Inference for covariate-adjusted time-dependent prognostic accuracy measures.

Evaluating the prognostic performance of candidate markers for future disease onset or progression is one of the major goals in medical research. A marker's prognostic performance refers to how well it separates patients at the high or low risk of a future disease state. Often the discriminative performance of a marker is affected by the patient characteristics (covariates). Time-dependent receiver operating characteristic (ROC) curves that ignore the informativeness of the covariates will lead to biased estimates of the accuracy parameters. We propose a time-dependent ROC curve that accounts for the informativeness of the covariates in the case of censored data. We propose inverse probability weighted (IPW) estimators for estimating the proposed accuracy parameters. We investigate the performance of the IPW estimators through simulation studies and real-life data analysis.

Screening histories and contact with physicians as determinants of cervical cancer risk in Montreal, Quebec.

BACKGROUND: Cervical cancer (cca) is largely a preventable disease if women receive regular screening, which allows for the detection and treatment of preinvasive lesions before they become invasive. Having been inadequately screened is a common finding among women who develop cca. Our primary objective was to determine the Pap screening histories of women diagnosed with cca in Montreal, Quebec. Secondary objectives were to determine the characteristics of women at greatest risk of cca and to characterize the level of physician contact those women had before developing cca. METHODS: The Invasive Cervical Cancer Study, a population-based case-control study, consisted of Greater Montreal residents diagnosed with histologically confirmed cca between 1998 and 2004. Respondents to the 2003 Canadian Community Health Survey and a sample of women without cca obtained from Quebec medical billing records served as controls. RESULTS: During the period of interest, 568 women were diagnosed with cca. Immigrants and women speaking neither French nor English were at greatest risk of cca. Most of the women in the case group had been screened at least once during their lifetime (84.8%-90.4%), but they were less likely to have been screened within 3 years of diagnosis. Having received care from a family physician or a medical specialist other than a gynecologist within the 5 years before diagnosis was associated with a greater risk of cca development. CONCLUSIONS: Our findings provide evidence of the need for an organized population-based screening program. They also underscore the need for provider education to prevent missed opportunities for cca screening when at-risk women seek medical attention.

Impact of HLA mismatch at first kidney transplant on lifetime with graft function in young recipients.

As HLA matching has been progressively de-emphasized in the American deceased donor (DD) kidney allocation algorithm, concerns have been raised that poor matching at first transplant may lead to greater sensitization and more difficulty finding an acceptable donor for a second transplant should the first transplant fail. We compared proportion of total observed lifetime with graft function after first transplant, and waiting times for a second transplant between individuals with different levels of HLA mismatch (MM) at first transplant. We studied patients recorded in the United States Renal Data System (1988-2009) who received a first DD transplant at age ≤21 years (n = 8433), and the subgroup who were listed for a second DD transplant following first graft failure (n = 2498). Compared with recipients of 2-3 MM first grafts, 4-6 MM graft recipients spent 12% less of their time and 0-1 MM recipients 15% more time with a functioning graft after the first transplant (both p < 0.0001); 4-6 MM recipients were significantly less likely (hazard ratio [HR] 0.87 [95% confidence interval 0.76, 0.98]; p = 0.03), and 0-1 MM recipients more likely (HR 1.26 [0.99, 1.60]; p = 0.06) to receive a second transplant after listing. The benefits of better HLA matching at first transplant on lifetime with graft function are significant, but relatively small.

Risk of prostate cancer death in long-term users of warfarin: a population-based case-control study.

PURPOSE: Recent evidence suggests that warfarin use may be associated with a reduced risk of prostate cancer. We aimed to determine whether exposure to warfarin is also associated with a reduced risk of prostate cancer death. METHODS: A nested case-control study was conducted within a population-based cohort of 10,012 men aged ≥50 years with newly diagnosed prostate cancer between 1985 and 2002 and with no history of cancer since 1970 using the linked records of Saskatchewan Health and Saskatchewan Cancer Agency registry. We identified 2,309 cases who died of prostate cancer during follow-up. For each case, one control alive at the time of the case's death and matched for length of follow-up (±6 months) was randomly selected. Prescription counts were used to define warfarin exposure. Multivariate conditional logistic regression analysis was used to calculate the adjusted incidence rates of prostate cancer death in relation to warfarin use while adjusting for confounding by age, year of prostate cancer diagnosis, clinical stage and grade of cancer at diagnosis, Chronic Disease Score, and use of warfarin before diagnosis. RESULTS: Ever use of warfarin following a diagnosis of prostate cancer was associated with an adjusted rate ratio of 1.44 (95 % confidence interval (CI) 1.33-1.84) for prostate cancer death. The adjusted rate ratio with one-year use of warfarin was 1.77 (95 % CI 1.25-2.50) compared to never use. The unadjusted rate ratio with five-year use of warfarin was 0.64 (95 % CI 0.40-1.00) and remained unchanged in the adjusted analysis (0.65, 95 % CI 0.37-1.13), although no longer statistically significant. CONCLUSION: Our study does not provide conclusive evidence for a protective effect of long-term warfarin on prostate cancer-specific mortality. Moreover, short-term warfarin use may be associated with an increased risk of prostate cancer death.

Trajectories of endurance activity following pulmonary rehabilitation in COPD patients.

Maintenance of physical activity following pulmonary rehabilitation remains a challenge for patients with chronic obstructive pulmonary disease (COPD). The objectives of this study were to identify patterns of endurance activity after completion of pulmonary rehabilitation and to characterise people who succeed and those who have difficulty maintaining endurance activity. In a longitudinal study embedded within a randomised clinical trial, 206 individuals with COPD underwent a 3-month pulmonary rehabilitation programme. Weekly duration of endurance activity was assessed at 4, 6, 8 and 12 months after the start of rehabilitation. Trajectory modelling was used to determine the most common patterns of activity during the post-rehabilitation phase from 4-12 months. Three distinct patterns were identified, two of which indicated difficulty in maintaining endurance activity: 61 individuals reported a high activity level at 4 months (2.7 h·week(-1)) and stayed high; 114 individuals started at a low activity level (mean 1.0 h·week(-1)) and stayed low; and 31 individuals started high (3.0 h·week(-1)) and declined. The low activity group was characterised by more severe disease and greater respiratory impairment. The high and declined group had less severe disease and respiratory impairment, but reported greater barriers to exercise. Pulmonary rehabilitation should include interventions aimed at minimising barriers, in order to induce long-term behaviour change.

Change in mortality risk over time in young kidney transplant recipients.

Mortality risk for kidney transplant recipients may change with increasing accumulated exposure to the "transplantation milieu." We sought to characterize changes over time in mortality rate and in age-, sex- and race-standardized mortality ratios (SMR) relative to the general population, and to estimate the association between increasing time since first transplant and mortality risk. A total of 18 911 patients who received a first transplant at <21 years old (1983-2006), and whose data were recorded in the USRDS, were studied. There were 2713 deaths over a median follow-up of 8.9 (interquartile range 4.0-14.5; maximum 23) years. Among those with graft function, mortality was highest in the first post transplant year; beyond the first year of the first transplant, age-adjusted mortality rates and SMRs decreased slightly over follow-up. Cause of death was cardiovascular for 34.6%, infection for 19.5%, malignancy for 5.8%, other for 21.4% and unknown for 18.7%. For every 1-year time increment after the end of the first post transplant year, age-adjusted all-cause and cardiovascular mortality rates fell by 1% (p = 0.06) and 16% (p = 0.007), respectively; infection-related mortality rate did not change over time (p = 0.5). These results suggest that exposure to the transplantation milieu has no cumulative negative effects on cardiovascular health over the long term.

Measuring insulin sensitivity in youth: How do the different indices compare with the gold-standard method?

AIM: The objective of the study was to examine the correlation between three methods of measuring insulin sensitivity (IS) - namely, the frequently sampled intravenous glucose tolerance test (FSIVGTT), indices derived from the oral glucose tolerance test (OGTT) and fasting indices (HOMA-IR, QUICKI, fasting insulin [INS(0)]) - and the gold-standard method, the hyperinsulinaemic-euglycaemic clamp (HEC) test, in children. METHODS: A total of 20 children [nine boys and 11 girls; mean (SD) age: 9 (2) years] were studied. Their mean (SD) BMI Z score was 1.5 (0.8). All participants had normal glucose metabolism. Each child underwent a 3-h HEC (40 mU/m(2)/min of insulin), an insulin-modified minimal-model FSIVGTT and a 3-h OGTT. The clamp-derived IS was calculated, using DeFronzo's metabolized glucose index and Bergman's IS index. Correlations were established using Spearman's rank correlations. RESULTS: The two clamp-derived measures were highly correlated (r=0.85), and the IS measured from the FSIVGTT was well correlated with both clamp measures [r=0.69, 0.74]. Of the nine indices derived from the OGTT, the three with the highest correlation with clamp results were the ISI Matsuda [r=0.63, 0.68], SI(is)OGTT [r=0.53, 0.65] and log sum insulin [r=-0.64, -0.75]. Fasting indices of IS had similar correlations to clamp results: HOMA-IR [r=-0.55, -0.56]; QUICKI [r=0.55, 0.57]; and INS(0) [r=-0.59, -0.63]. CONCLUSION: While fasting-based indices of IS are a suitable option for large cohorts, OGTT-derived indices may represent a useful compromise for obtaining both clinical (glucose tolerance) and physiological (insulin sensitivity) information, making them particularly useful for large-scale physiological and epidemiological studies.

Antidepressants and risk of prostate cancer: a nested case-control study.

Although the association between antidepressant drug use and risk of cancer has received considerable attention in the past years, no work has been done specifically on prostate cancer. We carried out a population-based case-control study to assess the risk of prostate cancer in association with exposure to tricyclic antidepressants (TCAs) and selective serotonin reuptake inhibitors (SSRIs). 7767 prostate cancer cases diagnosed between 1981 and 2000 were accrued through the Saskatchewan Cancer Agency. Saskatchewan Health identified a total of 31,068 male controls who were matched on age and calendar time. Data on exposure to TCAs and SSRIs were compiled from the Saskatchewan outpatient prescription drug database, and covered a period upto 24 years. A positive significant association was found between TCA use and risk of prostate cancer, when exposure took place 2-5 years before diagnosis, with rate ratios of 1.31, 1.58, and 2.42 at the low, medium and high average daily dose levels, respectively. Exposure to SSRIs was not found to be significantly associated with the risk of prostate cancer. TCA use 2-5 years in the past was associated with a small dose-dependent increase in the risk of prostate cancer. Nevertheless, detection bias could have contributed to the observed association.

A task-orientated intervention enhances walking distance and speed in the first year post stroke: a randomized controlled trial.

OBJECTIVE: To evaluate the efficacy of a task-orientated intervention in enhancing competence in walking in people with stroke. DESIGN: Two-centre observer-blinded stratified block-randomized controlled trial. SETTING: General community. SUBJECTS: Between May 2000 and February 2003, 91 individuals with a residual walking deficit within one year of a first or recurrent stroke consented to participate. INTERVENTIONS: The experimental intervention comprised 10 functional tasks designed to strengthen the lower extremities and enhance walking balance, speed and distance. The control intervention involved the practice of upper extremity activities. Subjects in both groups attended sessions three times a week for six weeks. MAIN MEASURES: Six-minute walk test (SMWT), 5-m walk (comfortable and maximum pace), Berg Balance Scale, timed 'up and go'. RESULTS: At baseline, subjects in the experimental (n = 44) and control (n = 47) groups walked an average distance of 209 m (SD = 126) and 204 m (SD =131), respectively, on the SMWT. Mean improvements of 40 m (SD =72), and 5 m (SD =66) were observed following the experimental and control interventions, respectively. The between-group difference was 35 m (95% confidence interval (CI) 7, 64). Significant between-group effects of 0.21 m/s (95% CI 0.12, 0.30) and of 0.11 m/s (95% CI 0.03, 0.19) in maximum and comfortable walking speed, respectively, were observed. People with a mild, moderate or severe walking deficit at baseline improved an average of 36 (SD =96), 55 (SD = 56) and 18 m (SD = 23), respectively, in SMWT performance following the experimental intervention. CONCLUSIONS: Study findings support the efficacy of a task-orientated intervention in enhancing walking distance and speed in the first year post stroke, particularly in people with moderate walking deficits.

Insulin resistance syndrome in a representative sample of children and adolescents from Quebec, Canada.

OBJECTIVES: To estimate the prevalence of insulin resistance syndrome (IRS) in a representative sample of youth. To test for the independent contribution of insulin resistance (IR) and adiposity to clustering of metabolic risk factors. To identify the underlying components of IRS. To examine the relationship between adiposity and fasting plasma levels of free fatty acids (FFA). METHODS: In 1999, we conducted a school-based survey of a representative sample of youth aged 9, 13 and 16 y in Quebec, Canada. Age-specific questionnaire data, standardized clinical measurements and a fasting blood sample were available for 2244 subjects. Fasting insulin and HOMA were used as surrogate measures of IR. RESULTS: In all age-sex groups, adiposity indices, blood pressure (BP), plasma glucose and triglycerides (TG) increased significantly with increasing insulin quartiles while HDL cholesterol (HDL-C) decreased. The overall prevalence of IRS defined as hyperinsulinaemia combined with two or more risk factors including overweight, high systolic BP, impaired fasting glucose, high TG and low HDL-C, was 11.5% (95% CI: 10.2-12.9). There were no significant differences in the prevalence of IRS across ages or between sexes. The independent contribution of adiposity to clustering of risk factors was stronger than that of fasting insulin (or HOMA-IR). Factor analysis revealed three factors (BMI/insulin/lipids, BMI/insulin/glucose and diastolic/systolic BP) consistent across ages suggesting that more than one pathophysiologic process underlies IRS. Although elevation of FFA might be in the causal pathway linking obesity to IR, we did not detect any consistent association between measures of fatness and fasting plasma FFA. CONCLUSION: IRS is highly prevalent in youth, even among children as young as age 9 y. Factor analysis identifies three physiologic domains within IRS with a unifying role for markers of IR and adiposity.

The effects of transdermal and oral oestrogen replacement therapy on colorectal cancer risk in postmenopausal women.

The aim of this study was to examine the effects of oral and transdermal oestrogen replacement therapy on the risk of colorectal cancer. Data from a nested case-control study, designed to investigate the effect of hormone replacement therapy (HRT) on colorectal cancer were analysed. New cases of colorectal cancer, diagnosed between 1992 and mid-1998 (N=1197), were identified using the Saskatchewan Cancer Agency cancer registry. Women >/=50 years of age, eligible for coverage by the Saskatchewan Prescription Drug Plan, were included in the study. Four controls per case were age matched to cases, using incidence density sampling. The outpatient prescription drug plan database was used to ascertain oestrogen prescriptions. Women were classified according to history of transdermal (TDE) and oral (OE) oestrogen use. Conditional logistic regression was used to estimate odds ratios (ORs) and 95% confidence intervals (CIs). Compared with women who had never used HRT, ORs for <3 and >/=3 years of TDE use and colorectal cancer were 0.69 (95% CI: 0.43-1.10) and 0.33 (95% CI: 0.12-0.95), and for OE use were 0.90 (95% CI: 0.73-1.01) and 0.75 (95% CI: 0.60-0.93), respectively. The risk reduction for colorectal cancer with TDE may be greater in magnitude than that which has been reported for oral HRT.

Effectiveness of the hands-free technique in reducing operating theatre injuries.

BACKGROUND: Operating theatre personnel are at increased risk for transmission of blood borne pathogens when passing sharp instruments. The hands-free technique, whereby a tray or other means are used to eliminate simultaneous handling of sharp instruments, has been recommended. AIMS: To prospectively evaluate the effectiveness of the hands-free technique in reducing the incidence of percutaneous injuries, contaminations, and glove tears arising from handling sharp instruments. METHODS: For each of 3765 operations carried out in main and surgical day care operating theatres in a large urban hospital, over six months, circulating nurses recorded the proportion of use of the hands-free technique during each operation, as well as other features of the operation. The hands-free technique, considered to be used when 75% or more of the passes in an operation were done in this way, was used in 42% of operations. The relative rate of incidents (percutaneous injuries, contaminations, and glove tears) in operations where the hands-free technique was used and not used, with adjustment via multiple logistic regression for the different risk profiles of the two sets of operations, was calculated. RESULTS: A total of 143 incidents (40 percutaneous injuries, 51 contaminations, and 52 glove tears) were reported. In operations with greater than 100 ml blood loss, the incident rate was 4% (18/486) when the hands-free technique was used and 10% (90/880) when it was not, approximately 60% less. When adjusted for differences in type and duration of surgery, emergency status, noisiness, time of day, and number present for 75% of the operation, the reduction in the rate was 59% (95% CI 23% to 72%). In operations with less than 100 ml blood loss, the corresponding rates were 1.4% (15/1051) when the hands-free technique was used and 1.5% (19/1259) when it was not used. Adjustment for differences in risk factors did not alter the difference. CONCLUSIONS: Although not effective in all operations, use of the hands-free technique was effective in operations with more substantial blood loss.

The effects of tricyclic antidepressants on breast cancer risk.

To test the hypothesis that tricyclic antidepressant use increases invasive female breast cancer incidence, we carried out a case-control study within the population of female beneficiaries of the Saskatchewan Prescription Drug Plan aged 35 years from 1981-995 with no history of cancer since 1970. This agency has provided full or partial coverage for outpatient prescriptions to Saskatchewan residents since 1975. We accrued 5882 histologically proven cases and 23,517 controls, randomly selected from the source population and individually matched on age and sampling time. Heavy exposure to any tricyclic antidepressants was associated with an elevated rate ratio for breast cancer 11-15 years later (2.02, 95% confidence interval: 1.34-3.04). Post hoc analyses based on the results of genotoxicity studies carried out using Drosophila melanogaster suggested that the increased risk could be attributed to the use of the six genotoxic tricyclic antidepressants, and not to the use of the four nongenotoxic tricyclic antidepressants. However, our results may have been confounded by the effects of other determinants of breast cancer associated with tricyclic antidepressant use.

A population-based case-control study of lung cancer and green tea consumption among women living in Shanghai, China.

Epidemiologic evidence regarding the association between the consumption of green tea and lung cancer is limited and inconclusive, although experimental studies have shown consistently that tea preparations and tea polyphenols may inhibit the induction of a variety of cancers, including lung cancer. In this population-based case-control study, we examined the association between past consumption of green tea and the risk of lung cancer. We identified 649 incident cases of primary lung cancer among women diagnosed from February 1992 through January 1994 using the population-based Shanghai Cancer Registry. We randomly selected a control group of 675 women from the Shanghai Residential Registry, frequency-matched to the expected age distribution of the cases. Green tea consumption was ascertained through face-to-face interviews. We estimated adjusted odds ratios (ORs) and 95% confidence intervals (95% CIs) using unconditional logistic regression. Among nonsmoking women, consumption of green tea was associated with a reduced risk of lung cancer (OR = 0.65; 95% CI = 0.45-0.93), and the risks decreased with increasing consumption. We found little association, however, among women who smoked (OR = 0.94; 95% CI = 0.40-2.22). The inconsistency in the association between drinking tea and the risk of lung cancer reported in previous studies may in part be due to inadequate control of confounding of active smoking.

A heuristic approach to the formulas for population attributable fraction.

BACKGROUND: As the definitional formula for population attributable fraction is not usually directly usable in applications, separate estimation formulas are required. However, most epidemiology textbooks limit their coverage to Levin's formula, based on the (dichotomous) distribution of the exposure of interest in the population. Few present or explain Miettinen's formula, based on the distribution of the exposure in the cases; and even fewer present the corresponding formulas for situations with more than two levels of exposure. Thus, many health researchers and public health practitioners are unaware of, or are not confident in their use of, these formulas, particularly when they involve several exposure levels, or confounding factors. METHODS/RESULTS: A heuristic approach, coupled with pictorial representations, is offered to help understand and interconnect the structures behind the Levin and Miettinen formulas. The pictorial representation shows how to deal correctly with several exposure levels, and why a commonly used approach is incorrect. Correct and incorrect approaches are also presented for situations where estimates must be aggregated over strata of a confounding factor.

Association between vitamin A status and lung function level in children aged 6--9 years in Wukro wereda, Northern Ethiopia.

BACKGROUND: In developing countries, studies using morbidity recalls to evaluate the benefits of vitamin A on respiratory health in children under 6 years of age have been inconclusive. This relationship has not been examined in older children. Spirometric measurements, an objective means of assessing respiratory health, require the subject's collaboration and have been successfully used in children over 6 years of age. This report describes a cross-sectional analysis of the relationship between lung function and vitamin A status in an area endemic to vitamin A deficiency. METHODS: The data on which this report is based were gathered prior to the implementation of a prospective trial of the effect of vitamin A supplementation on lung function level in Northern Ethiopia. Vitamin A status was assessed by the Modified Relative Dose Response (MRDR) method and lung function assessed by spirometry in 702 rural children aged 6--9 years. Demographic, personal health, household, environmental and socioeconomic data were gathered by questionnaire. RESULTS: In children with low vitamin A reserve, the unadjusted forced expiratory volume in one second (FEV(1)) was 48.8 ml (P = 0.006) lower than in those with adequate reserve. This difference was 23.1 ml (P = 0.04) when adjusted for age, gender and height and 14.1 ml (P = 0.20) when adjusted for children's demographic, general health, lung function and household-related characteristics. CONCLUSION: Although these findings suggest that vitamin A plays a relatively minor role in determining FEV(1) level, interpretation is limited by the cross-sectional design. Further clarification of its role requires a trial of vitamin A supplementation.

Adverse events associated with prescription drug cost-sharing among poor and elderly persons.

CONTEXT: Rising costs of medications and inequities in access have sparked calls for drug policy reform in the United States and Canada. Control of drug expenditures by prescription cost-sharing for elderly persons and poor persons is a contentious issue because little is known about the health impact in these subgroups. OBJECTIVES: To determine (1) the impact of introducing prescription drug cost-sharing on use of essential and less essential drugs among elderly persons and welfare recipients and (2) rates of emergency department (ED) visits and serious adverse events associated with reductions in drug use before and after policy implementation. DESIGN AND SETTING: Interrupted time-series analysis of data from 32 months before and 17 months after introduction of a prescription coinsurance and deductible cost-sharing policy in Quebec in 1996. Separate 10-month prepolicy control and postpolicy cohort studies were conducted to estimate the impact of the drug reform on adverse events. PARTICIPANTS: A random sample of 93 950 elderly persons and 55 333 adult welfare medication recipients. MAIN OUTCOME MEASURES: Mean daily number of essential and less essential drugs used per month, ED visits, and serious adverse events (hospitalization, nursing home admission, and mortality) before and after policy introduction. RESULTS: After cost-sharing was introduced, use of essential drugs decreased by 9.12% (95% confidence interval [CI], 8.7%-9.6%) in elderly persons and by 14.42% (95% CI, 13.3%-15.6%) in welfare recipients; use of less essential drugs decreased by 15.14% (95% CI, 14.4%-15.9%) and 22.39% (95% CI, 20.9%-23.9%), respectively. The rate (per 10 000 person-months) of serious adverse events associated with reductions in use of essential drugs increased from 5.8 in the prepolicy control cohort to 12.6 in the postpolicy cohort in elderly persons (a net increase of 6.8 [95% CI, 5.6-8.0]) and from 14.7 to 27.6 in welfare recipients (a net increase of 12.9 [95% CI, 10.2-15.5]). Emergency department visit rates related to reductions in the use of essential drugs also increased by 14.2 (95% CI, 8.5-19.9) per 10 000 person-months in elderly persons (prepolicy control cohort, 32.9; postpolicy cohort, 47.1) and by 54.2 (95% CI, 33.5-74.8) among welfare recipients (prepolicy control cohort, 69.6; postpolicy cohort, 123.8). These increases were primarily due to an increase in the proportion of recipients who reduced their use of essential drugs. Reductions in the use of less essential drugs were not associated with an increase in risk of adverse events or ED visits. CONCLUSIONS: In our study, increased cost-sharing for prescription drugs in elderly persons and welfare recipients was followed by reductions in use of essential drugs and a higher rate of serious adverse events and ED visits associated with these reductions.

Nested case-control study of the effects of non-steroidal anti-inflammatory drugs on breast cancer risk and stage.

We carried out a nested case-control study to measure the rate ratio (RR) for invasive female breast cancer in relation to non-steroidal anti-inflammatory drug (NSAID) use. The source population consisted of the female beneficiaries of the Saskatchewan Prescription Drug Plan from 1981 to 1995 with no history of cancer since 1970. Four controls/case, matched on age and sampling time, were randomly selected. Dispensing rates during successive time periods characterized NSAID exposure. RRs associated with exposure during each period were adjusted for exposure during the others. Confounding by other determinants was studied in analyses adjusted with data obtained by interviewing samples of subjects accrued from mid-1991 to mid-1995. We accrued 5882 cases and 23,517 controls. Increasing NSAID exposure 2-5 years preceding diagnosis was associated with a trend towards a decreasing RR (P-trend = 0.003); for the highest exposure level RR = 0.76, 95% confidence interval 0.63-0.92. This protective effect could not be attributed to confounding by other determinants. In analyses involving only the cases, NSAID exposure 2-5 and 6-10 years preceding diagnosis was associated with significantly reduced risks of presenting with a large tumour (> 5 cm diameter) or distant metastasis, but not regional lymph node metastasis. The use of NSAIDs may retard the growth of breast cancers and prevent distant metastasis.