PRECISION - Comparison of different NSAIDs in patients with arthritis and an increased CV risk

Nissen SE, et al. Cardiovascular safety of celecoxib, naproxen, or ibuprofen for arthritis. N Engl J Med 2016;online

Bottom line: For patients with arthritis and a ~1%/year risk of CVD and low risk of GI bleeds, no NSAID (between celecoxib, ibuprofen or naproxen) is clearly safer. Consider a strategy of using the cheapest NSAID, PRN if possible, and optimizing other strategies to decrease the CV risk of NSAIDs (smoking cessation, statins, BP control, ASA for secondary prevention, etc). NSAIDs should be discontinued in patients who do not receive a noticeable improvement in pain & function from these agents.


Issues with internal validity?

  • Unclear risk of allocation, performance and detection bias  (likely low risk, but inadequate description in protocol and published report)
  • High risk of attrition and contamination bias
    • 68.8% stopped taking the study drug, though no differences in rates or timing between intervention groups
    • 27.4% discontinued follow-up, and 7.2% were outright lost-to-follow-up
    • The above decrease the study's power to find a difference between groups, especially in the ITT analysis, biasing the conclusions towards non-inferiority. With that said, estimates from ITT and on-treatment populations were almost identical

Special considerations: Non-inferiority trial

  • This study was designed as a non-inferiority safety study to determine that celecoxib was "not much more dangerous" from a CV perspective vs naproxen
  • Criteria to be met to achieve this study's non-inferiority conclusion:
    • Hazard ratio (HR) point estimate <1.12 for the primary outcome
    • Upper limit of HR 97.5% confidence interval <1.33
    • The above had to be met in both intention-to-treat (all patients randomized) and on-treatment populations (events occurring while patient taking study drug + extra 30 days after they discontinued it) to conclude non-inferiority
  • When developing these non-inferiority criteria, assumed would require 762 primary outcome events, and that these would occur at a rate of 2%/year (actual ~1%) & that 40% of patients would discontinue treatment (actual 68%)
    • Actual primary outcome event n=607

Patients (n=24,081)

  • Included
    • Age 18+ y
    • Need for daily NSAID for arthritis pain
    • Increased risk for/established atherosclerotic cardiovascular disease (ASCVD), defined as any of the following:
      • CAD
      • Cerebrovascular disease
      • Symptomatic peripheral vascular disease
      • Arterial surgery/angioplasty for any ASCVD >3 months prior to randomization
      • Diabetes (extra eligibility criterion for women: on insulin)
      • 3+ ASCVD risk factors from the following: Age >55 y, family history of CVD/stroke, current smoker >15 cigarettes/day, HTN, LVH on EKG, dyslipidemia, microalbuminuria or urine protein:creatinine ratio >2, ABI <0.9, waist:hip ratio 0.90 or greater)
  • Excluded
    • ACS, stroke/TIA or CABG within 3 months
    • HF with NYHA class III-IV symptoms or LVEF 35% or less
    • Planned revascularization procedure
    • Uncontrolled arrhythmia within 3 months
    • BP >140/90 mm Hg
    • Malignancy within 5 years
    • Any significant GI, hepatic, renal or coagulation disorders
    • Concomitant meds: Warfarin, prednisone equivalent >20 mg/d, lithium
  • Screened 31,857 -> randomized 24,222 -> analyzed 24,081
  • "Average" patient
    • Age 63 y
    • Female 64%
    • White 75%
    • Arthritis: OA (90%), RA (10%)
      • HAQ disability score 1.1 (0= no disability, 3= complete disability)
      • Pain visual analog score 54 mm (/100)
    • CV risk category: Primary prevention (77%), secondary prevention (23%)
    • PMHx
      • Current smoker 21%
      • HTN 78%
      • Diabetes 35%
      • Dyslipidemia 63%
    • BP 125/75 mm Hg
    • SCr 80 umol/L
    • Meds
      • ASA 46%
      • Statin 54%
      • DMARD 7%

Generalizability: Who does this apply to?

  • Middle-aged to elderly adults with arthritis (mainly OA) with a low-intermediate risk of CV events (~10% over 10 years), and minimal non-CV comorbidity
    • Despite enrolling patients with CV risk factors, actual CV event rates were closer to low-intermediate risk (~1%/year), likely resulting from concomitant treatments to lower CV disease & enrolment of a sample of patients generally healthier than the underlying population.


  • Celecoxib
    • Initial dose 100 mg PO BID
    • (RA only) dose could be increased to 200 mg PO BID
    • Mean treatment duration: 20.8 months
  • Ibuprofen
    • Initial dose 600 mg PO TID
    • Dose could be increased to 800 mg PO TID
    • Note: In practice, most clinicians will recommend a lower dose than what was used in this trial (usually ~200-400 mg PO TID-QID)
    • Mean treatment duration: 19.6 months
  • Naproxen
    • Initial dose 375 mg PO BID
    • Dose could be increased to 500 mg PO BID
    • Mean treatment duration: 20.5 months
  • Co-intervention for all: Gastroprotection with daily PPI (esomeprazole 20-40 mg)

Results (reported in the following order: celecoxib, ibuprofen, naproxen)

ITT analyses (mean follow-up 2.8 years)

  • Death: 1.6% vs 1.8% vs 2.0%
    • Celecoxib vs naproxen: HR 0.80 (0.63-1.00)
  • CV
    • Primary outcome (CV death, non-fatal MI, non-fatal stroke): 2.3% vs 2.7% vs 2.5%
      • Celecoxib vs naproxen: HR 0.93 (0.75-1.13)
      • Ibuprofen vs naproxen: HR 1.08 (0.90-1.31)
    • Major adverse cardiovascular event (primary outcome, coronary revascularization, hospitalization for unstable angina, or TIA): 4.2% vs 4.8% vs 4.3%
      • Ibuprofen vs naproxen: HR 1.11 (0.96-1.29)
    • HF hospitalization: 0.6% in all groups
    • Hospitalization for HTN: 0.3% vs 0.5% vs 0.4%
  • GI - Clinically significant events (GI hemorrhage, obstruction, perforation, or symptomatic gastric/duodenal ulcer): 0.7% vs 0.9% vs 0.7%
    • Celecoxib vs naproxen: HR 0.97 (0.67-1.40)
  • Renal (creatinine increase to >177 umol/L, hospitalization for AKI, or initiation of dialysis): 0.7% vs 1.1% vs 0.9%
    • Celecoxib vs naproxen: HR 0.79 (0.56-1.12)
  • Change in pain visual analogue score from baseline (/100 mm): -9.3 vs -9.5 vs -10.2

On-treatment analyses (mean follow-up 1.8 y [mean treatment duration + 1 month])

  • Primary outcome: 1.7% vs 1.9% vs 1.8%
    • Celecoxib vs naproxen: HR 0.90 (0.71-1.15)
    • Ibuprofen vs naproxen: HR 1.12 (0.89-1.40)


  • Caveats to consider
    • The authors performed over 100 analyses including subgroup analyses in addition to the primary non-inferiority analysis, so chance is a plausible explanation for any differences (or lack of differences) in secondary & tertiary outcomes
    • The upper limit of the non-inferiority margin of 1.33 (a 33% relative risk increase) conceptually accepts that NSAIDs could offset the effect of 1-2 interventions to reduce CV events (i.e. CV risk reduction for smoking cessation, ASA, statins, BP & diabetes control are all in the ballpark of a 15-35% relative reduction)
    • Event rates were lower than predicted based on enrolment criteria, likely due to concomitant treatments to reduce CV events
    • Patients discontinued study treatment at a mean 1.7 years, limiting long-term follow-up and assessment of long-term risk
    • No comparison group that received placebo/no NSAID
  • In the context of the above caveats, this study provides some evidence that
    • Celecoxib, ibuprofen and naproxen have similar CV and renal safety profiles at the doses and duration used in this trial, but cannot absolutely refute other studies demonstrating a greater risk with celecoxib;
    • GI event rates between NSAIDs & celecoxib differed only when iron-deficiency anemia of presumed GI origin was added to the predefined outcome of clinically-significant GI events. The difference was not different when considering only serious events. This weakens the rationale for using celecoxib over a non-selective NSAID;
    • Ibuprofen used at doses higher than routinely prescribed may pose a greater risk of CV and non-serious GI events than celecoxib or naproxen. The implications of this for lower doses are unclear;
  • None of the drugs in this study achieved a minimally clinically important change in the pain score (13.7/100). It's likely that a subset achieved a demonstrable reduction in pain, but for the majority of the trial participants, the benefits were non-existent and therefore risks outweighed the benefits.