## Similar protocols

## Protocol publication

[…] The data were tested for normal distribution using histograms and the Kolmogorov-Smirnoff test or Shapiro-Wilk test as appropriate. The categorical and continuous data are presented as percentages and mean ± SD values (or medians and interquartile ranges (IQR), respectively. The categorical variables were compared using a χ2 test or Fisher’s exact test as appropriate. The quantitative continuous variables were compared using an unpaired t-test and the Mann–Whitney U test for normally and non-normally distributed variables, respectively.To analyze the effect of the campaign, we planned a quasi-experimental design using interrupted time series (ITS) to control for secular trends [-]. The ITS design is the strongest quasi-experimental design for evaluating the effects of time-delimited interventions. The data were collected at multiple instances over time, before and after an intervention, to detect whether the intervention had a significantly greater effect than the underlying secular trend. An advantage of the ITS design is that it can adjust for time trends, estimate changes in time trends and account for autocorrelation. Furthermore, it allows the investigation of potential biases together with the secular trend, such as the duration of the intervention (that is, that the intervention might have an effect for the first 2 months only, rather than a sustained effect) and seasonal or cyclical effects.To model the effect of the campaign, we run autoregressive integrated moving average (ARIMA) models. Two parameters define each segment (before and after the intervention) of a time series: level and trend. The level is the value of the series at the beginning of a given time interval. The trend is the rate of change of a measure (slope) during a segment. To examine the results, we might analyze whether there are changes in level and trend that follow an intervention. In general, a change in level constitutes an immediate intervention effect, and a change in trend represents a gradual variation in the outcome.Multiple time series require different approaches, which consist of assessing not only the absolute value but also the shape of each series. To enhance our analysis while correctly addressing our multicenter data, we performed two approaches for our main outcomes (length of MV and midazolam consumption). First, patient data were aggregated by month. In one approach, we retrieved an average value per month representing the network, which was an arithmetic mean value per unit per month. In another approach, we fit a hierarchical time series [] by using a bottom-up method. This approach involves first providing independent time series at the bottom level of the hierarchy (each ICU, level 1) and then aggregating the independent time series upward to produce a revised time series for the whole hierarchy (network, level 0). This method accounts for noise and variability between time series and provides additional information in comparison to a crude average.To deal with variation in case mixes over time at two levels (patient and unit), we conducted a sensitivity analysis for our main outcome fitting a generalized linear mixed model, achieving the better fit for correlated responses, in particular for the analysis of our longitudinal and clustered data. The length of MV for each patient was the dependent variable. For the first level (patient level), we adjusted for (Simplified Acute Physiology Score III (SAPS III), Charlson index score and vasoactive drugs. For the second level (unit level), we fit random intercepts for each one of ten units and nested into the variable units random components for type of admission and reasons for admitting: sepsis syndrome, cardiac surgery and respiratory conditions. The model was built with a Poisson distribution with a log link function, and the covariance structure for the random effects was first-order autoregressive [,].We evaluated the association between midazolam consumption and length of MV fitting a linear mixed model. The monthly consumption of midazolam adjusted per number of mechanically ventilated patients per each ICU was the dependent variable. The length of MV per month per each ICU was the outcome evaluated. In the model, both variables were transformed with the use of their natural logarithms to reduce the influence of extreme outliers. To adjust for severity, we entered SAPS III score as a covariate. We used splines to allow nonlinear associations between midazolam and length of MV [,].P <0.05 was considered to be statistically significant for all of the analyses. The R free source statistical package version 2.15.2 ([]; **The** R Project for Statistical Computing, Vienna Austria), CRAN ([])-specific libraries and the **SPSS** 19.0 package for Windows software (IBM SPSS, Chicago, IL, USA) were used in all of the analyses.Further information on the statistical analysis plan is available in Additional file . […]

## Pipeline specifications

Software tools | The R Project for Statistical Computing, SPSS |
---|---|

Application | Miscellaneous |

Organisms | Homo sapiens |