Marta Avalos

This paper considers the problem of estimation and variable selection for large high-dimensional data (high number of predictors p and large sample size N, without excluding the possibility that N < p) resulting from an individually matched case-control study. We develop a simple algorithm for the adaptation of the Lasso and related methods to the conditional logistic regression model. Our proposal relies on the simplification of the calculations involved in the likelihood function. Then, the proposed algorithm iteratively solves reweighted Lasso problems using cyclical coordinate descent, computed along a regularization path. This method can handle large problems and deal with sparse features efficiently. We discuss benefits and drawbacks with respect to the existing available implementations. We also illustrate the interest and use of these techniques on a pharmacoepidemiological study of medication use and traffic safety.

Purpose In exploratory analyses of pharmacoepidemiological data from large populations with large number of exposures, both a conceptual and computational problem is how to screen hypotheses using probabilistic reasoning, selecting drug classes or individual drugs that most warrant further hypothesis testing. Methods We report the use of a shrinkage technique, the Lasso, in the exploratory analysis of the data on prescription drugs and road traffic crashes, resulting from the case-crossover matched-pair interval approach described by Orriols and colleagues (PLoS Med 2010; 7:e1000366). To prevent false-positive results, we consider a bootstrap-enhanced version of the Lasso. To highlight the most stable results, we extensively examine sensitivity to the choice of referent window. Results Antiepileptics, benzodiazepine hypnotics, anxiolytics, antidepressants, antithrombotic agents, mineral supplements, drugs used in diabetes, antiparkinsonian treatment, and several cardiovascular drugs showed suspected associations with road traffic accident involvement or accident responsibility. Conclusion These results, in relation to other findings in the literature, provide new insight and may generate new hypotheses on the association between prescription drugs use and impaired driving ability. Copyright © 2013 John Wiley & Sons, Ltd.

Background: Large data sets with many variables provide particular challenges when constructing analytic models. Lasso-related methods provide a useful tool, although one that remains unfamiliar to most epidemiologists. Methods: We illustrate the application of lasso methods in an analysis of the impact of prescribed drugs on the risk of a road traffic crash, using a large French nationwide database (PLoS Med 2010; 7:e1000366). In the original case-control study, the authors analyzed each exposure separately. We use the lasso method, which can simultaneously perform estimation and variable selection in a single model. We compare point estimates and confidence intervals using (1) a separate logistic regression model for each drug with a Bonferroni correction and (2) lasso shrinkage logistic regression analysis. Results: Shrinkage regression had little effect on (bias corrected) point estimates, but led to less conservative results, noticeably for drugs with moderate levels of exposure. Carbamates, carboxamide derivative and fatty acid derivative antiepileptics, drugs used in opioid dependence, and mineral supplements of potassium showed stronger associations. Conclusion: Lasso is a relevant method in the analysis of databases with large number of exposures and can be recommended as an alternative to conventional strategies.

L swimmers were observed weekly for 2 yr. The indexes of cardiac autonomic regulation in supine and orthostatic position were assessed as explanatory variables by time domain (SD1, SD2) and spectral analyses (high frequency [HF] = 0.15-0.40 Hz, low frequency [LF] = 0.04-0.15 Hz, and HF/LF ratio) of HRV. Logistic mixed models described the relationship between the explanatory variables and the risk of upper respiratory tract and pulmonary infections, muscular affections, and all-type pathologies. Results: The risk of all-type pathologies was higher for national swimmers and in winter (P < 0.01). An increase in the parasympathetic indexes (HF, SD1) in the supine position assessed 1 wk earlier was linked to a higher risk of upper respiratory tract and pulmonary infections (P < 0.05) and to a higher risk of muscular affections (increase in HF, P < 0.05). Multivariate analyses showed (1) a higher all-type pathologies risk in winter and for an increase in the total power of HRV associated with a decline SD1 in supine position, (2) a higher all-type pathologies risk in winter associated with a decline in HF assessed 1 wk earlier in orthostatic position, and (3) a higher risk of muscular affections in winter associated with a decrease SD1 and an increase LF in orthostatic position. Conclusions: Swimmers' health maintenance requires particular attention when autonomic balance shows a sudden increase in parasympathetic indices in the supine position assessed 1 wk earlier evolving toward sympathetic predominance in supine and orthostatic positions....

The aim of this study was to carry out a statistical analysis of the Banister model to verify how useful it is in monitoring the training programmes of elite swimmers. The accuracy, the ill-conditioning and the stability of this model were thus investigated. The training loads of nine elite swimmers, measured over one season, were related to performances with the Banister model. First, to assess accuracy, the 95% bootstrap confidence interval (95% CI) of parameter estimates and modelled performances were calculated. Second, to study ill-conditioning, the correlation matrix of parameter estimates was computed. Finally, to analyse stability, iterative computation was performed with the same data but minus one performance, chosen at random. Performances were related to training loads for all participants (R ² = 0.79 ± 0.13, P < 0.05) and the estimation procedure seemed to be stable. Nevertheless, the range of 95% CI values of the most useful parameters for monitoring training was wide: τ _a = 38 (17, 59), τ _f = 19 (6, 32), t_n = 19 (7, 35), t_g = 43 (25, 61). Furthermore, some parameters were highly correlated, making their interpretation worthless. We suggest possible ways to deal with these problems and review alternative methods to model the training – performance relationships.

The aim of this study was to model the residual effects of training on the swimming performance and to compare a model that includes threshold saturation (MM) with the Banister model (BM). Seven Olympic swimmers were studied over a period of 4 +/- 2 years. For 3 training loads (low-intensity w(LIT), high-intensity w(HIT), and strength training w(ST)), 3 residual training effects were determined: short-term (STE) during the taper phase (i.e., 3 weeks before the performance [weeks 0, 1, and 2]), intermediate-term (ITE) during the intensity phase (weeks 3, 4, and 5), and long-term (LTE) during the volume phase (weeks 6, 7, and 8). ITE and LTE were positive for w(HIT) and w(LIT), respectively (p < 0.05). Low-intensity training load during taper was related to performances by a parabolic relationship (p < 0.05). Different quality measures indicated that MM compares favorably with BM. Identifying individual training thresholds may help individualize the distribution of training loads.