This study was based on a well-established prospective cohort and reflected the effect of routine monitoring in a real-world population with geographically and economically diverse areas in China. However, a major methodological problem in this observational study is that investigators have no control over the intervention assignment, which is likely to result in confounding and biased estimation of intervention effects [19
]. To account for these differences in observed covariates, propensity scores were applied in this analysis.
Among several propensity score methods, SMRW and matching estimated most closely approached results observed in the clinical trials [23
]. In comparison with matching, SMRW has more advantages [24
]: 1) data from all patients are used so that variance is more closely to the study population; 2) process for SMRW is much easier than propensity score matching. Based on all above, we used SMRW to balance all available covariates and estimate the effect of routine monitoring in this study.
In the present study, 1/3 of the patients with liver dysfunction did not have any clinical symptoms, including 8 with severe hepatotoxicity, which indicated the important role of routine monitoring in identifying asymptomatic liver injury. Liver injury could be fatal if it was not recognized in time [12
], thus routine monitoring could be helpful to detect liver damage early so as to apply appropriate interventions in time and improve the prognosis. Our results showed that there was no death and only 2 patients (1.8%) in hospital in scheduled monitoring group. 35.1% changed their anti-TB treatment. In contrast, in passive detection group 11.1% required hospitalization and 2 patients (1.3%) died. More than half of the patients changed their anti-TB treatment. Therefore, routine liver function monitoring was associated with less hospitalization, better prognosis and better compliance of anti-TB treatment in this study. A study by Agal et al. [25
] showed that 21 (10.5%) of 200
TB patients who accepted regular liver function monitoring (every week for the first month, then fortnightly for the next 2
months and then monthly until the end of therapy) developed ATLI and no one died, whereas 16.6% of the patients without periodic monitoring died and 75% developed icteric hepatitis. In a study by Mcneill et al. [26
], the rate of severe ATLI in patients receiving RZ prophylaxis for latent tuberculosis infection (LTBI) reduced from 5% to 0% with periodic monitoring. Another study [27
] with H prophylaxis for LTBI indicated that no severe ATLI developed on routine monitoring.
Biochemical abnormalities generally occur before clinical symptoms or signs of liver injury develop, thus monitoring of liver function can detect liver injury ahead of the symptomatic period, prevent serious ATLI and avoid incompliance of anti-TB treatment [12
]. In our study, the median liver dysfunction finding time of passive detection group was 39.0
days, which indicated that monitoring in the first month after the initiation of anti-TB treatment was important. Subgroup analysis for patients with liver dysfunction finding time less than 1
month showed that significantly more patients required hospitalization in passive detection group. More patients in passive detection group changed their anti-TB treatment and were likely to have a poor prognosis of hepatotoxicity, although there was no significant statistically difference between the two groups.
Some limitations need to be noted. First, the ADACS study was not specially designed to evaluate the effect of regular liver function monitoring, only two routine tests (baseline and within 2
months after initiation of anti-TB treatment respectively) were conducted, thus it may limit the ability of detecting liver dysfunction earlier. Secondly, although we controlled a large set of factors that potentially differed between the groups at baseline using propensity score method, there are still some unknown confounders which may not be balanced. Propensity scores are not magic bullets capable of eliminating all the bias of observational studies [28
]. Besides, the study did not collect patients’ information on HIV infection and alcohol consumption which are important risk factors of hepatotoxicity. Finally, owing to the only two routine tests in our study, it is still not clear what frequency or intervals between tests may be optimal. And further studies on compliance and cost effectiveness of monitoring are needed.