TY - JOUR
AU - Dreyfus, Jill
AB - We thank Ohneberg et al. for their letter (1) on our recently published article (2), which brings up some interesting points for discussion. The authors first point out that our study was not the first application of risk-set matching to hospital-acquired infections, and they highlight some previous applications of the method (1). They then comment on our analysis following risk-set matching. Their argument is that even though we accounted for time of infection with risk-set matching, by not using the time dimension explicitly in the statistical analysis, we somehow did not adequately incorporate time of infection into our estimation strategy (1). This argument seems odd to us. For example, consider a paired design that matches exposed units to unexposed controls exactly on some covariate, say age. Ohneberg et al.’s argument suggests that the average of the paired differences does not adequately incorporate age because age was not explicitly used in the model. However, age was used in the matched-pairs design and thus was adequately incorporated into the estimation strategy, as we did in our study using the day of an inpatient stay on which a laboratory-positive bloodstream culture was drawn (2). We acknowledge that our analysis was straightforward and that there are still challenges involved in estimation following risk-set matching—for example, how to handle a matched, at-risk patient who goes on to become positive for a bloodstream infection. However, we stand by the main point of our approach, that for a patient infected at some point during their hospital encounter (e.g., on day 5), a reasonable comparator is a patient who is still at risk of hospital-acquired infection (e.g., still an inpatient on day 5) and has comparable risk factors for infection. On more specific points, we think the approach of competing risks within a multistate model is a reasonable approach to differentiating between discharge and death. Another approach is the principal stratification framework (3, 4), which can account for censoring or truncation due to death (e.g., if infected patients have shorter hospital stays because they die). Regarding estimation of costs, we used the cost of the entire encounter. Our concern with left-truncating costs at the time of infection is that this method might exclude costs that accumulate before infection is diagnosed (e.g., costs due to culture collection, empiric therapy, etc.), which in some cases may be substantial. The authors’ last point is that the full cohort should be used rather than reducing the cohort via matching. Indeed, one drawback of matching is that it reduces the size of the sample, which can limit statistical power. However, the benefits are that we can ensure that the comparison groups are comparable and that we are making inferences for the relevant population. In our analysis, the sample size was sufficiently large to allow us to utilize this approach. Methods like subclassification and full matching are other options that do not exclude as many patients (5), but to our knowledge they have not been applied to risk-set approaches. Acknowledgments This work was supported by the Internal Research Grant Program at Children’s Hospitals and Clinics of Minnesota (grant 47601). J.D. is an employee and shareholder of Premier, Inc. (Charlotte, North Carolina). The other authors declare no conflicts of interest. References 1 Ohneberg K , Wolkewitz M , Schumacher M . Re: “Risk-set matching to assess the impact of hospital-acquired bloodstream infections” [letter] . Am J Epidemiol . 2019 ; 188 ( 6 ): 1192 – 1193 . 2 Watson D , Spaulding AB , Dreyfus J . Risk-set matching to assess the impact of hospital-acquired bloodstream infections . Am J Epidemiol . 2019 ; 188 ( 2 ): 461 – 466 . Google Scholar Crossref Search ADS PubMed 3 Frangakis CE , Rubin DB . Principal stratification in causal inference . Biometrics . 2002 ; 58 ( 1 ): 21 – 29 . Google Scholar Crossref Search ADS PubMed 4 Rubin DB . Causal inference through potential outcomes and principal stratification: application to studies with “censoring” due to death . Stat Sci . 2006 ; 21 ( 3 ): 299 – 309 . Google Scholar Crossref Search ADS 5 Stuart EA . Matching methods for causal inference: a review and a look forward . Stat Sci . 2010 ; 25 ( 1 ): 1 – 21 . Google Scholar Crossref Search ADS PubMed © The Author(s) 2019. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
TI - THE AUTHORS REPLY
JF - American Journal of Epidemiology
DO - 10.1093/aje/kwz036
DA - 2019-06-01
UR - https://www.deepdyve.com/lp/oxford-university-press/the-authors-reply-6lwfm0PDhq
SP - 1192
EP - 1193
VL - 188
IS - 6
DP - DeepDyve
ER -