Reply to Rezahosseini.

To the Editor—We thank Dr Rezahosseini for his valuable comments [1] on our study. We agree that multiple measures must be undertaken to increase the robustness of results in observational studies and that the results need to be interpreted carefully.

The combination of diagnoses used to define acute liver injury in our study has been previously evaluated in different settings with reported positive predictive values ranging from 75% to 95% [2, 3]. Furthermore, we excluded patients who were hospitalized for any other diagnoses, which in this case would include patients admitted for sepsis. We do agree with the author’s comment that the fluoroquinolones are more often used in gram-negative infections as compared to amoxicillin and that the lack of indication of treatment is a weakness, as described in the article. However, since we excluded any treatment episodes that were preceded by hospitalization in the past 2 months, we find it unlikely that the choice of antibiotic in an outpatient setting is based on a specific pathogen (ie, gram negative or gram positive) rather than a presumed site of infection for which there are overlapping areas between the compared antibiotics. Furthermore, biochemical markers of liver insults, such as increased values of aminotransferases, bilirubin, and international normalized ratio, are sometimes seen in septic patients (without regard of causative organism). However, in a large epidemiologic study from 2017 looking at patients with severe sepsis according to Sepsis-2, only 2151 of 197724 (1.1%) patients had a hepatic Sequential Organ Failure Assessment score ≥2, making it a rare occurrence [4].