Comparison of the statistical efficiency of case-crossover and case-control designs: application to severe cutaneous adverse reactions.

Although case-crossover analyses have lately emerged as an alternative to case-control analyses in epidemiological studies, it is not yet known in which situations they give reliable conclusions. In this work, the case-crossover and the case-control designs were first compared on the basis of a dataset from a published study of severe cutaneous adverse reactions resulting from drug exposures of various durations and prevalences of use (245 cases, 1147 controls, and exposures to 23 drug classes). Next, the statistical efficiency of each design was compared via Monte Carlo simulations. Eight of the 13 risk factors identified by case-control analysis of the published data were also identified by the case-crossover analysis, with fairly good agreement on ranks of risk estimates (Spearman's correlation coefficient = 0.71, P < 0.001 ). Simulation studies showed that for relative risks below 8, the case-crossover design (250 cases, 4 control periods/case) had a higher power than the case-control design (250 cases, 4 controls/case), and that the case-crossover design was more conservative than the case-control design for prevalences of drug use below 10%. We conclude that the case-crossover design is not suitable for long-term exposures, but is an appropriate alternative for assessing rare risks associated with transient to short-term exposures.

Risk of Stevens-Johnson syndrome and toxic epidermal necrolysis during first weeks of antiepileptic therapy: a case-control study. Study Group of the International Case Control Study on Severe Cutaneous Adverse Reactions.

BACKGROUND: There is still controversy about whether all antiepileptic drugs are associated with the severe cutaneous reactions Stevens-Johnson syndrome (SJS) and toxic epidermal necrolysis (TEN). We have studied the role of antiepileptic drugs in SJS and TEN, taking into account potential cofactors that might confound or modify the risk. METHODS: The case-control study in France, Italy, Germany, and Portugal identified cases of SJS/TEN that developed when the patient was not in hospital and were validated by an expert committee. Controls were patients admitted to the same hospital as the case for an acute illness or an elective procedure. FINDINGS: 73 (21%) of the 352 SJS/TEN cases and 28 (2%) of the 1579 controls reported intake of antiepileptic drugs. Among the 73 exposed SJS and TEN patients, 36 reported intake of phenobarbital, 14 of phenytoin, 21 of carbamazepine, 13 of valproic acid, and three of lamotrigine. Risk was highest in the first 8 weeks after onset of treatment. For individual antiepileptic drugs the univariate relative risk of SJS/TEN for 8 weeks or less of use was 57 (95% CI 16-360; multivariate risk 59 [12-302]) for phenobarbital; 91 (26-infinity) for phenytoin; 120 (34-infinity) for carbamazepine; 25 (5.6-infinity) for lamotrigine, and 24 (5.9-infinity) for valproic acid. The result for valproic acid was based on four case users, all of whom reported concurrent use of other associate drugs. The univariate relative risk for more than 8 weeks of use was 6.2 (2.4-17.0; multivariate risk 2.1 [0.5-9.3]) for phenobarbital, 1.2 (0-5.4) for phenytoin, 0.4 (0.02-2.1) for carbamazepine, and 7.0 (2.4-21.0; multivariate risk 2.0 [0.3-15.0]) for valproic acid. INTERPRETATION: SJS and TEN are associated with short-term therapy with phenytoin, phenobarbital, and carbamazepine. The association with valproic acid seems to be confounded by concomitant short-term therapy with other causal drugs. Lamotrigine also has the potential for severe skin reactions. The period of increased risk is largely confined to the first 8 weeks of treatment.

E3N, a French cohort study on cancer risk factors. E3N Group. Etude Epidémiologique auprès de femmes de l'Education Nationale.

This paper describes the design of E3N, a prospective cohort study conducted in France on risk factors for female cancers. The cohort comprises 100,000 women, aged 40-65 years at baseline in 1990. Participants were asked to complete questionnaires every 18 months. The main hypotheses studied concern the relationship between diet and cancer and between hormonal treatments and cancer. All cancers diagnosed are registered, together with other diseases (cardiovascular diseases, diabetes, osteoporosis). The study population and the follow-up procedure are described.

Medication use and the risk of Stevens-Johnson syndrome or toxic epidermal necrolysis.

BACKGROUND: Toxic epidermal necrolysis and Stevens-Johnson syndrome are rare, life-threatening, drug-induced cutaneous reactions. We conducted a case-control study to quantify the risks associated with the use of specific drugs. METHODS: Data were obtained through surveillance networks in France, Germany, Italy, and Portugal. Drug use before the onset of disease was compared in 245 people who were hospitalized because of toxic epidermal necrolysis or Stevens-Johnson syndrome and 1147 patients hospitalized for other reasons (controls). Crude relative risks were calculated and adjusted for confounding by multivariate methods when numbers were large enough. RESULTS: Among drugs usually used for short periods, the risks were increased for trimethoprim-sulfamethoxazole and other sulfonamide antibiotics (crude relative risk, 172; 95 percent confidence interval, 75 to 396), chlormezanone (crude relative risk, 62; 21 to 188), aminopenicillins (multivariate relative risk, 6.7; 2.5 to 18), quinolones (multivariate relative risk, 10; 2.6 to 38), and cephalosporins (multivariate relative risk, 14; 3.2 to 59). For acetaminophen, the multivariate relative risk was 0.6 (95 percent confidence interval, 0.2 to 1.3) in France but 9.3 (3.9 to 22) in the other countries. Among drugs usually used for months or years, the increased risk was confined largely to the first two months of treatment, when crude relative risks were as follows: carbamazepine, 90 (95 percent confidence interval, 19 to infinity); phenobarbital, 45 (19 to 108); phenytoin, 53 (11 to infinity); valproic acid, 25 (4.3 to infinity); oxicam nonsteroidal antiinflammatory drugs (NSAIDs), 72 (25 to 209); allopurinol, 52 (16 to 167); and corticosteroids, 54 (23 to 124). For many drugs, including thiazide diuretics and oral hypoglycemic agents, there was no significant increase in risk. CONCLUSIONS: The use of antibacterial sulfonamides, anticonvulsant agents, oxicam NSAIDs, allopurinol, chlormezanone, and corticosteroids is associated with large increases in the risk of Stevens-Johnson syndrome or toxic epidermal necrolysis. But for none of the drugs does the excess risk exceed five cases per million users per week.

An international collaborative case-control study of severe cutaneous adverse reactions (SCAR). Design and methods.

A multicenter international case-control study has been designed to elucidate the etiology of Stevens-Johnson syndrome and toxic epidermal necrolysis (TEN). Although these diseases occur rarely, the morbidity is high and the mortality for TEN is of the order of 30%. These serious dermatologic conditions have often been linked to exposure to drugs. Infective and autoimmune diseases, as well as other non-drug risk factors, have also been postulated to be of importance in increasing the risk. The design and methods are described, with particular attention to the unique challenges for an epidemiologic study of these conditions.

Reproductive factors and breast cancer risk. Effect of age at diagnosis.

The data from a French case-control study of 495 patients with breast cancer and 542 control subjects interviewed in five French public hospitals, were analyzed to assess the effect of reproductive factors (age at menarche, age at first full-term pregnancy, the time interval between these two ages, and parity) on the risk of breast cancer. Age at menarche, age at first full-term pregnancy, the time interval between these two ages, and parity appeared to have a limited influence on breast cancer risk. However, the relationship between these factors and the risk of breast cancer varied according to the age at breast cancer diagnosis. In the youngest group of women, the most consistent effects came from factors occurring early in life (menarche, first full-term pregnancy, and consequently the time interval between these two events). These factors had a null or weak effect on the oldest group of women. The protective effect of high parity was confined to the oldest group of women.

[How to measure the risk of breast cancer in a population?]. / Comment mesurer le risque de cancer du sein dans une population?

One can read that one woman in 11 in France and one in nine in the USA will get breast cancer. This high risk is the result of the irrealistic assumption that all women remain exposed to the risk of breast cancer until age 90, and ignores the risk of death before that age. It is possible to allow for the occurrence of death before age 90. We describe the method, and find a risk of getting breast cancer before age 90 of one in 14 in France. The estimation of the risk of breast cancer over limited periods of life may also be useful. We describe how these estimates can be obtained, and present some results. For instance the risk of breast cancer between age 50 and 69 is one in 29, and the risk between 50 and 79 is one in 19.

Triangular test and randomized trials: practical problems in a small cell lung cancer trial.

The triangular test has been used to monitor survival data from a randomized trial in patients with small cell lung cancer. The results of consecutive interim analyses and the problems met by the data monitoring committee and the co-ordinators are described. The methods as well as the consequences of the early stopping on the analysis and the results of this trial are discussed. From this experience, we believe that statistical stopping rules--only one of the factors to be taken into account when deciding to stop a trial--should be used with caution. Independent data monitoring committees may be useful in helping to review the ongoing results and advise the participants in the trial.

Differential effects of tar content, type of tobacco and use of a filter on lung cancer risk in male cigarette smokers.

BACKGROUND: To date, the effect of tar exposure, the use of a filter and the type of tobacco (light or dark) on the risk of lung cancer have not been studied together. METHODS: Data from a case-control study on lung cancer were used to evaluate the specific effects of these three parameters. Cases of lung cancer and controls were interviewed in hospitals in France from 1976 to 1980. The past tar content of cigarettes distributed by the French Tobacco Monopoly was obtained. The results presented concern an exclusively male population of lifelong smokers of French cigarettes and lifelong smokers of light imported cigarettes. The population comprised 1114 cases of histologically confirmed lung cancer and 1466 hospitalized controls. RESULTS: An increase in risk was found among smokers of both dark and light cigarettes (relative risk [RR] = 2.6, 95% confidence interval [CI]:1.1-6.5) and among lifelong smokers of dark cigarettes (RR = 1.7, 95% Cl: 0.9-3.2) compared to lifelong smokers of light cigarettes. Similarly, the risk of lung cancer was higher among smokers of both non-filter and filter cigarettes (RR = 1.6, 95% Cl: 0.9-2.7) and among lifelong smokers of non-filter cigarettes (RR = 1.6, 95% Cl: 0.9-2.8) than among lifelong smokers of filter cigarettes. A borderline significant increase in risk was observed for smokers of French cigarettes compared to smokers of light imported cigarettes (RR = 2.6, 95% Cl: 0.9-7.7). No significant difference in the risk of lung cancer was observed among smokers of French cigarettes according to the proportion of years of smoking high tar cigarettes. CONCLUSION: This study tends to demonstrate a separate effect of type of tobacco, use of a filter and tar content in addition to the usual characteristics of cigarette smoking on lung cancer risk.

Variations in the risk of breast cancer associated with a family history of breast cancer according to age at onset and reproductive factors.

In a case-control study of 495 breast cancer patients and 785 controls between 20 and 56 years of age, the risk of breast cancer associated with a family history of breast cancer was studied according to age and reproductive factors. The familial risk of breast cancer was not significantly modified by age at onset, age at menarche, number of children, age at first full-term pregnancy, menstrual cycle length or age at menopause. However, the familial risk significantly increased with the number of abortions (p < 0.05) and seemed to decrease after a natural menopause (p = 0.08). These results suggest that a familial predisposition to breast cancer exerts the same influence during the first six decades of life, except maybe when there are isolated or repeated events such as abortions or artificially imposed menopause, in which case the risk is apparently greater.

Post-mastectomy megavoltage radiotherapy: the Oslo and Stockholm trials.

The ability of adjuvant radiotherapy to prevent distant metastasis and to prolong survival in patients with early breast cancer is much debated. The paper presents a joint analysis of long-term results (13-16 years' follow-up) from the Oslo and Stockholm randomised trials of post-operative megavoltage radiotherapy versus surgery alone. Among node-positive patients there was a significant 37% relative reduction of distant metastases with radiation (P less than 0.01) and an overall survival difference in favor of the irradiated patients which corresponded with a 22% relative reduction of deaths of borderline significance (P less than 0.06). No significant benefit with radiation in terms of distant metastasis-free survival or overall survival was observed among node-negative patients. The results show that effective local treatment can prevent distant dissemination in some patients and contradict the contention that node-positive breast cancer invariably is a systemic disease already at primary diagnosis.

Long-term risk of sarcoma following radiation treatment for breast cancer.

Between 1954 and 1983, 7620 patients were treated for breast carcinoma at Institut Gustave Roussy (France). Of these patients, 6919 were followed for at least 1 year. Out of these, 11 presented with sarcomas thought to be induced by irradiation, 2 of which were Steward-Treves Syndrome, and 9 of which were sarcomas within the irradiated fields. All histological slides were reviewed and a comparison with those of breast cancer was done. The sites of these sarcomas were: parietal wall, 1 case; second costal cartilage, 1 case; infraclavicular region, 1 case; supraclavicular region, 2 cases; internal third of the clavicle, 2 cases; axillary region 2 cases; and the internal side of the upper arm (Stewart-Treves syndrome), 2 cases. The median age of these 11 patients at the diagnosis of sarcomas was 65.8 (49-83). The mean latent period was 9.5 years (4-24). Three patients underwent radical mastectomy and nine modified radical mastectomy. Only one patient received chemotherapy. The radiation doses received at the site of the sarcoma were 45 Gy/18 fr. for 10 cases and 90-100 Gy for 1 case (due to overlapping between two fields). The histology was as follows: malignant fibrous histiocytoma, 5 cases; fibrosarcoma, 3 cases; lymphangiosarcoma, 2 cases; and osteochondrosarcoma, 1 case. The median survival following diagnosis of sarcoma was 2.4 years (4 months-9 years). Two patients are still alive: one with recurrence of her breast cancer, the other in complete remission, with 7 and 3 years follow-up, respectively. All other patients died from their sarcomas. The cumulative incidence of sarcoma following irradiation of breast cancer was 0.2% (0.09-0.47) at 10 years. The standardized incidence ratio (SIR) of sarcoma (observed n# of cases (Obs)/expected n# of cases (Exp) computed from the Danish Cancer Registry for the same period) was 1.81 (CI 0.91-3.23). This is significantly higher than one, with a p = 0.03 (One Tailed Exact Test). The mean annual excess (Obs-Exp)/100.000 person-years at risk during the same period/(100,000) was 9.92. This study suggests that patients treated by radiation for breast cancer have a risk of subsequent sarcomas that is higher than the general population. However, the benefit from adjuvant radiation therapy in the treatment of breast cancer exceeds the risk of second cancer; therefore, the potential of radiation-induced sarcomas should not be a factor in the selection of treatment for patients with breast cancer.

Prognostic factors in 951 nonmetastatic rhabdomyosarcoma in children: a report from the International Rhabdomyosarcoma Workshop.

A total of 951 newly diagnosed patients with nonmetastatic rhabdomyosarcoma (RMS) under 21 years of age from four different cooperative study groups were analyzed to identify the most important pretreatment tumor characteristics in predicting survival. The patient characteristics considered were tumor invasiveness (T), tumor size, status of locoregional lymph nodes, and primary site. The cooperative groups and studies analyzed were: SIOP [RMS-75 (1975-1984)], the Intergroup Rhabdomyosarcoma Study [IRS-II (1978-1982)], Federal Republic of Germany [CWS-81 (1981-1986)] and Italy [RMS-79 (1979-1986), each study involving a different therapeutic protocol. Patient characteristics identified as having a significant relationship to survival by univariate analysis were (favorable features given in brackets): tumor invasiveness [T1], tumor size [less than or equal to 5 cm], status of locoregional lymph nodes [clinically negative], and primary site [orbit and genitourinary non-bladder prostate (GU-non-BP) sites]. T2 tumors tended to have large size (greater than 5 cm), to be associated with positive nodes, and to have a different distribution by primary site than T1 tumors. A multivariate Cox regression analysis of the pooled data identified tumor invasiveness (T), primary site, and the interaction between T and primary site as significant predictive factors for survival. The prognosis of orbital tumors was consistently favorable regardless of T status, whereas the prognosis for "other sites" was consistently unfavorable. International cooperation in this study has permitted the identification of good and poor risk patient subgroups, permitting the focusing of therapeutic protocols to particular risk groups and setting the stage for common future therapeutic protocols for rare subgroups of patients.

Association between breast cancer and family malignancies.

In a case-control study, the relationship between a family history of cancer of the breast, ovary, colon, uterus or prostate and the risk of breast cancer was investigated. The data consisted of family histories from 495 breast cancer cases and 785 controls aged 20-56 years. A positive association was found between the occurrence of breast cancer and a history of breast cancer in the families of the subjects affected. This relationship increased linearly with both the degree of kinship of the affected relatives and with their number. The risk of breast cancer associated with other types of cancer in the family was not significantly different from unity.

Occupational risk factors for cancer of the pancreas: a case-control study.

The role of the occupational environment in the occurrence of cancer of the pancreas was analysed in a case-control study of 171 cases of pancreatic cancer and 317 controls matched for age at interview, sex, hospital, and interviewer. The study was conducted in France between 1982 and 1985 and covered 15 important industries and various other occupations. The results are given for the whole population, a subgroup of manual workers (46 cases and 108 controls), and a subgroup of French nationals (114 cases and 253 controls). After adjustment for cigarette smoking, and coffee and alcohol consumption, workers in the food industry (OR = 1.86) and the leather industry (OR = 1.63) showed higher risks than other industries. In the sub-group of French nationals only the risk associated with the textile industry was significantly higher than unity (OR = 2.30). No significant increase in risk was associated with work in any of the other branches studied; printing showed a moderate increase in risk (OR = 1.54). The subgroup of manual workers showed an increase in risk for cancer of the pancreas among those working in the building materials and building trades classification (OR = 2.16) and transportation (OR = 1.57).

Risk factors for male breast cancer: a Franco-Swiss case-control study.

In order to evaluate risk factors in male breast cancer, a case-control study of 91 histologically diagnosed cases and 255 cancer controls, matched for age and year of diagnosis, was conducted in Villejuif (France) and Geneva (Switzerland). Factors studied included marital history, occupational exposure, associated diseases, family history of breast cancer, history of radiation exposure and drugs which increase or are suspected of increasing prolactin or estrogen levels. Cases were significantly more likely to be bachelors or to work as butchers than were the controls. Other statistically significant risk factors identified were family history of breast cancer and therapeutic use of digitalis. Elevated risks, although not significant, were observed for individuals with a past history of tuberculosis, hyperthyroidism, liver cirrhosis or gallstone, who had been treated with isoniazid, or who had worked in high environmental temperatures. Further studies are needed to confirm our results particularly as regards hormonal imbalance and genetic factors.

Changes in patterns of cigarette smoking and lung cancer risk: results of a case-control study.

Data from a case-control study on lung cancer were used to evaluate how changes in cigarette habits, mainly smoking cessation, switch from non-filter to filter brands, from dark to light tobacco, or from handrolled to manufactured cigarettes, and reduction in daily consumption influence lung cancer risk. The results presented concern all males, exclusive cigarette smokers, involved in the study, i.e. 1,057 histologically confirmed lung cancer and 1,503 matched controls. The general decrease in lung cancer risk with the years since cessation was also found in each subgroup of cigarette exposure defined by duration of smoking, daily consumption and type of cigarettes smoked. Among smokers who had given up smoking from less than 10 years earlier, the lung cancer risks were two-fold higher for those who had stopped smoking for coughing or health reasons than for those who had stopped smoking for reasons other than health problems. A decrease in lung cancer risk, although not significant, was found in people who switched from non-filter brands to filter brands and from dark to light tobacco and in smokers who reduced their daily consumption of cigarettes by more than 25% as compared to smokers who had not changed habits.

Coffee, alcohol, smoking and cancer of the pancreas: a case-control study.

A hospital case control study on smoking, alcohol and coffee consumption as risk factors for pancreatic cancer, in which 161 cases were matched to 268 control patients, was conducted in France between 1982 and 1985. The results showed no association between tobacco or alcohol consumption and cancer of the pancreas, whereas coffee consumption was associated with an increased risk [OR of 2 or more cups per day vs. less: 2.27 (1.11-4.64) and 1.45 (0.82-2.55) among females and males respectively]. A dose-response relationship was observed with coffee consumption in both sexes, stronger among females than among males: the linear increase estimates were 2.00 (1.22-3.29) and 1.32 (0.91-1.92) respectively. No interaction was found between coffee and tobacco, whereas the effect of coffee appeared to be limited to non-drinkers of alcohol.