely to stop at least one NSAID prescription (n = 57; 83%) than were patients for whom these complaints or medical findings were not documented (n = 260; 68%) (P = 0.014). A similar trend was observed for discontinuation of all NSAIDs (23% versus 18%) (data not shown).

We next examined patterns and predictors of NSAID discontinuation at the individual prescription level (Table 4). Of all NSAID prescriptions written (n = 1,142), 615 (54%) were discontinued during followup after a mean ± SD cumulative duration of 5.8 ± 5.9 months per agent. Compared with traditional NSAIDs (411 prescriptions; 58%), coxibs (204 prescriptions; 47%) were significantly less likely to be discontinued (OR 0.6, 95% CI 0.5–0.8). Of the 615 documented NSAIDs that were stopped, 273 (44%) had a documented reason for discontinuation. Among these, lack of efficacy was the most frequent reason (n = 113, 41%) followed by an adverse event (n = 93, 34%). GI complaints (abdominal pain, GI bleeding, or GI complications) were the most frequently documented medical problems considered to be an adverse event by the managing physicians (n = 50, 54%). There was no significant association between NSAID type (coxibs or traditional) and reporting of an adverse event as a reason for discontinuation. Compared with coxibs, traditional NSAIDs were significantly more likely to be discontinued because of a GI adverse event (OR 4.2, 95% CI 1.6–10.9) or a history of GI disease (OR 1.7, 95% CI 1.1–2.6). In contrast, coxibs were more likely to be discontinued when the patient had a history of hypertension (OR 1.5, 95% CI 1.0–2.1).

Table 4. Reasons for discontinuation of traditional nonsteroidal antiinflammatory drugs and coxibs*
ReasonTraditional NSAIDs (n = 707 prescriptions)Coxibs (n = 435 prescriptions)
  • *

    Values are the number (%). Gastrointestinal events include abdominal pain, gastrointestinal bleeding, or gastrointestinal complications. Non-gastrointestinal events include dyspnea, headache, dizziness, renal complications, and central nervous system complaints. NSAIDs = nonsteroidal antiinflammatory drugs.

  • P < 0.01.

  • Explicit reason known for 44% of all NSAIDs discontinued.

  • §

    P < 0.05.

Drug discontinued, n = 615411 (58.1)204 (46.9)
Known reason for discontinuation, n = 273  
 Lack of efficacy72 (10.2)41 (9.4)
 Adverse event66 (9.3)27 (6.2)
 Type of adverse event  
  Gastrointestinal42 (5.9)8 (1.8)
  Non-gastrointestinal3 (0.4)5 (1.1)
  Other or not mentioned21 (3.0)14 (3.2)
 Therapeutic course completed24 (3.4)3 (0.7)
 Too expensive4 (0.6)9 (2.1)§
 Other factors21 (3.0)6 (1.4)

Multivariable regression analyses exploring reasons for NSAID discontinuation showed that patients with a history of GI disease (OR 2.5, 95% CI 1.2–5.3) and those seeing an internist (OR 1.7, 95% CI 1.0–2.7) or a rheumatologist (OR 3.1, 95% CI 1.7–5.6) (referent to those seeing a family or general practitioner) remained more likely to discontinue at least one NSAID (data not shown).


  1. Top of page
  2. Abstract

Despite the increasing worldwide use of coxibs (25, 26), patterns and predictors of coxib use have not been previously studied at a population level. A plausible explanation for the greater use of coxibs by specialists in our study is that internists (and particularly rheumatologists) see referred patients who already have tried traditional NSAIDs and either experienced no benefit or experienced toxicity. It is also possible that these same patients are also more likely to be exposed to and/or influenced by direct-to-consumer advertising. Practice pattern variations among the therapies for musculoskeletal disorder are well documented (20, 22, 27–29). Variations in practice patterns and unequal levels of patient care may be partially attributable to inadequate dissemination of the latest knowledge. Specialists such as rheumatologists are quicker to adopt new treatments when choosing medications (22, 27, 30–34).

It should be noted that of all pharmaceutical products being marketed, rofecoxib had the largest recent advertising budget of any current pharmaceutical product (35). Accordingly, rheumatologists and internists are more frequently exposed to pharmaceutical detailing of newer arthritis products than are family or general practitioners, and this may preferentially influence their prescribing behaviors.

Comorbidity, especially GI disease and cardiac failure, appeared to be a factor favoring the selection of coxibs. This may be a reflection of physicians prescribing these medications to “sicker” patients. Our finding that generalists are more likely to selectively prescribe coxibs to their patients at higher risk for GI disease is important, because >60% of patients with musculoskeletal disorders are treated primarily by generalists (36), and >20% of all encounters with primary care providers are for musculoskeletal complaints (37, 38). Generalists tend to be more conservative and often more cost-effective in their adoption of newer musculoskeletal therapies (30), which may influence their willingness to selectively prescribe coxibs to a patient with a history of GI disease.

Compared with nonselective NSAIDs, coxibs were significantly less likely to be discontinued, suggesting that they may be better tolerated. During the first months of the study period, only traditional NSAIDs were prescribed, and it is possible that coxibs were too new to the drug market and had not been used for a long enough time to be as commonly discontinued. However, the duration of NSAID use did not significantly differ between the 2 classes of drugs in our study.

GI problems have been consistently reported as the most frequently documented adverse event leading to discontinuation of an NSAID (39). As expected, a history of GI disease was the factor most associated with discontinuation of at least one NSAID. Although the numbers were small, nonselective NSAIDs had more discontinuations for GI reasons than did the coxibs. These findings support the premise that GI symptoms or adverse effects may be a major factor associated with switching between NSAIDs. Absence of more comprehensive data on discontinuation prohibits us from deeper exploration of these findings.

This study did not addresses important questions on physicians' prescribing behaviors, such as overprescription of NSAIDs or adherence to community standards of care. We did not define absolute standards against which to measure prescribing quality, because evidence on which to generate consensus remains incomplete.

Because medical record review is constrained by the completeness and accuracy of the information recorded (40–42), our analysis may partially reflect quality of documentation rather than quality of care. In addition, medical charts do not provide accurate information on adherence to medical treatment or on the magnitude of use of over-the-counter medications (43–46). However, our detailed record review protocol with stringent quality control and excellent interrater reliability increase confidence in our results. Another limitation in the present study is that we were not able to ascertain patient race/ethnicity, because the managed care organization did not routinely collect racial or ethnic information. Last, given the restriction of our sample to one geographic area, our results may not be generalizable to all groups of chronic NSAID users. Despite these potential limitations, the data for this study came from a large regional managed health care organization, allowing linkage of data from medical records, pharmacy claims, and administrative files.

In conclusion, we observed significantly greater use of coxibs among specialists, even after accounting for important case-mix covariates. We also determined that patient factors, such as GI disease, had an important impact on both initiation and discontinuation of traditional NSAIDs compared with coxibs. Further research is needed to relate these interesting NSAID use patterns with long-term patient safety.


  1. Top of page
  2. Abstract