Abstract
Background
Most women with ovarian cancer (OC) are diagnosed with advanced disease. They often experience recurrence after primary treatment, and their subsequent prognosis is poor. Our goal was to evaluate the association between use of nonsteroidal antiinflammatory drugs (NSAIDs), including regular and low-dose aspirin, and 5-year cancer-specific survival after an OC diagnosis.
Methods
The Ovarian cancer Prognosis And Lifestyle study is a prospective population-based cohort of 958 Australian women with OC. Information was gathered through self-completed questionnaires. We classified NSAID use during the year prediagnosis and postdiagnosis as none or occasional (<1 d/wk), infrequent (1-3 d/wk), and frequent (≥4 d/wk) use. We measured survival from the start of primary treatment: surgery or neoadjuvant chemotherapy for analyses of prediagnosis use, or 12 months after starting treatment (postdiagnosis use) until the earliest of date of death from OC (other deaths were censored) or last follow-up to 5 years. We used Cox proportional hazards regression to estimate adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) and applied inverse-probability of treatment weighting to minimize confounding. We also calculated restricted mean survival times.
Results
Compared with nonusers and infrequent users, we observed better survival associated with frequent NSAID use prediagnosis (HR = 0.73, 95% CI = 0.55 to 0.97) or postdiagnosis (HR = 0.65, 95% CI = 0.45 to 0.94). Estimates were similar for aspirin and nonaspirin NSAIDs, new and continuous users and in weighted models. These differences would translate to a 2.5-month increase in mean survival by 5 years postdiagnosis. There was no association with acetaminophen.
Conclusions
Our findings confirm a previous study suggesting NSAID use might improve OC survival.
Ovarian cancer (OC) is the sixth-leading cause of cancer-related death in women worldwide, with 5-year survival rates lower than 50% (1). Most women are diagnosed with advanced disease and commonly experience recurrence after primary treatment, at which point their prognosis is poor (2,3).
Nonsteroidal antiinflammatory drugs (NSAIDs), including aspirin, are commonly used analgesic and antipyretic drugs. Low-dose aspirin is also used for primary and secondary prevention of cardiovascular diseases (CVD). NSAIDs inhibit the cyclooxygenase enzymes (COX-1/COX-2) that convert arachidonic acid to prostaglandins (4); thus, their pharmacological effect is mainly through their impact on prostaglandin production and prostaglandin-mediated pathways, including homeostasis and inflammation (4). COX-2 plays an essential role in the production of prostaglandins involved in regulating angiogenesis, migration and invasion of tumor cells, and apoptosis, all of which could promote cancer progression (5,6). Aspirin could also improve survival by reducing thromboembolism, a leading cause of mortality in elderly cancer patients (7), or by reducing platelet interactions with tumor cells and/or release of cytokines, chemokines, and other tumor growth factors (8).
In vitro findings have suggested an association between higher expression of COX-2 and angiogenesis, lymph node metastasis, early cancer recurrence, and worse survival (9,10). Although several studies have investigated the association between NSAIDs and OC survival, a recent meta-analysis found most of those that reported associations were likely to have been affected by immortal time bias (11).
We sought to clarify the relation between NSAID use and survival. For comparison, we considered acetaminophen (paracetamol), another common analgesic that relieves symptoms via a different mechanism of action (12). We hypothesized that NSAID use, particularly after successful treatment, would reduce disease recurrence and improve survival, but there would be no association for acetaminophen.
Methods
The Ovarian cancer Prognosis And Lifestyle (OPAL) study is a national prospective cohort of Australian women aged 21-79 years diagnosed with histologically confirmed invasive epithelial ovarian, primary peritoneal, or fallopian tube cancer from 2012 to 2015. Women were identified via the gynecological-oncology clinics at 18 major public and private hospitals across the country. The median time between diagnosis and enrolment into the study was 1.8 (range = 0-12) months. Overall, 958 women (78% of those approached) consented to participate (Figure 1).
Figure 1.
Flow diagram of Ovarian cancer Prognosis And Lifestyle (OPAL) study participants included in analyses. NA-NSAIDs = nonaspirin nonsteroidal antiinflammatory drugs; NSAIDs = nonsteroidal antiinflammatory drugs.
Women completed a baseline questionnaire at recruitment, providing information about their ancestry, education, height, average weight during the 5 years prediagnosis, smoking habit, physical activity, preexisting conditions, and medication use. The options provided for ancestry included Indigenous Australian, Asian, British/Irish, Western/Northern European, Southern European, Eastern European, Middle Eastern, Pacific Islander, South American, Black African, Maori and Other; this information was used to code ethnicity as White/presumed White or Other/mixed as only a minority reported other specific ethnicities. Women also completed questionnaires every 3 months for the first year, specifying their current weight, smoking, physical activity, and medication use. Excluding 15 women missing prediagnosis medication data and 1 with a concurrent diagnosis of pancreatic cancer, we included 942 women in the current study (Figure 1).
For analyses of postdiagnosis use, we additionally excluded women who died within the first year or did not have a complete response to primary treatment, usually based on normalization of CA125 (Figure 1). To avoid misclassifying use because of missing questionnaires, we required women to have completed at least 2 questionnaires during the first year (see Supplementary Methods, available online for information about how missing data were handled). Women who died during the first year or did not respond well to treatment had more aggressive cancers and other comorbidities, and those excluded due to missing questionnaires had more comorbidities and were less active than those included (Supplementary Table 1, available online).
Medication use
We classified women as prediagnosis users if they reported use 1 and more day(s)/week within the year prediagnosis. We classified women as postdiagnosis users if they reported use on 1 and more postdiagnosis questionnaires. We also separated women who first used NSAIDs postdiagnosis (new users) from those who used NSAIDs both before and after diagnosis (continuous users) and compared them with women who did not use NSAIDs in the year before or after diagnosis (never users). We further classified exposure as 1-3 d/wk and 4 and more d/wk (frequent users).
Survival outcomes
Information about cancer treatment, recurrence, and date and cause of death was abstracted from medical records every 12 months to December 2020.
The primary outcome was OC-specific survival (OVS) and, for these analyses, we censored deaths from any other cause; we also assessed overall and progression-free survival (PFS). If a woman died from OC and her progression status was unknown (n = 2), we assumed she experienced progression on her date of death. Because OC was the cause of death for almost all women (98%), the results for overall survival and OVS were essentially the same; thus, we present only OVS and PFS.
To avoid immortal time bias, we ensured that exposure measurement preceded or was simultaneous with the beginning of the follow-up (13). In models assessing prediagnosis use, follow-up started from the date a woman started primary cancer treatment (surgery or neoadjuvant chemotherapy)—this was the same as the date of histological diagnosis for women treated with primary surgery. In models assessing postdiagnosis use, follow-up started from the date of 12-month questionnaire. End of follow-up was determined as the earliest of date of death (OS and OVS), date of disease progression (PFS), date of last follow-up, or 5 years (only 24 women [3%] were lost to follow-up before 4 years). If women had completed the baseline questionnaire (or 12-month questionnaire for postdiagnosis use) after the start of follow-up, we left-truncated their follow-up to the questionnaire date.
Analysis
We used the log-rank test to compare OVS between groups. We estimated hazard ratios (HRs) and 95% confidence intervals (CIs) using Cox proportional hazards regression, adjusted for potential confounders determined by a directed acyclic graph. Prediagnosis models were adjusted for age at diagnosis and Charlson comorbidity score. We additionally adjusted postdiagnosis models for prediagnosis NSAID use (this may have affected the type of tumors that developed) and for cancer stage and residual tumor after debulking surgery to reduce confounding by prognostic factors that might influence medication use. Adjustment for other potential confounders, including education, physical activity, body mass index, smoking, concurrent use of other medications (eg, statins, steroids), germline pathogenic BRCA1/2 variant, and histotype (for postdiagnosis analyses), did not change the results, so we did not include these variables in the final models. No variables violated the proportional hazards assumption.
In all NSAID models, the reference group was women who used neither aspirin nor nonaspirin (NA)-NSAIDs. We first assessed outcomes for aspirin and NA-NSAIDs separately; however, because their mechanism of action is similar, we also combined all NSAIDs (aspirin or NA-NSAIDs) to obtain greater statistical power.
We applied propensity score–based inverse probability (IP) of treatment weighting to minimize confounding by observed potential confounders (Supplementary Methods; Supplementary Table 2, available online). Using IP-weighted models, we estimated the average treatment effect in the entire population (ATE; weighted for all women) (14) and the average treatment effect among treated women (weighted for users) (15).
We also estimated the restricted mean survival time (RMST) (16) at 5 years for NSAID users and nonusers using both unweighted and weighted flexible parametric survival models (17). We calculated the difference in the adjusted RMST between the 2 groups as an alternative measure of the association between NSAID use and survival (18).
Finally, we evaluated the association between acetaminophen use (compared with nonuse) and survival, adjusted for NSAID use. Unlike NSAIDs, women commonly took acetaminophen following surgery; therefore, we repeated these analyses reclassifying women who only reported use on the 3-month questionnaire as nonusers (n = 76).
To assess the potential for confounding to explain our results, we calculated E-values for hazard ratios and the confidence limits closest to the null. An E-value determines the strength of association (on a risk ratio scale) an unmeasured confounder would need to have with both NSAID use and OC mortality to move the hazard ratio (or confidence limit) to the null (19).
Stratified and sensitivity analyses
We also restricted the postdiagnosis analyses to women who received chemotherapy and separately evaluated the outcomes for NSAID use during and after chemotherapy. Furthermore, we evaluated the associations (1) separately for women younger and older than 60 years [higher risk of NSAID-related complications in older women (2,20) in women with early (I/II) and advanced (III/IV) stage disease, and (3) among those with high-grade serous carcinoma, the most common histotype.
In sensitivity analyses we ran models excluding acetaminophen users from the reference group for NSAID comparisons and vice versa. We also repeated analyses including women who did not have a complete response to primary treatment and excluding women who reported CVD at baseline or had a germline BRCA mutation. We also considered restricting models to women who did not receive steroids, but this was only 30% of women so we had insufficient power for this.
All statistical analyses were performed using SAS software v9.4 (SAS Institute Inc., Cary, NC, USA) and the “stpm2” and “stpm2_standsurv” commands (21,22) in Stata version 15 (Stata Corporation, College Station, TX, USA).
Ethics approvals
The OPAL study was approved by the Human Research Ethics Committees of QIMR Berghofer Medical Research Institute and all participating centers, and women provided signed informed consent.
Results
Descriptive characteristics
Overall, 874 women (93%) received chemotherapy (≥3 cycles); 605 (64%) experienced disease progression and 421 (45%) died within 5 years postdiagnosis; OC was the underlying cause of death for 411 of these women (98%). Prediagnosis, 209 (22%) women used NSAIDs (aspirin = 101, NA-NSAIDs = 92, both = 16) and 171 (18%) used acetaminophen. Postdiagnosis, 222 (32%) used NSAIDs (aspirin = 79, NA-NSAIDs = 122, both = 21) and 126 (57%) of these were new users; 394 (57%) used acetaminophen, of whom 290 (74%) were new users.
Nonsteroidal antiinflammatory drugs
Compared with women who did not use any NSAIDS prediagnosis, aspirin users were older and had more comorbidities, such as hypertension, hyperlipidemia, and diabetes, and so were more likely to use other chronic disease medications. Prediagnosis NA-NSAID users, on average, had a higher body mass index than nonusers (Table 1).
Table 1.
Baseline and clinical characteristics of study participants by regular use (≥1 d/wk) of aspirin and nonaspirin NSAIDs prediagnosis (total = 942)
| Variable | No NSAIDs | Aspirina | NA-NSAIDa |
|---|---|---|---|
| (n = 733) | (n = 117) | (n = 108) | |
| Age, mean (SD), y | 59.0 (11.0) | 66.2 (8.0) | 60.3 (9.8) |
| BMI, mean (SD), kg/m2 | 26.6 (5.6) | 27.8 (6.4) | 28.6 (6.9) |
| Ethnicity, No. (%) | |||
| Other/mixed | 93 (12) | 14 (12) | 10 (9) |
| White/presumed White | 637 (88) | 102 (88) | 98 (91) |
| Missing | 3 | 1 | 0 |
| Education (highest level), No. (%) | |||
| High school diploma | 328 (45) | 69 (60) | 55 (51) |
| Technical college | 199 (27) | 22 (19) | 19 (18) |
| University | 206 (28) | 24 (21) | 34 (31) |
| Smoking habit, No. (%) | |||
| Never | 401 (55) | 62 (53) | 53 (49) |
| Former smoker | 246 (34) | 41 (35) | 42 (39) |
| Current smoker | 86 (12) | 14 (12) | 13 (12) |
| Comorbidities, No. (%) | |||
| Hypertension | 212 (29) | 69 (59) | 37 (34) |
| Hyperlipidemia | 206 (28) | 67 (57) | 34 (31) |
| Cardiovascular disease | 11 (1) | 11 (10) | 4 (4) |
| Diabetes | 42 (6) | 23 (20) | 6 (6) |
| Use of other medications, No. (%) | |||
| Statins | 124 (17) | 56 (48) | 22 (20) |
| ARB/ACEI | 168 (23) | 60 (51) | 31 (29) |
| Beta-blockers | 24 (3) | 23 (20) | 5 (5) |
| Calcium channel blockers | 46 (6) | 17 (15) | 8 (7) |
| Metformin | 24 (3) | 11 (9) | 6 (6) |
| Steroids | 6 (1) | 0 (0) | 0 (0) |
| Acetaminophen | 92 (13) | 29 (25) | 59 (55) |
| Charlson comorbidity score, No. (%) | |||
| 0 | 562 (76) | 67 (57) | 71 (66) |
| 1 | 114 (16) | 24 (21) | 24 (22) |
| ≥2 | 58 (8) | 26 (22) | 13 (12) |
| Histologic subtype, No. (%) | |||
| High-grade serous carcinoma | 523 (71) | 93 (79) | 71 (66) |
| Mucinous carcinoma | 38 (5) | 5 (4) | 4 (4) |
| Endometrioid carcinoma | 59 (8) | 5 (4) | 14 (13) |
| Clear cell carcinoma | 45 (6) | 5 (4) | 7 (6) |
| Low-grade serous carcinoma | 31 (4) | 2 (2) | 4 (4) |
| Carcinosarcoma | 26 (4) | 6 (5) | 5 (5) |
| Other | 11 (2) | 1 (1) | 3 (3) |
| FIGO stage at diagnosis, No. (%) | |||
| I | 159 (22) | 11 (9) | 17 (16) |
| II | 56 (8) | 11 (9) | 13 (12) |
| III | 422 (58) | 78 (67) | 68 (63) |
| IV | 96 (13) | 17 (15) | 10 (9) |
| Germline pathogenic BRCA1 or BRCA2 variant, No. (%) | |||
| No | 503 (84) | 82 (84) | 75 (86) |
| Yes | 99 (16) | 16 (16) | 12 (14) |
| Unknown | 131 | 19 | 21 |
| Residual tumor, No. (%) | |||
| No residual | 425 (60) | 63 (56) | 69 (67) |
| Any residual | 280 (40) | 49 (44) | 34 (33) |
| Unknown residual | 28 | 5 | 5 |
| Primary chemotherapy, No. (%) | |||
| Yes <3 cycles | 98 (13) | 7 (6) | 10 (9) |
| Yes ≥3 cycles | 681 (93) | 110 (94) | 98 (91) |
| Unknown | 1 | 0 | 0 |
Regular use represents use at least 1 d/wk of aspirin (low dose or standard dose) or NA-NSAIDs; nonusers are women who did not regularly use aspirin or NA-NSAIDs. The totals sum to over than 942 because 16 women used both aspirin and NA-NSAIDs. ARB/ACEI = angiotensin-receptor blockers/angiotensin-converting-enzyme inhibitors; BMI = body mass index; FIGO = The International Federation of Gynecology and Obstetrics; NA-NSAIDs = nonaspirin nonsteroidal antiinflammatory drugs; NSAIDs = nonsteroidal antiinflammatory drugs.
Crude OVS at 5 years was similar for prediagnosis NSAID users (≥1 day/wk aspirin and/or NA-NSAIDs) and nonusers (approximately 55%); however, it was statistically significantly higher for postdiagnosis users than nonusers (70% vs 61%) (Figure 2).
Figure 2.
Kaplan-Meier curves for cumulative survival of study participants by nonsteroidal antiinflammatory drug (NSAID; aspirin or nonaspirin) use during the year prediagnosis (A) and postdiagnosis (B).
Because the estimates for aspirin and NA-NSAIDs were generally similar (Supplementary Table 3, available online), we report estimates for all NSAIDs (aspirin or NA-NSAIDs) combined. In the unweighted models, we observed a suggestive association between prediagnosis NSAID use and OVS (HR = 0.82, 95% CI = 0.65 to 1.05) and a similar association for PFS. However, when we considered frequency of use, the survival advantage was restricted to frequent users (HR = 0.73, 95% CI = 0.55 to 0.98; E-value = 1.79 [CL = 1.13]) (Table 2); this was primarily daily low-dose aspirin use (73%).
Table 2.
Any use of NSAIDS (aspirin or nonaspirin) or acetaminophen pre- and postdiagnosis and survival in women with ovarian cancer
| Regular use (≥1 d/wk) | No. (event) | Ovarian cancer–specific survival | Progression-free survivala |
|---|---|---|---|
| HR (95% CI)b | HR (95% CI)b | ||
| NSAIDs | |||
| Prediagnosis | |||
| No | 733 (324) | Referent | Referent |
| Yes | 209 (87) | 0.82 (0.65 to 1.05) | 0.85 (0.70 to 1.04) |
| 1-3 d/wk | 65 (29) | 1.07 (0.73 to 1.56) | 1.02 (0.74 to 1.40) |
| ≥4 d/wk | 144 (58) | 0.73 (0.55 to 0.98) | 0.79 (0.62 to 1.00) |
| Postdiagnosis | |||
| No | 476 (181) | Referent | Referent |
| Yes | 222 (64) | 0.65 (0.48 to 0.89) | 0.86 (0.67 to 1.11) |
| 1-3 d/wk | 76 (20) | 0.74 (0.46 to 1.18) | 1.06 (0.74 to 1.52) |
| ≥4 d/wk | 146 (44) | 0.62 (0.43 to 0.89) | 0.76 (0.56 to 1.04) |
| New or continuous use postdiagnosis | |||
| No use pre- or postdiagnosis | 406 (150) | Referent | Referent |
| New use | 126 (38) | 0.72 (0.50 to 1.03) | 0.99 (0.74 to 1.33) |
| 1-3 d/wk | 56 (16) | 0.89 (0.53 to 1.50) | 1.07 (0.70 to 1.63) |
| ≥4 d/wk | 70 (22) | 0.63 (0.40 to 1.00) | 0.94 (0.65 to 1.35) |
| Continuous use | 81 (25) | 0.60 (0.39 to 0.94) | 0.79 (0.54 to 1.16) |
| 1-3 d/wk | 14 (4) | 0.54 (0.20 to 1.48) | 1.29 (0.62 to 1.68) |
| ≥4 d/wk | 67 (21) | 0.62 (0.38 to 1.00) | 0.70 (0.46 to 1.08) |
| Acetaminophen | |||
| Prediagnosis | |||
| No | 770 (331) | Referent | Referent |
| Yes | 171 (80) | 1.07 (0.83 to 1.38) | 1.12 (0.91 to 1.39) |
| 1-3 d/wk | 105 (48) | 1.01 (0.74 to 1.39) | 1.08 (0.83 to 1.40) |
| ≥4 d/wk | 66 (32) | 1.08 (0.74 to 1.57) | 1.15 (0.84 to 1.57) |
| Postdiagnosis | |||
| No | 305 (97) | Referent | Referent |
| Yes | 396 (149) | 1.33 (1.01 to 1.73) | 1.19 (0.94 to 1.50) |
| 1-3 d/wk | 178 (64) | 1.29 (0.94 to 1.79) | 1.04 (0.78 to 1.39) |
| ≥4 d/wk | 217 (85) | 1.34 (0.99 to 1.83) | 1.28 (0.98 to 1.68) |
Excluding women missing information about disease recurrence (pre- and postdiagnosis models) and women who experienced progression within the first year of follow-up (postdiagnosis models). CI = confidence interval; HR = hazard ratio; NSAIDs = nonsteroidal antiinflammatory drugs.
Unweighted Cox models are adjusted for age and comorbidity score. Postdiagnosis models are further adjusted for prediagnosis use of the medication of interest, stage at diagnosis, and residual disease.
The association was stronger for postdiagnosis NSAID use, suggesting an approximately 35% improvement in survival associated with any use (HR = 0.65, 95% CI = 0.48 to 0.89; E-value = 2.03 [CL = 1.39]), which was again stronger for frequent users (HR = 0.62, 95% CI = 0.43 to 0.89; infrequent use 0.74, 95% CI = 0.46 to 1.18). In models comparing new (OVS = 0.72, 95% CI = 0.50 to 1.03) or continuing NSAID use (OVS = 0.60, 95% CI = 0.39 to 0.94) postdiagnosis with never use, the estimates were very similar to the overall results (Table 2). The associations were weaker for PFS.
Because any potential benefit associated with NSAID use appeared to be restricted to frequent use, we combined infrequent NSAID users (1-3 d/wk) with nonusers in the reference group (≥4 vs <4 d/wk) for all further analyses. This did not appreciably alter the estimates for frequent use (Table 3).
Table 3.
Weighted models for the association between frequent use of NSAIDs (aspirin or nonaspirin) and acetaminophen and survival in women with ovarian cancer
| Medication | No. (event) | Ovarian cancer–specific survival |
Progression-free survivala |
||||
|---|---|---|---|---|---|---|---|
| Unweightedb | ATTc | ATEc | Unweightedb | ATTc | ATEc | ||
| HR (95% CI) | HR (95% CI) | HR (95% CI) | HR (95% CI) | HR (95% CI) | HR (95% CI) | ||
| NSAIDs | |||||||
| Prediagnosis | |||||||
| <4 d/wk | 798 (353) | Referent | Referent | Referent | Referent | Referent | Referent |
| ≥4 d/wk | 142 (58) | 0.73 (0.55 to 0.97) | 0.82 (0.58 to 1.14) | 0.67 (0.46 to 0.98) | 0.79 (0.62 to 0.99) | 0.86 (0.66 to 1.14) | 0.77 (0.56 to 1.04) |
| Postdiagnosis | |||||||
| <4 d/wk | 552 (201) | Referent | Referent | Referent | Referent | Referent | Referent |
| ≥4 d/wk | 146 (44) | 0.65 (0.46 to 0.93) | 0.74 (0.48 to 1.12) | 0.69 (0.46 to 1.03) | 0.75 (0.56 to 1.02) | 0.76 (0.54 to 1.07) | 0.89 (0.64 to 1.23) |
| Acetaminophen | |||||||
| Prediagnosis | |||||||
| <1 d/wk | 770 (331) | Referent | Referent | Referent | Referent | Referent | Referent |
| ≥1 d/wk | 171 (80) | 1.07 (0.83 to 1.38) | 0.96 (0.75 to 1.24) | 0.94 (0.71 to 1.24) | 1.12 (0.91 to 1.39) | 0.98 (0.80 to 1.22) | 0.96 (0.76 to 1.22) |
| Postdiagnosis | |||||||
| <1 d/wk | 305 (97) | Referent | Referent | Referent | Referent | Referent | Referent |
| ≥1 d/wk | 396 (149) | 1.33 (1.01 to 1.73) | 1.31 (1.00 to 1.72) | 1.29 (0.99 to 1.67) | 1.19 (0.94 to 1.50) | 1.14 (0.89 to 1.45) | 1.12 (0.89 to 1.41) |
Excludes women missing information about disease recurrence (pre- and postdiagnosis models) and women who experienced progression within the first year of follow-up (postdiagnosis models). ATE = average treatment effect in the entire population; ATT = average treatment effect in treated; CI = confidence interval; HR = hazard ratio; NSAIDs = nonsteroidal antiinflammatory drugs.
Unweighted Cox models are adjusted for age and comorbidity score. Postdiagnosis models are further adjusted for prediagnosis use of medication of interest, stage at diagnosis, and residual disease.
Variables included in the propensity score models: age, body mass index, smoking status, physical activity, comorbidity score, hypertension, cardiovascular diseases, hyperlipidemia, diabetes, concurrent use of other medications (statin, beta-blockers, metformin, calcium channel blockers, angiotensin-converting enzyme inhibitors/angiotensin receptor-blockers), stage at diagnosis, histologic subtype, and residual tumor after surgery.
The results did not appreciably differ by age (OVS [postdiagnosis] ≤60 years= 0.67, 95% CI = 0.37 to 1.21 vs >60 years = 0.64, 95% CI = 0.40 to 1.00) or for women with high-grade serous carcinoma (Supplementary Table 4, available online). The potential survival benefit appeared to be restricted to women with stage III or IV disease (OVS [postdiagnosis] stage III or IV HR = 0.60, 95% CI = 0.41 to 0.87; stage I or II = 1.25, 95% CI = 0.36 to 4.31; Pinteraction = .25); however, the confidence interval for the latter estimate was very wide.
The results also did not appreciably differ when we excluded women with CVD, but the association was slightly attenuated when we included women who did not have a complete response to primary treatment (OVS HR = 0.78, 95% CI = 0.58 to 1.05), especially for new users (HR = 0.82, 95% CI = 059 to 1.15 vs continuing use 0.61, 95% CI = 0.40 to 0.93) (Supplementary Table 5, available online). Exclusion of acetaminophen users from the reference group (OVS [postdiagnosis] HR = 0.67, 95% CI = 0.44 to 1.02) and women with a BRCA mutation (HR = 0.66, 95% CI = 0.45 to 0.98) did not alter the results. The results also did not markedly differ when we looked at timing of use postdiagnosis among women who received chemotherapy (Supplementary Table 6, available online), although we had insufficient power to look at frequent use.
In IP-weighted models, the results for frequent vs nonusers or infrequent users were very similar to the unweighted estimates, suggesting confounding was not a major problem (Table 3). The average treatment effect among treated women models suggested approximately 30% improved OVS among frequent users of NSAIDs postdiagnosis compared with their expected survival if they had been nonusers or infrequent users (HR = 0.74, 95% CI = 0.48 to 1.12). The ATE models suggested a similar survival advantage if all women had used NSAIDs postdiagnosis (HR = 0.69, 95% CI = 0.46 to 1.03).
The survival difference between frequent users and nonusers or infrequent users translates to a difference of approximately 2.5 months in RMST by 5 years after diagnosis (prediagnosis = 2.73, 95% CI = 0.01 to 5.46; postdiagnosis = 2.33, 95% CI = 0.13 to 4.54) (Figure 3). When we repeated this analysis in the weighted models (ATE), the RMST difference for prediagnosis use was slightly different from unweighted models (3.68, 95% CI = 0.55 to 6.82), but it did not change for postdiagnosis use (2.36, 95% CI = 0.04 to 4.67).
Figure 3.
Unweighted survival curves by nonsteroidal antiinflammatory drug (NSAID; aspirin or nonaspirin) use prediagnosis (A) and postdiagnosis (B). CI = confidence interval; diff = difference; RMST = restricted mean survival time (months).
Acetaminophen
On average, prediagnosis acetaminophen users were older and had a lower level of education, higher body mass index , and more comorbidities than nonusers; as a result, they were also more likely to use other chronic disease medications (Supplementary Table 7, available online). Our data suggested no association between prediagnosis use and survival, but worse OVS and PFS associated with postdiagnosis use (Table 2). The results for infrequent and frequent use did not vary markedly (Table 2); therefore, we report results for any use (≥1 vs <1 d/wk). The associations remained unchanged in the weighted models (Table 3). When we reclassified women who reported acetaminophen use only on the 3-month questionnaire as nonusers to exclude use associated with recovery from surgery, the estimates were closer to the null (Supplementary Table 8, available online). The associations were attenuated when we excluded NSAID users from the reference group (OVS [postdiagnosis] HR = 1.08, 95% CI = 0.82 to 1.41).
Discussion
Our data suggest improved survival associated with both prediagnosis and postdiagnosis use of NSAIDs, with little difference between new and continuous users (vs never users). The association was driven by frequent use (which was primarily daily low-dose aspirin) and would translate to a 2.5-month difference in mean survival at 5 years. The IP-weighted models support these findings. Although the survival increase is small, other treatments used for OC may not confer a much greater benefit. For example, a randomized controlled trial comparing addition of bevacizumab vs placebo with standard chemotherapy showed only a 5-month increase in restricted mean survival among women with advanced disease (23).
A recent meta-analysis reported that most previous studies investigating the association between NSAIDs and OC survival were likely affected by immortal time bias (11). Of the 3 unaffected studies, 1 from the United States also suggested a survival advantage among self-reported postdiagnosis users (≥2 d/wk), with a particularly strong association for new use after diagnosis (aspirin HR = 0.44, 95% CI = 0.26 to 0.74, NA-NSAIDs = 0.46, 95% CI = 0.29 to 0.73) (24). The other studies investigating low-dose aspirin and NA-NSAIDs only suggested a possible survival benefit for high-cumulative dose or high-intensity NA-NSAID use (25) and not for low-dose aspirin (26). However, these studies identified users via prescription data so may have misclassified women who bought the medications over-the-counter as nonusers. This would have made the groups more similar, potentially underestimating the association. Although self-reported data are potentially affected by recall bias, they would capture both prescription and over-the-counter use.
The substantial survival benefit seen for new users in the US study (24) raises questions about the similarity between women who initiate NSAIDs postdiagnosis and those who do not, because this could be confounded by disease severity (reverse causality) (27). In women with less severe disease and better prognosis, symptoms might be more manageable with common analgesics (eg, NSAIDs, acetaminophen). They might also be more likely to initiate a preventive medication like low-dose aspirin. In contrast, women with progressive disease might need more potent pain medications. We tried to mitigate this bias by excluding women who did not have a complete response to treatment or survive the first year postdiagnosis. Additional adjustment of postdiagnosis models for the stage of disease, residual tumor, and histotype, and applying weighted models should also reduce the imbalance between users and nonusers regarding disease severity. Furthermore, our data indicating no marked difference between the survival of continuous and new NSAID users (vs never users) provide further reassurance that the observed association between NSAIDs and survival is unlikely to be due to reverse causality (27).
In contrast to NSAIDs, there was no suggestion of an association between acetaminophen use and OC survival; if anything, acetaminophen users experienced worse survival than nonusers. Unlike NSAIDs, acetaminophen does not possess strong peripheral antiinflammatory properties (12), and its antipyretic or analgesic effects are mainly related to its function on central serotonergic neuronal pathways (12). Because any recall error is likely to equally apply to NSAIDs and acetaminophen, this differential pattern of association provides further reassurance that the observed survival advantage with NSAID use might be real.
The main strength of our analysis is that we used several approaches to reduce the risk of bias (reverse causality, immortal time, and confounding) and conducted extensive sensitivity analyses that did not change our results. We used both covariate-adjustment and IP-weighted models to balance the comparison groups for known or suspected confounders.
The main limitations were the relatively limited statistical power and possible unknown or unmeasured confounders. We were able to control for known confounders, but our E-value analysis suggested that an unknown confounder would have to be strongly associated (RR ≥ 2.0) with both NSAID use and OVS to explain away our observed association for postdiagnosis use, and a moderate association (RR ≥ 1.4) could move the upper confidence interval to the null. There might also be misclassification due to recall error. We were unable to assess the reliability of NSAID data, but for other common medications, there was excellent agreement between self-reported and prescription data (κ > 0.8). Furthermore, such error is unlikely to differ based on subsequent survival, so any misclassification would likely be nondifferential, biasing results towards the null. Because our study included mostly White women, we could not consider other racial and ethnic groups. Finally, we did not ask women why they used these medications so could not consider possible confounding by other indications of use.
Our results are consistent with previous data suggesting a possible survival benefit associated with postdiagnosis NSAID use among women with OC. Future research should consider whether any potential benefit might be restricted to subsets of patients.
Supplementary Material
Contributor Information
Azam Majidi, Population Health Program, QIMR Berghofer Medical Research Institute, Brisbane, QLD, Australia; School of Public Health, Faculty of Medicine, University of Queensland, Brisbane, QLD, Australia.
Renhua Na, Population Health Program, QIMR Berghofer Medical Research Institute, Brisbane, QLD, Australia.
Susan J Jordan, Population Health Program, QIMR Berghofer Medical Research Institute, Brisbane, QLD, Australia; School of Public Health, Faculty of Medicine, University of Queensland, Brisbane, QLD, Australia.
Anna DeFazio, Department of Gynaecological Oncology, Westmead Hospital, Westmead, NSW, Australia; Centre for Cancer Research, The Westmead Institute for Medical Research, The University of Sydney, Westmead, NSW, Australia; The Daffodil Centre, The University of Sydney, A Joint Venture with Cancer Council NSW, Sydney, NSW, Australia.
Andreas Obermair, Queensland Centre for Gynaecological Cancers, Royal Brisbane and Women’s Hospital, Brisbane, QLD, Australia.
Michael Friedlander, Department of Medical Oncology, Prince of Wales Hospital and Prince of Wales Clinical School UNSW, Sydney, NSW, Australia.
Peter Grant, Gynecological Oncology Unit, Mercy Hospital for Women, Melbourne, VIC, Australia.
Penelope M Webb, Population Health Program, QIMR Berghofer Medical Research Institute, Brisbane, QLD, Australia; School of Public Health, Faculty of Medicine, University of Queensland, Brisbane, QLD, Australia.
Funding
The OPAL study was funded by the National Health and Medical Research Council (NHMRC) of Australia (GNT1025142, GNT1120431), RN was supported by an NHMRC Program Grant (GNT1073898), PMW was supported by an NHMRC Fellowship (GNT1173346), and AM by a QIMR Berghofer International PhD Scholarship.
Notes
Role of the funder: The study sponsor had no role in the study design, in the collection, analysis or interpretation of the data or the writing of the manuscript.
Disclosures: PMW, MF and ADF have received funding from AstraZeneca for an unrelated study of ovarian cancer.
Author contributions: PMW: Supervision. AM, PMW, RN: Conceptualization, investigation, methodology, and formal analysis. AM, PMW: Writing—original draft. RN, AO, ADF, PG, SJJ, MF: Writing—review and editing.
Acknowledgements: We acknowledge the OPAL Study team and all the clinicians and participating institutions who helped make this study possible (see opalstudy.qimrberghofer.edu.au for a complete list). We also thank consumer representatives Karen Livingstone AM, Hélène O'Neill and Merran Williams and all the women who participated.
Data availability
The data that support the findings of this study are not publicly available due to privacy and ethical restrictions; however, they can potentially be made available on reasonable request to the corresponding author.
References
- 1. Sung H, Ferlay J, Siegel RL, et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2021;71(3):209-249. [DOI] [PubMed] [Google Scholar]
- 2. Siegel RL, Miller KD, Jemal A.. Cancer statistics, 2020. CA A Cancer J Clin. 2020;70(1):7-30. [DOI] [PubMed] [Google Scholar]
- 3. Conte C, Fagotti A, Avesani G, et al. Update on the secondary cytoreduction in platinum-sensitive recurrent ovarian cancer: a narrative review. Ann Transl Med. 2021;9(6):510-510. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. DuBois RN, Abramson SB, Crofford L, et al. Cyclooxygenase in biology and disease. FASEB J. 1998;12(12):1063-1073. [PubMed] [Google Scholar]
- 5. Vane JR, Bakhle YS, Botting RM.. Cyclooxygenases 1 and 2. Annu Rev Pharmacol Toxicol. 1998;38(1):97-120. [DOI] [PubMed] [Google Scholar]
- 6. Wang D, DuBois RN.. Prostaglandins and cancer. Gut. 2006;55(1):115-122. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Li P, Ning Y, Li M, et al. Aspirin is associated with reduced rates of venous thromboembolism in older patients with cancer. J Cardiovasc Pharmacol Ther. 2020;25(5):456-465. [DOI] [PubMed] [Google Scholar]
- 8. Tao DL, Tassi Yunga S, Williams CD, et al. Aspirin and antiplatelet treatments in cancer. Blood. 2021;137(23):3201-3211. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Gu P, Su Y, Guo S, et al. Over-expression of COX-2 induces human ovarian cancer cells (CAOV-3) viability, migration and proliferation in association with PI3-k/Akt activation. Cancer Invest. 2008;26(8):822-829. [DOI] [PubMed] [Google Scholar]
- 10. Sun H, Zhang X, Sun D, et al. COX-2 expression in ovarian cancer: an updated meta-analysis. Oncotarget. 2017;8(50):88152-88162. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Majidi A, Na R, Dixon-Suen S, et al. Common medications and survival in women with ovarian cancer: a systematic review and meta-analysis. Gynecol Oncol. 2020;157(3):678-685. [DOI] [PubMed] [Google Scholar]
- 12. Jóźwiak-Bebenista M, Nowak JZ.. Paracetamol: mechanism of action, applications and safety concern. Acta Pol Pharm. 2014;71(1):11-23. [PubMed] [Google Scholar]
- 13. Chubak J, Boudreau DM, Wirtz HS, McKnight B, Weiss NS.. Threats to validity of nonrandomized studies of postdiagnosis exposures on cancer recurrence and survival. J Natl Cancer Inst. 2013;105(19):1456-1462. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Hernán MA, Robins JM.. Causal Inference: What If. Boca Raton, FL: Chapman & Hall/CRC; 2020. [Google Scholar]
- 15. Lunt M, Solomon D, Rothman K, et al. ; British Society for Rheumatology Biologics Register Control Centre Consortium. Different methods of balancing covariates leading to different effect estimates in the presence of effect modification. Am J Epidemiol. 2009;169(7):909-917. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Hasegawa T, Misawa S, Nakagawa S, et al. ; JPMA Task Force Members. Restricted mean survival time as a summary measure of time to event outcome. Pharm Stat. 2020;19(4):436-453. [DOI] [PubMed] [Google Scholar]
- 17. Lambert PC, Royston P.. Further development of flexible parametric models for survival analysis. Stata J. 2009;9(2):265-290. [Google Scholar]
- 18. Conner SC, Sullivan LM, Benjamin EJ, et al. Adjusted restricted mean survival times in observational studies. Stat Med. 2019;38(20):3832-3860. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Linden A, Mathur MB, VanderWeele TJ.. Conducting sensitivity analysis for unmeasured confounding in observational studies using E-values: the evalue package. Stata J. 2020;20(1):162-175. [Google Scholar]
- 20. Lanza FL, Chan FKL, Quigley EMM, et al. Guidelines for prevention of NSAID-related ulcer complications. Am J Gastroenterol. 2009;104(3):728-738. [DOI] [PubMed] [Google Scholar]
- 21. Lambart PC. stpm2 (software). https://pclambert.net/software/stpm2/. Updated 2017. Accessed March 23, 2022.
- 22. Lambart PC. stpm2 (software). https://pclambert.net/software/stpm2_standsurv/ Updated 2017. Accessed March 23, 2022.
- 23. Oza AM, Cook AD, Pfisterer J, et al. Standard chemotherapy with or without bevacizumab for women with newly diagnosed ovarian cancer (ICON7): overall survival results of a phase 3 randomised trial. Lancet Oncol. 2015;16(8):928-936. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Merritt MA, Rice MS, Barnard ME, et al. Pre-diagnosis and post-diagnosis use of common analgesics and ovarian cancer prognosis (NHS/NHSII): a cohort study. Lancet Oncol. 2018;19(8):1107-1116. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Verdoodt F, Dehlendorff C, Friis S, et al. Non-aspirin NSAID use and ovarian cancer mortality. Gynecol Oncol. 2018;150(2):331-337. [DOI] [PubMed] [Google Scholar]
- 26. Verdoodt F, Kjaer SK, Dehlendorff C, et al. Aspirin use and ovarian cancer mortality in a Danish nationwide cohort study. Br J Cancer. 2018;118(4):611-615. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Hernan MA, Sauer BC, Hernandez-Diaz S, et al. Specifying a target trial prevents immortal time bias and other self-inflicted injuries in observational analyses. J Clin Epidemiol. 2016;79:70-75. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data that support the findings of this study are not publicly available due to privacy and ethical restrictions; however, they can potentially be made available on reasonable request to the corresponding author.