Re: Lung protective mechanical ventilation and two year survival in patients with acute lung injury: prospective cohort study
Needham and colleagues present much needed data regarding long term hard outcomes in this day and age of acute lung injury (ALI) management (1) which reflects ‘effectiveness’ rather than ‘efficacy’ (2) of lung protective mechanical ventilation. Kudos to them on a well-designed observational (over a 2-year period) prospective multi-center cohort study.
As eloquently described by Needham and colleagues in their ‘limitations’ section (1), potential biases are inherent in any observational study. It is interesting that 10% of each group (’adherent’ and ‘non-adherent’ groups) had an arterial PH of < 7.25. This needs to be carefully analyzed in the setting of the study’s overall results. Presenting these data as % with PH < 7.25 is appropriate and makes clinical sense, and the fact that this cut-point was decided a priori greatly adds to the quality of the paper. However now that the results are as above, taking it one step further to try to differentiate the 2 groups by highlighting any potential confounders, a continuous variable for the arterial PH for each group would add a lot to the study’s analyses. The question becomes: Are the % of ‘adequate’ PH in the 2 study groups comparable because the clinicians increased the tidal volume to correct for severe acidosis before the time of variables measurement (6am and 6pm in this study)? Or is it lack of clinicians’ knowledge / training or lack of protocolization of ALI management? One measure that may shed some light on this question is presenting the arterial PH data as a continuous variable for each group (acknowledging that this was not decided a priori) for a more powerful comparison.
It is also typically difficult to guesstimate the intra-class correlation (ICC) in general since ICC’s are rarely reported in the literature. It would be greatly appreciated if the investigators could provide some guidance / input on how they estimated and then incorporated / adjusted for intra-class (i.e., intra-patient) correlation. This is especially relevant when comparing the ‘alive’ and the ‘dead’ groups (e.g., table 2 shows data on 4,157 ‘dead’ observations in the setting of a total of 485 total patients and 311 dead patients). Such intra-patient correlation also applies when analyzing the other variables in table 2, including the arterial PH and the respiratory rate.
This study (1) highlights the huge challenge to transform a rapidly changing medical management within an intensive care unit [based on a multitude of data input throughout the ICU stay (e.g., rapidly responding to severe hypoxemia, acidosis, and / or hypotension)] into a consistent a priori - decided numerical variable that could make mathematical sense to be included in a regression model. Making the balance between the integrity of the mathematical model and study design, and reflecting the patient’s continuously changing medical condition is a fine but occasionally blurry line. The study by Needham and colleagues (1) is a good example of getting as close as practically possible to such a balance.
1 Needham DM, Colantuoni E, Mendez-Tellez PA, Dinglas VD, Sevransky JE, Dennison Himmelfarb CR, et al. Lung protective mechanical ventilation and two year survival in patients with acute lung injury: prospective cohort study BMJ 2012;344:e2124.
2 Lauer MS, Collins FS. Using science to improve the nation's health system: NIH's commitment to comparative effectiveness research. JAMA 2010;303:2182-3.
Competing interests: No competing interests