Paired-data analysis is required in the study by May et al.

We read with great interest the article by May et al. who evaluated the effectiveness of argon plasma coagulation (APC) for small-bowel bleeding. We would like to highlight some issues regarding the statistical analysis used in the study. The authors have used a t test to compare continuous variables (e. g. hemoglobin) and Fisher’s (exact) test for categorical variables (e. g. blood transfusion). For the before and after comparison of the APC group these tests are not appropriate because these are not independent observations. Therefore, a paired t test would be appropriate to compare hemoglobin (Table 3 of the manuscript); if the distribution of hemoglobin is skewed then a non-parametric test should be used. Similarly, in Table 4 of the manuscript comparing need for blood transfusion, a McNemar’s test should have been used. Furthermore, only one P value has been reported in the same row as the “Before APC” group whereas the comparison is between before and after APC groups; it is not clear whether this P value is for the Osler or the Non-Osler group or whether or not it included all of the patients. Considering the large differences reported in the study, we think that the results of significance tests would not change considerably. However, applying correct statistical analysis is crucial to any research study in order to avoid the danger of incorrect interpretation based on statistical tests.

After undergoing APC, patients were followed up to assess the need for blood transfusion, and the rates of rebleeding and mortality. The average follow-up period was 55 months (range 42 – 72 months). It is possible that an outcome was not observed in a patient with short follow-up duration simply because the duration of follow-up was not long enough to experience any outcome. If the patient had been observed for a longer period, it is possible that the outcome of interest may have been observed. Furthermore, in clinical care/research studies, participants/patients are routinely lost to follow-up and no inference regarding their outcome can be made (censoring). However, the authors have not mentioned whether any patients were lost to follow-up in their study. With a considerably variable duration of follow-up and attrition of the original cohort, calculation of cumulative incidence (risk), and therefore, the risk ratio would not be appropriate. In such follow-up studies, time-to-event (survival) analysis would be more appropriate.

References

• 1 May A, Friesing-Sosnik T, Manner H et al. Long-term outcome after argon plasma coagulation of small-bowel lesions using double-balloon enteroscopy in patients with mid-gastrointestinal bleeding.  Endoscopy. 2011;  9 759-765