Design and statistical analysis reporting among interrupted time series studies in drug utilization research: a cross-sectional survey

Table 5 Additional methodological considerations (parameters setting, individual-level covariates and hierarchical data structure) for ITS studies

Characteristics	n	%
Incorrect interpretation of level change due to time parameterization ^a
Reported the regression model and interpreted the coefficients (N = 153)
Yes	60	39.2
Where did the author report the regression model and the interpretation of coefficients? (N = 60)
In article	47	78.3
In supplementary material	13	21.7
The interpretation of level change due to time parameterization was incorrect ^b
Yes	15	-
Individual-level characteristics
Has individual-level data (N = 153)
Yes	127	83.0
Consider individual-level characteristics (N = 127)
Yes	24	18.9
How to control individual-level characteristics (N = 24) ^c
Add covariates	21	87.5
Stratified Analysis	7	29.2
Other	4	16.7
Hierarchical data structure
Data structure for ITS analysis (N = 149)^d
Hierarchical data (more than one level) ^e	97	65.1
Whether the author handled hierarchical data (N = 97)
Yes	23	23.7
Methods for handling hierarchical data (N = 23)
Stratified by sites	13	56.5
Mixed effect model	6	26.1
Generalized estimating equation	2	8.7
Fixed-effect model	1	4.3
Two-stage analysis	1	4.3
Considered cluster effects in which level? (N = 23)
Hospital/clinic/other healthcare provider	15	65.2
Province/State/Region	4	17.4
Nation	2	8.7
Unclear	2	8.7
Whether the author reported the differences across sites (N = 23)
Yes	16	69.6
If yes, how to present the differences across site? (N = 16)
Figure	7	43.8
Both table and figure	6	37.5
Table	3	18.8

^aIf the researchers set an ITS model with both level change and slope change, and used the product between their calendar time variable and the indicator variable indicating pre- versus post-intervention time periods to represent the post-intervention linear segment, then the interpretation was wrong (More details in Appendix 1)
^bFor this item, we did not calculate the proportion as the denominator is difficult to define. We believe that using either 60 (the number of studies reporting regression models) or 139 (the number of models including level change and slope change) as the denominator would be inappropriate
^cSome studies used more than one method to control individual-level characteristics
^dThis part only included ITS studies with aggregated analysis units (n = 149) because the mishandling of data hierarchy only takes place in the ITS study with aggregated analysis unit
^eFor the studies that contained individual-level data, we calculated how many levels are there in the dataset excluded individual data (which cannot be repeated measured). For example, the raw data was a three-level hierarchy of patient, hospital and region and the repeated measured level were hospital and region. We defined this dataset as a two-level hierarchical data for ITS analysis. For the studies that only contained aggregated data, we calculated how many levels are there in the dataset directly

ISSN: 1471-2288