In praise of Prais-Winsten: An evaluation of methods used to account for autocorrelation in interrupted time series.

Interrupted time series are increasingly being used to assess the population impact of public health interventions. These data are usually correlated over time (auto correlated) and this must be accounted for in the analysis. Typically, this is done using either the Prais-Winsten method, the Newey-West method, or autoregressive-moving-average (ARMA) modeling. In this paper, we illustrate these methods via a study of pneumococcal vaccine introduction and explore their performance under 20 simulated autocorrelation scenarios with sample sizes ranging between 20 and 300. We show that in terms of mean square error, the Prais-Winsten and ARMA methods perform best, while in terms of coverage the Prais-Winsten method generally performs better than other methods. All three methods are unbiased. As well as having good statistical properties, the Prais-Winsten method is attractive because it is decision-free and produces a single measure of autocorrelation that can be compared between studies and used to guide sample size calculations. We would therefore encourage analysts to consider using this simple method to analyze interrupted time series.

Investigating the effects of long-term dornase alfa use on lung function using registry data.

BACKGROUND: Dornase alfa (DNase) is one of the commonest cystic fibrosis (CF) treatments and is often used for many years. However, studies have not evaluated the effectiveness of its long-term use. We aimed to use UK CF Registry data to investigate the effects of one-, two-, three-, four- and five-years of DNase use on lung function to see if the benefits of short-term treatment use are sustained long term. METHODS: We analysed data from 4,198 people in the UK CF Registry from 2007 to 2015 using g-estimation. By controlling for time-dependent confounding we estimated the effects of long-term DNase use on percent predicted FEV1 (ppFEV1) and investigated whether the effect differed by ppFEV1 at treatment initiation or by age. RESULTS: Considering the population as a whole, there was no significant effect of one-year's use of DNase; change in ppFEV1 over one year was -0.1% in the treated compared to the untreated (p = 0.51) and this did not change with long-term use. However, treatment was estimated to be more beneficial in people with lower lung function (p < 0.001); those with ppFEV1 < 70% at treatment initiation, showed an increase in lung function over one year that was sustained out to five years. The estimated effect of DNase did not depend on age (p = 0.35). CONCLUSIONS: DNase improved lung function in individuals with reduced lung function, bringing a step-change in lung function, but no change in the slope of decline. There was no evidence for a benefit in lung function in those initiating treatment with ppFEV1 > 70%.

Dietary patterns derived with multiple methods from food diaries and breast cancer risk in the UK Dietary Cohort Consortium.

BACKGROUND/OBJECTIVES: In spite of several studies relating dietary patterns to breast cancer risk, evidence so far remains inconsistent. This study aimed to investigate associations of dietary patterns derived with three different methods with breast cancer risk. SUBJECTS/METHODS: The Mediterranean Diet Score (MDS), principal components analyses (PCA) and reduced rank regression (RRR) were used to derive dietary patterns in a case-control study of 610 breast cancer cases and 1891 matched controls within four UK cohort studies. Dietary intakes were collected prospectively using 4- to 7-day food diaries and resulting food consumption data were grouped into 42 food groups. Conditional logistic regression models were used to estimate odds ratios (ORs) for associations between pattern scores and breast cancer risk adjusting for relevant covariates. A separate model was fitted for post-menopausal women only. RESULTS: The MDS was not associated with breast cancer risk (OR comparing first tertile with third 1.20 (95% CI 0.92; 1.56)), nor the first PCA-derived dietary pattern, explaining 2.7% of variation of diet and characterized by cheese, crisps and savoury snacks, legumes, nuts and seeds (OR 1.18 (95% CI 0.91; 1.53)). The first RRR-derived pattern, a 'high-alcohol' pattern, was associated with a higher risk of breast cancer (OR 1.27; 95% CI 1.00; 1.62), which was most pronounced in post-menopausal women (OR 1.46 (95% CI 1.08; 1.98)). CONCLUSIONS: A 'high-alcohol' dietary pattern derived with RRR was associated with an increased breast cancer risk; no evidence of associations of other dietary patterns with breast cancer risk was observed in this study.

Vitamin C intake from diary recordings and risk of breast cancer in the UK Dietary Cohort Consortium.

BACKGROUND/OBJECTIVES: Vitamin C intake has been inversely associated with breast cancer risk in case-control studies, but not in meta-analyses of cohort studies using Food Frequency Questionnaires, which can over-report fruit and vegetable intake, the main source of vitamin C. This is the first study to investigate associations between vitamin C intake and breast cancer risk using food diaries. SUBJECTS/METHODS: Estimated dietary vitamin C intake was derived from 4-7 day food diaries pooled from five prospective studies in the UK Dietary Cohort Consortium. This nested case-control study of 707 incident breast cancer cases and 2144 matched controls examined breast cancer risk in relation to dietary vitamin C intake using conditional logistic regression adjusting for relevant covariates. Additionally, total vitamin C intake from supplements and diet was analysed in three cohorts. RESULTS: No evidence of associations was observed between breast cancer risk and vitamin C intake analysed for dietary vitamin C intake (odds ratios (OR)=0.98 per 60 mg/day, 95% confidence interval (CI): 0.88-1.09, P (trend)=0.7), dietary vitamin C density (OR=0.97 per 60 mg/day, 95% CI: 0.87-1.07, P (trend)=0.5 ) or total vitamin C intake (OR=1.01 per 60 mg/day, 95% CI: 0.99-1.03, P (trend)=0.3). Additionally, there was no significant association for post-menopausal women (OR=1.02 per 60 mg/day, 95% CI: 0.99-1.05, P (trend)=0.3). CONCLUSIONS: This pooled analysis of individual UK women found no evidence of significant associations between breast cancer incidence and dietary or total vitamin C intake derived uniquely from detailed diary recordings.

Self-reported and measured anthropometric data and risk of colorectal cancer in the EPIC-Norfolk study.

BACKGROUND: Epidemiological studies have shown inconsistent results for the association between body size and colorectal cancer (CRC) risk. Inconsistencies may be because of the reliance on self-reported measures of body size. OBJECTIVE: We examined the association of self-reported and directly assessed anthropometric data (body height, weight, body mass index (BMI), waist, hip, waist-to-hip ratio (WHR) and chest circumference) with CRC risk in the EPIC-Norfolk study. DESIGN: A total of 20,608 participants with complete self-reported and measured height and weight and without any history of cancer were followed up an average of 11 years, during which 357 incident CRC cases were recorded. Hazard Ratios (HRs) and 95% confidence intervals (CIs) were estimated using Cox proportional hazards models. RESULTS: After adjustment for confounders, HRs among women in the highest quintile of the body size measure relative to the lowest quintile showed that measured height (HR=1.98, 95% CI=1.19-3.28, P (trend)=0.009), measured waist circumference (HR=1.65, 95% CI=0.97-2.86, P (trend)=0.009) and measured WHR (HR=2.07, 95% CI=1.17-3.67, P (trend)=0.001) were associated with increased CRC risk. Associations using corresponding self-reported measures were attenuated and not statistically significant. Conversely, the association of BMI with CRC risk in women was weaker using measured BMI (HR=1.57, 95% CI=0.91-2.73, P (trend)=0.05) compared with self-reported BMI (HR=1.97, 95% CI=1.18-3.30, P (trend)=0.02). In men no significantly increased CRC risk was observed with any of the anthropometric measures. CONCLUSIONS: Measured height, waist circumference and WHR were associated with CRC risk in women, whereas any significant associations with those measures were attenuated when self-reported data were used.

Effects of body size and sociodemographic characteristics on differences between self-reported and measured anthropometric data in middle-aged men and women: the EPIC-Norfolk study.

BACKGROUND/OBJECTIVES: To investigate the effects of body size and sociodemographic characteristics on differences between self-reported (SR) and measured anthropometric data in men and women. SUBJECTS/METHODS: This study comprises 9933 men and 11,856 women aged 39-79 years at baseline survey (1993-1997) in the EPIC-Norfolk study (Norfolk arm of the European Investigation into Cancer and Nutrition Study). The effects of sex, measured height, weight, age group, educational level and social class on differences between SR and measured weight, height, body mass index (BMI), waist, hip and waist-to-hip ratio (WHR) were examined. RESULTS: There were systematic differences between SR and measured anthropometric measurements by sex, measured height, weight and sociodemographic characteristics. Height was overestimated in both sexes while weight, waist, hip, and consequently, BMI and WHR were underestimated. Being male, shorter, heavier, older, and having no educational qualifications and manual occupation were independently associated with overreporting of height, and underreporting of weight was associated independently with being female, shorter, heavier, younger age, and higher education level and social class. Underreporting of waist circumference was strongly associated with being female and higher measured waist circumference, while underreporting of hip circumference was associated with being male and higher measured hip circumference. Furthermore, there was substantial degree of misclassification of BMI and waist circumference categories for both general and central obesity associated with SR data. CONCLUSIONS: These findings suggest that errors in SR anthropometric data, especially waist and hip circumference are influenced by actual body size as well as sociodemographic characteristics. These systematic differences may influence associations between SR anthropometric measures and health outcomes in epidemiological studies.

Alcohol intake and risk of colorectal cancer: results from the UK Dietary Cohort Consortium.

BACKGROUND: Epidemiological studies have suggested that excessive alcohol intake increases colorectal cancer (CRC) risk. However, findings regarding tumour subsites and sex differences have been inconsistent. METHODS: We investigated the prospective associations between alcohol intake on overall and site- and sex-specific CRC risk. Analyses were conducted on 579 CRC cases and 1996 matched controls nested within the UK Dietary Cohort Consortium using standardised data obtained from food diaries as a main nutritional method and repeated using data from food frequency questionnaire (FFQ). RESULTS: Compared with individuals in the lightest category of drinkers (>0-<5 g per day), the multivariable odds ratios of CRC were 1.16 (95% confidence interval (95% CI): 0.88, 1.53) for non-drinkers, 0.91 (95% CI: 0.67, 1.24) for drinkers with 5-<15 g per day, 0.90 (95% CI: 0.65, 1.25) for drinkers with 15-<30 g per day, 1.02 (95% CI: 0.66, 1.58) for drinkers with 30-<45 g per day and 1.19 (95% CI: 0.75, 1.91) for drinkers with >or=45 g per day. No clear associations were observed between site-specific CRC risk and alcohol intake in either sex. Analyses using FFQ showed similar results. CONCLUSION: We found no significantly increased risk of CRC up to 30 g per day of alcohol intake within the UK Dietary Cohort Consortium.

Inverse sampling of controls in a matched case control study.

A method of inverse sampling of controls in a matched case-control study is described in which, for each case, controls are sampled until a discordant set is achieved. For a binary exposure, inverse sampling is used to determine the number of controls for each case. When most individuals in a population have the same exposure, standard case-control sampling may result in many case-control sets being concordant with respect to exposure and thus uninformative in the conditional logistic analysis. The method using inverse control sampling is proposed as a solution to this problem in situations when it is practically feasible. In many circumstances, inverse control sampling is found to offer improved statistical efficiency relative to a comparable study with a fixed number of controls per case.