Department of International Health, Johns Hopkins University Bloomberg School of Public Health Baltimore, Maryland e-mail: email@example.com
Department of Ophthalmology Wilmer Eye Institute Johns Hopkins University School of Medicine Baltimore, Maryland
We read with interest the study by Chalmers et al.1 This letter addresses some concerns with study design and data analyses. It also asks specific questions with the hope of better interpreting the study findings in light of whether it actually answers the question of whether there was an association between lens care product (LCP) use and corneal infiltrative events (CIEs).
The chief clinical issue and likely the underlying principal aim of the research are to identify risk factors for acute symptomatic corneal infiltrates in the setting of contact lens wear. Therefore, it is unclear why the authors included microbial keratitis, contact lens acute red eye, solution hypersensitivity, superior epithelial arcuate lesion, viral keratoconjunctivitis, contact lens peripheral ulcer, and asymptomatic infectious keratitis (IK) in the case definition of CIE. If only symptomatic IK patients were included as cases and compared against their matched controls, would the estimates of the risk factor associations be similar to what was found when these other conditions were included? By diluting the case definition with events unlikely to be related to the type of contact lens solution, it is more likely that a true association would be missed. In addition, what would the power of the study be to detect potentially important associations? With this more restricted definition, the number of cases would now be reduced from 166 to 74. Furthermore, because daily wear users are the primary subjects exposed to disinfecting solutions, it makes sense, as the authors do in Table 3, to limit the analysis to this group alone. What would the association between Opti-Free RepleniSH use and this more limited definition of CIEs be when restricted to daily wearers only? This is certainly the clinical question of interest, which does not seem to be adequately addressed in the Chalmers analysis and presentation.
There were 44 (27%) cases and 114 (23%) controls where Opti-Free was the stated LCP, but it was unclear if the LCP was RepleniSH or Express. The percentage of missing data is even larger if one restricts the calculations only to those using Opti-Free because not all cases and controls used Opti-Free. The authors state that these cases and controls were assigned to RepleniSH or Express at random based on the market share data within the three different years covering the study period (2007 to 2009). This seems like a reasonable approach to take for the controls, but if there was a real difference between the use of these two products among cases and controls, assigning both groups to the market share at random would lead to a lower estimated odds ratios than truly exists. In the statistical methods, the authors state that “cases were allocated separately from the controls to maintain balance in the samples.” Could the authors state more clearly how the imputation was done separately for the cases and controls? What was the odds ratio for an analysis that used only cases and controls for which the type of Opti-Free solution was known? How does it compare with the odds ratio of 1.63 obtained with the imputed values? This would be helpful in understanding whether some bias was introduced into the estimation of risk by the imputation of missing Opti-Free type.
In Table 3, the univariate association between all CIEs and Opti-Free RepleniSH is statistically and clinically significant (odds ratio, 1.63; 95% confidence interval, 1.11 to 2.39). The authors state that “increased risk in univariate analysis for LCP was not significant after adjustment for other factors.” What was the nonsignificant adjusted odds ratio for RepleniSH (which could be of clinical significance) in the multivariate model and which specific confounders attenuated this association? And although the power is almost certainly insufficient, what would the estimated odds ratio be if cases were limited to users of daily wear contact lenses who experienced acute symptomatic infiltrates (i.e., just the IK cases)?
Finally, the authors conclude that “our results did not find a significantly increased risk of CIEs with the use of Opti-Free RepleniSH by itself or in combination with Acuvue OASYS lenses…” Given that they acknowledge that “the sample size was not sufficient to detect an increased risk less than 80%,” the conclusion should more reasonably be that this study did not have the statistical power to detect a clinically meaningful increase in risk associated with Opti-Free RepleniSH. A true increased risk of CIEs of even 50% would be of clinical and public health importance. The sample size required to detect a 50% increased odds of having a CIE when using Opti-Free RepleniSH compared with all other LCP use with power of 80% and type I error of 5% is 263 cases and 789 controls (three controls per case).
In summary, the clinical heterogeneity of the cases and lack of detail and clarity in the presentation of the study results render this a difficult article to interpret. However, these issues are dwarfed by our concerns regarding adequate sample size (i.e., power) of consistently defined cases that would allow any definitive conclusions to be reached.
Joanne Katz, ScD
Department of International Health
Johns Hopkins University Bloomberg
School of Public Health
Oliver Schein, MD, MPH
Department of Ophthalmology
Wilmer Eye Institute, Johns Hopkins
University School of Medicine
Joanne Katz was a paid consultant to Bausch and Lomb to review the Chalmers article. Oliver Schein is a consultant to Bausch and Lomb and Abbott Medical Optics.
1. Chalmers RL, Keay L, McNally J, Kern J. Multicenter case-control study of the role of lens materials and care products on the development of corneal infiltrates. Optom Vis Sci 2012; 89: 316–25.