Volume of blood cultured and APACHE II score: an intriguing diptych for diagnosis of bloodstream infections?

In a recent issue of the Journal of Clinical Microbiology, Bouza et al. provided interesting insight into the putative effect that the volume of blood cultured per patient had upon the rate of detection of bloodstream infections (BSI) (1). According to the authors, in the total of the sample, the volume of blood cultured was not an independent predictor capable of affecting blood culture results. On the other hand, the acute physiology and chronic health evaluation II (APACHE II) score was associated with the obtention of a low-volume specimen. Subsequently, the authors conducted a subanalysis of the cases with APACHE scores higher than 18, which concluded that “higher volumes were associated with better yields of detection of BSI.” This intriguing observation may be worth explaining and investigating further. Concerning the subanalysis, an interesting question arises: within the APACHE 18-score subgroup, has the APACHE score per se been retained in the multivariable model? This has not been explained in Materials and Methods and seems of particular interest, as the APACHE score may not lose its predictive power in the aggregated APACHE 18-score subgroup; for instance, a patient with an APACHE score near 30 may represent a completely different entity than a patient with an APACHE score near 18. In other words, the division-stratification of the sample into 18 and 18 subgroups may not be sufficient for the omission of the APACHE score as putative predictor. In case a putatively significant predictor is not included in the subanalysis model, other less significant factors and confounders may become significant. If the APACHE score has not been included in the subanalysis, would its introduction into the model lead to the disappearance of the effect mediated by the volume of blood? On the other hand, if the APACHE score has been included in the model but has not been reported for reasons of brevity, has it retained its predictive power (as in the total of the sample)? In this scenario, reporting the odds ratio (OR) of the APACHE score in the subanalysis may be desirable. In any case, in regard to the total of the sample, a simple maneuver might have been invaluable, i.e., the introduction of the statistical interaction of the APACHE score multiplied by the volume of blood cultured in the full model. If this interaction proved to be statistically significant, it would be a reliable proof that the APACHE score modifies the effect of the volume of blood upon BSI detection. It remains for the authors to proceed to such further and detailed analyses in the future.