Medicina Intensiva  (English Edition) Medicina Intensiva  (English Edition)
Med Intensiva 2018;42:134-5 - Vol. 42 Num.2 DOI: 10.1016/j.medine.2018.01.004
Letter to the Editor
Comparative statistical analysis in observational epidemiological studies
Análisis estadísticos comparativos en los estudios epidemiológicos observacionales
A. Bosch-Gaya, L. Matute-Blanco, D. Fernández-Rodríguez, , F. Worner
Servicio de Cardiología, Hospital Universitari Arnau de Vilanova, Lérida, Spain
Dear Editor,

We wish to congratulate Socias et al.1 for their work on the prognostic impact of ST segment elevation acute coronary syndrome (STE-ACS) in Mallorca, Spain.

Subpopulations with STE-ACS from two observational registries were compared in different time frames. In the IBERICA-Mallorca registry, fibrinolysis was the main reperfusion therapy, while the Código infarto-Illes Balears (CI-IB) registry implemented one healthcare network for the management of STE-ACS based on primary angioplasty (PA). From both registries, only 49.7 and 58.8 per cent of the populations were included for the analysis. Features from both samples varied significantly in aspects such as history; clinical profile; and effective medical therapy in the management of STE-ACS (betablockers and ACE).2

The main conclusion from the study1 is that the implementation of one PA network reduced mortality rate. However, the statistical analysis conducted deserves certain considerations:

  • (a)

    Survival analysis: in observational studies where the main goal is to assess the time elapsed until the occurrence of one event, survival analyses3,4 are normally used. The most widely used statistical tool to know the effect of one independent variable in time is the KaplanMeier method. However, it is not good for the simultaneous assessment of more than one independent variable, or to estimate the effect size on the risk of occurrence of one event.3 In a setting where different variables coexist that may have an impact on the main goal, the survival analysis using the Cox regression model is key to assess to what extent multiple variables can modify the risk of occurrence of a single event.4 This study showed that the mortality rate at 28 days varied after the implementation of this network (IBERICA-Mallorca=12.2 vs CI-IB=7.2 per cent; HR=0.560; 95 per cent CI: 0.360–0.872; p=0.010) based on the KaplanMeier method. However, when the Cox regression analysis was conducted, no statistical significance was achieved in either model.

  • (b)

    Propensity score matching: one option for the inter-group comparison of one outcome variable is observational studies is the Propensity score matching.5 This statistical analysis assesses the effect that one treatment strategy has based on the covariables that predict that one patient will be assigned to the assessed treatment, and not the control group. Thus, by pairing patients based on the estimated propensity scores, we can make inter-group comparisons with an approximate balance on the relevant variables, instead of just one simple comparison between patients who received the treatment strategy, and those who did not.

In sum, even though the implementation of one PA network for the management of STE-ACS improved prognosis,2 the KaplanMeier analysis was not powerful enough to confirm that the reported reduced mortality rate is exclusively due to the effect of such healthcare network. This is why we believe that conducting one Cox regression analysis to assess not only the baseline differences and type of reperfusion, but also different therapies received; or one propensity score matching analysis to balance all relevant covariables will make better evaluations of the real impact of this network on the prognosis of this population.

Conflicts of interest

The authors declare no conflict of interests associated with this article whatsoever.

References
1
L. Socias,G. Frontera,C. Rubert,A. Carrillo,V. Peral,A. Rodriguez
Comparative analysis between 2 periods of acute myocardial infarction after a decade in Mallorca. IBERIA Study (1996–1998) and Infarction-Code (2008–2010)
Med Intensiva, 40 (2016), pp. 541-549 http://dx.doi.org/10.1016/j.medin.2016.04.001
2
P.G. Steg,S.K. James,D. Atar,L.P. Badano,C. Blömstrom-Lundqvist,M.A. Borger
ESC Guidelines for the management of acute myocardial infarction in patients presenting with ST-segment elevation
Eur Heart J, 33 (2012), pp. 2569-2619 http://dx.doi.org/10.1093/eurheartj/ehs215
3
E.L. Kaplan,P. Meier
Non-parametric estimation from incomplete observations
J Am Stat Assoc, 53 (1958), pp. 457-481
4
D.R. Cox
Regression models and life-tables
J R Stat Soc, 34 (1972), pp. 187-220
5
P.C. Austin
An introduction to propensity score methods for reducing the effects of confounding in observational studies
Multivariate Behav Res, 46 (2011), pp. 399-424 http://dx.doi.org/10.1080/00273171.2011.568786

Please cite this article as: Bosch-Gaya A, Matute-Blanco L, Fernández-Rodríguez D, Worner F. Análisis estadísticos comparativos en los estudios epidemiológicos observacionales. Med Intensiva. 2018;42:134–135.

Corresponding author. (D. Fernández-Rodríguez d.fernan.2@hotmail.com)
Copyright © 2017. Elsevier España, S.L.U. and SEMICYUC
Med Intensiva 2018;42:134-5 - Vol. 42 Num.2 DOI: 10.1016/j.medine.2018.01.004