As our study results show, we were able to obtain and validate a predictive score for TB treatment adherence for a large cohort of new TB cases undergoing standard TB treatment. The score could be easy for a clinician to calculate situation risk of lost to follow-up at the time of diagnosis because the evaluation of it is dependent on only clinical and epidemiological characteristics, not on results of complimentary tests. The results of the statistical model indicate this predicting system is valid to evaluate the probability of lost to follow-up outcome before treatment is even initiated, as shown by the positive relationship between rate of lost to follow-up and point score.
Other components of the Stop TB Strategy are contributing to health system strengthening and engaging all care providers  but the implementation of DOT is not possible in many countries and is not the rule in developed countries. This TB score is a useful to determine the proportion of patients for who DOT should be considered, depending of available resources of the TB program.
The independent predictors factors of lost to follow-up outcome such as living situation or previous TB treatment were also consistent to those of other studies [11–13]. Other predictive models have been created in recent years have to estimate the risk of certain diseases [14–17], including some for TB (on quality of respiratory health , preventative isolation , severity pulmonary TB by means of chest x-ray , clinical course  and risk of multi-drug resistance ), but to our knowledge none have been created for the risk of lost to follow-up outcome.
Our score takes into account some characteristics, such as country of origin. Some studies have shown that the immigrant patient population does not meet the treatment adherence objectives described by the WHO, but the native patient population does . Furthermore, the percentage of annual TB cases among immigrant populations of various countries is higher [24–27]. Thus it is important a predictive score that provides the probability of lost to follow-up outcome at the time of diagnosis considers characteristics such as country of residence, living alone or in an institution, history of TB or IDU, or poor understanding. It should be noted that poor patient understanding refers to not only a language barrier, but also difficulty in understanding treatment instructions which can occur in native and immigrant population .
According to our model, the probability of lost to follow-up outcome corresponds with a significantly elevated risk of all point categories and also increases with the total points acquired by each patient. Our analyses that examined concordance between the predicted and observed values yielded similar results and the CI of the ROC curves did not detect any significant differences. This confirms the reliability and discrimination capacity of our model.
Treatment adherence is directly related to many factors, such as gender, poverty, economic difficulty, social context, healthcare services, personal interpretation of the disease , drug addiction, country of origin , history of TB , alcoholism, homelessness and HIV infection . The score of our model allows a clinician to determine which TB patients have a higher risk of lost to follow-up by a simple score ≥2. The clinician can then decide what measure should be taken to improve treatment adherence , such as scheduling additional clinical visits, providing more information about TB, providing family or social support, reducing drug prescription costs, hasten future clinical visits, improving communication between involved healthcare professionals, collaborating with public health personnel or community health workers  and finally accurately allocating DOT resources to the patients who need it the most.
Systematic DOT is a proven effective intervention to achieve treatment adherence . For example, treatment compliance rates have reached over 95% in subgroups for which DOT was made a priority . As consequence of that DOT is more costly than self-administered treatment , all TB control programs must use efficiently their resources.
The score estimated in our study reflects the study setting, with a high quality and uniform healthcare system throughout the country. The score calculated in other settings may vary according to different patient characteristics and healthcare systems. In the setting of limited healthcare distribution, the score can represent the side of the medal dedicated to healthcare and can indicate where improvements should be made. Current studies such as this should be reviewed by control programs to facilitate the use of predicting factors of treatment non-compliance to identify high risk subgroups. A predictive score can be extremely helpful to best direct DOT interventions.
One study limitation is that patients with known drug resistance and those with a contraindication to start standard treatment of three or four drugs were excluded because the first study cohort (derivation cohort) was also used to study treatment completion among TB patients . Patients with known drug resistance were excluded because they required longer treatment, had poor compliance and represented few cases, which could have distorted the results.
Consequently, the second cohort (validation cohort) also followed the same exclusion criteria. Other limitation is the use of terciles of risk for the Hosmer-Lemeshow test instead of deciles of risk because of an insufficiently large sample size. Nonetheless, we still consider the model valid because the observed and expected values were similar.