Introduction

Nontuberculous mycobacteria (NTM) are environmental microorganisms that cause chronic refractory respiratory infections. The global incidence of NTM pulmonary disease (NTM-PD) has been increasing with a concomitant increase in mortality rates1,2,3,4. The causative species of NTM-PD exhibit regional variation; however, Mycobacterium avium–intracellulare complex (MAC) accounts for 90% of NTM-PD cases in Japan5. The clinical course of MAC pulmonary disease (MAC-PD) is significantly heterogeneous. Some patients remain stable with watchful waiting alone, whereas others develop bronchiectasis or cavitation and succumb to respiratory failure despite multidrug therapy. Owing to the relatively high incidence of adverse events associated with multidrug therapy, treatment is not universally recommended at diagnosis6. The adherence rate to guideline-based treatment is low (4.3–41.9%), likely due to frequent adverse effects7. Consequently, prognostic evaluation at the time of diagnosis is crucial for determining disease severity. Nevertheless, determining which patients to treat and when remains challenging for clinicians8.

Systematic reviews have reported 5–42% 5-year mortality rates for MAC-PD9, identifying various risk factors, including male sex, comorbidities, advanced age, and presence of cavities. Mortality rates tend to be lower in Asian studies than in European or North American studies. In 2021, a Korean research group proposed the BACES score, which incorporates 5 components—body mass index (BMI), age, cavitation, elevated erythrocyte sedimentation rate (ESR), and male sex—as a prognostic factor for NTM-PD10. However, this scoring model is not specific to MAC-PD, as Mycobacterium abscessus complex (MABC), which is associated with poor prognosis, accounts for approximately 20% of the causative mycobacteria10. Furthermore, several variables, including male sex, age > 70 years, malignancy, BMI, lymphocyte count, albumin (Alb) levels, and fibrocavitary (FC) pattern, have been proposed in other prognostic score models for MAC-PD11. However, these models were developed in single tertiary care institutions, potentially including patients with more severe disease, and have not been validated in general healthcare environments. Furthermore, previous studies have focused on identifying prognostic factors for all-cause mortality, rather than those specific to respiratory infection-related death.

We conducted a multicentre retrospective cohort study of patients with NTM-PD in Kyushu, Japan, encompassing both tertiary care and community hospitals12. We aimed to develop and validate a prognostic scoring system to predict respiratory infection-related mortality in patients with MAC-PD.

Methods

Study design

This multicentre, retrospective, observational cohort investigation was conducted at 18 acute-care hospitals in Kyushu, Japan, as part of the Nagasaki, Fukuoka, Oita, Miyazaki, and Saga NTM study (NFOMS NTM study)12. The study population included patients aged ≥ 18 years with newly diagnosed NTM-PD who presented at the study centres between 1 January 2010 and 31 December 2017. Respiratory medicine specialists at each facility reviewed the medical records to collect data on patient characteristics, comorbidities, causative mycobacteria, laboratory findings, radiological features at the time of definitive NTM-PD diagnosis, treatment, and outcomes during the observation period. Patient follow-up continued until 31 December 2022. All data were collected using Research Electronic Data Capture (REDCap®). Data regarding the duration of oral corticosteroid (OCS) administration and the total dosage were not collected. Given the nature of multicentre retrospective studies, it is difficult to determine whether death was attributable solely to MAC-PD. MAC-PD is frequently associated with structural lung destruction, which may lead to respiratory failure or predispose patients to secondary infections caused by non-MAC pathogens—both of which can contribute to mortality. Therefore, we defined respiratory infection-related mortality as deaths in which MAC-PD was likely involved and evaluated the associated prognostic factors. Respiratory infection-related deaths were determined at the discretion of the attending physicians at each facility, based on their review of medical records.

To develop a scoring model for risk factors associated with respiratory infection-related mortality, eligible patients were randomly allocated to a 4:1 ratio to the derivation and validation groups. The risk factors for respiratory infection-related mortality were identified in the derivation group, and a scoring model was developed based on these factors to ensure consistency in the validation group. This study was conducted in accordance with the guidelines of the Declaration of Helsinki and approved by the Institutional Review Board (IRB) of Nagasaki University Hospital (approval number: 22121903). Verbal informed consent was obtained whenever possible, and an opt-out option was provided to the patients as this was a retrospective observational study. The IRB approved the waiver of informed consent.

Definitions for NTM-PD

The 2008 Japanese Society for Tuberculosis and Nontuberculous Mycobacteriosis (JSTNM) criteria were used to diagnose NTM-PD13. These diagnostic criteria were based on the 2007 official American Thoracic Society (ATS)/Infectious Diseases Society of America (IDSA)14. However, the JSTNM criteria do not include clinical symptoms, in contrast to the current 2020 ATS/European Respiratory Society/European Society of Clinical Microbiology and Infectious Diseases/IDSA guideline6. Identification methodologies for NTM species vary by institution and include PCR for M. avium and M. intracellulare, DNA–DNA hybridisation, and matrix-assisted laser desorption/ionisation-time-of-flight mass spectrometry. Radiological classification was conducted based on chest computed tomography and categorised by experts as nodular bronchiectasis (NB), NB with a cavity, or FC patterns.

Exclusion criteria

The primary objective of this study was to identify risk factors for respiratory infection-related mortality in patient with MAC-PD and develop a prognostic scoring model. Consequently, patients without MAC-PD were excluded. Further exclusion criteria included referral to other institutions for initial treatment.

Statistical analysis

Continuous variables were analysed using the Mann–Whitney U test, whereas categorical variables were examined using Fisher’s exact test. Cox proportional hazards analysis included variables that showed significant differences as assessed by experts, as well as those considered clinically relevant. A MAC-PD mortality prediction scoring model was developed based on Cox proportional hazards analysis results, and a receiver operating characteristic (ROC) curve was generated to evaluate the model’s performance. Cases with missing data were excluded from the multivariable analysis, and multicollinearity was assessed using the variance inflation factor (VIF), with a cutoff value of VIF > 5.

Survival analysis was performed using the log-rank test. All P values were two-sided, and P ≤ 0.05 were considered statistically significant. Group comparisons, multivariable analysis, and survival analysis were performed using GraphPad Prism ver. 10.4.2 (GraphPad Software, CA, USA), while the evaluation of ROC curves for the scoring model was conducted using JMP Pro ver. 18.1.0 (SAS Institute, Cary, NC, USA).

Results

Patients characteristics

During the study period, 1,317 patients were newly diagnosed with NTM-PD. The final analysis included 1,165 cases, with 139 cases excluded because of non-MAC-PD (Fig. 1). The mean age was 70.7 years, and 371 (31.8%) were male. M. avium and M. intracellulare were detected in 517 (44.4%) and 648 (55.6%) patients, respectively (Table 1). Within 5 years, 656 (56.3%) patients received treatment for MAC-PD. During the observation period, all-cause mortality occurred in 183 patients (15.7%), and respiratory infection-related mortality occurred in 67 patients (5.8%). In total, 932 and 233 patients were randomly allocated to the derivation and validation groups, respectively.

Fig. 1
figure 1

Flowchart of study patients. JSTNM, Japanese Society for Tuberculosis and Nontuberculous Mycobacteriosis; MAC, Mycobacterium avium-intracellulare complex; NTM-PD, nontuberculous mycobacterial pulmonary disease.

Full size image
Table 1 Patients characteristics at diagnosis and during clinical course.
Full size table

Risk factors for mortality

To identify the risk factors for respiratory infection-related mortality in MAC-PD, we evaluated the backgrounds, comorbidities, and laboratory findings at diagnosis of 932 patients in the derivation group. Univariate analysis revealed that age ≥ 65 years, male sex, BMI < 18.5, interstitial pneumonia (IP), chronic obstructive pulmonary disease (COPD), bronchial asthma, oral corticosteroid (OCS) use, inhaled corticosteroid use (ICS), Alb < 3.5 g/dL, elevated ESR (> 15 mm/h in male and 20 mm/h in female), lymphocyte count < 1,000/µL, M. intracellulare, cavity formation, and cavitary nodular bronchiectatic pattern were associated with an increased mortality risk (Table 2). Cox proportional hazards analysis was performed using the following variables: age ≥ 65 years, male sex, IP, COPD, ICS use, Alb < 3.5 g/dL, and cavitary lesions ≥ 2 cm in diameter (Table 2). Multicollinearity among these variables was assessed using the VIF, and all variables had VIF values below 3, indicating negligible multicollinearity. ESR was excluded because of missing data for 563 patients (60.4%). The results indicated that age ≥ 65 years, male sex, IP, Alb < 3.5 g/dL, and cavitary lesions ≥ 2 cm in diameter were independent risk factors for respiratory-infection related mortality (Table 2).

Table 2 Prognostic analysis of risk factors for respiratory infection-related mortality in the derivation group.
Full size table

Overall, respiratory infection-related death occurred in 32 males (11.0%) and 19 females (3.0%). Male had a significantly higher respiratory infection-related mortality rate than female, as assessed by the log-rank test, with a crude hazard ratio of 5.12 (95% confidence interval [CI]: 2.63–9.99) (Fig. 2a). Age ≥ 65 years was significantly associated with respiratory infection-related mortality (HR 4.02, 95% CI: 2.18–7.44; Fig. 2b), as was IP (HR 5.32, 95% CI: 1.18–24.1; Fig. 2c). Hypoalbuminemia (Alb < 3.5 g/dL) demonstrated the strongest association (HR 8.69, 95% CI: 3.69–20.5; Fig. 2d), while cavitary lesions ≥ 2 cm in diameter were also significantly associated (HR 4.20, 95% CI: 1.56–11.3; Fig. 2e).

Fig. 2
figure 2

Kaplan–Meier analyses of patients in the derivation group. Male sex (a), age over 65 years (b), IP (c), hypoalbuminemia (d), and cavity formation (e) were associated with a significantly poorer prognosis. Without cavity ≥ 2 cm includes patients with no cavity or cavities < 2 cm. Cases with missing data were excluded from the analysis. IP, interstitial pneumonia.

Full size image

Throughout the study, 33 patients (3.5%) in the derivation group and 6 patients (2.6%) in the validation group had coinfection with chronic pulmonary aspergillosis (CPA), either at diagnosis or during follow-up (Table 1). In a univariate analysis of respiratory infection-related mortality, the development of CPA was significantly associated with mortality, with a crude HR of 6.64 (95% CI: 3.0–13.0). Because CPA typically developed after the diagnosis of NTM-PD, it was not included as a variable in the Cox proportional hazards analysis.

Construction of the prognostic scoring model

We included 5 independent risk factors identified through multivariate analysis to develop a prognostic scoring model. Based on the adjusted hazard ratios from the Cox proportional hazards model in the derivation group, we developed a scoring system (MAC prognostic score) by assigning 2 points to Alb < 3.5 g/dL, and 1 point each to age ≥ 65 years, male sex, IP, and cavitary lesions ≥ 2 cm in diameter.

Among the 795 patients in the derivation group without missing scoring data, the area under the ROC curve (AUC) for the MAC prognostic score was 0.86 (Fig. 3a). To determine the optimal prognostic cutoff value for respiratory infection-related mortality within 5 years, we calculated the Youden index (sensitivity + specificity − 1) based on the ROC curve. A score of ≥ 3 yielded the highest index. At this cutoff, the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), positive likelihood ratio, and negative likelihood ratio were 81.8%, 74.7%, 12.3%, 99.0%, 3.23, and 0.14, respectively. The treatment induction rates within 5 years of diagnosis were 75.0% (96/128) for score 0, 54.3% (242/447) for scores 1–2, 53.4% (99/185) for scores 3–4, and 62.9% (22/35) for scores 5–6. The Kaplan–Meier curves for the MAC prognostic score demonstrated significantly different prognoses for each score (Fig. 3b). The 5-year respiratory infection-related mortality rates were 0.0% (0/128) for score 0, 1.3% (6/447) for scores 1–2, 8.1% (15/185) for scores 3–4, and 34.3% (12/35) for scores 5–6. A higher score was associated with a poorer prognosis.

Fig. 3
figure 3

Receiver operating characteristic curve for MAC score and Kaplan–Meier analysis in the derivation group. (a) ROC curve of MAC score in the derivation group (n = 795). Cases with missing data were excluded from the analysis. The AUC was 0.86. The sensitivity and specificity were 81.8 and 74.7%, respectively, at a cutoff value of 3. (b) Kaplan–Meier analysis stratified by MAC score in the derivation group. MAC scores exceeding 3 points showed poor survival outcomes. AUC, area under the curve; ROC, receiver operating characteristic.

Full size image

Validation of the MAC prognostic score

The MAC prognostic score was subsequently adapted for the validation group and evaluated. Figure 4a shows the ROC curve based on 196 patients without missing scoring data, with an AUC of 0.81. At a cutoff value of 3 points, the sensitivity, specificity, PPV, and NPV for respiratory infection-related mortality within 5 years were 76.9%, 73.8%, 17.2%, and 97.8%, respectively. The treatment induction rates within 5 years of diagnosis were 76.5% (26/34) for score 0, 54.8% (57/104) for scores 1–2, 47.8% (22/46) for scores 3–4, and 66.7% (8/12) for scores 5–6. The Kaplan–Meier curves also demonstrated different prognoses for each score in the validation group (Fig. 4b). The 5-year respiratory infection-related mortality rates were 0.0% (0/34) for score 0, 2.9% (3/104) for scores 1–2, 8.7% (4/46) for scores 3–4, and 25.0% (3/12) for scores 5–6. In the validation group, scores of 3 or higher were associated with a significantly worse prognosis compared to scores below 3.

Fig. 4
figure 4

Receiver operating characteristic curve for MAC score and Kaplan–Meier analysis in the validation group. (a) ROC curve of MAC score in the validation group (n = 196). Cases with missing data were excluded from the analysis. The AUC was 0.81. The sensitivity and specificity were 76.9 and 73.8%, respectively, at a cutoff value of 3. (b) Kaplan–Meier analysis stratified by MAC score in the validation group. MAC scores exceeding 3 points showed poor survival outcomes. AUC, area under the curve; ROC, receiver operating characteristic.

Full size image

Comparison with the BACES score

The MAC prognostic scores were compared with the BACES scores. A total of 373 patients were included in the analysis, after excluding the cases lacking BMI data195 (16.7%), ESR data 698 (59.9%), or Alb data 173 (14.8%). The AUCs of the MAC prognostic score and BACES score were 0.82 and 0.88, respectively, with no statistically significant differences (Fig. 5a). The Kaplan–Meier curves for the MAC prognostic score and BACES score are presented (Figs. 5b, c). Compared with the BACES score, the MAC prognostic score demonstrated a non-inferior ability to identify the poor prognosis group.

Fig. 5
figure 5

Comparison of MAC and BACES scores. (a) ROC curves of MAC and BACES scores in the population without data loss (n = 373). The AUCs of the MAC and BACES scores were 0.82 and 0.88, respectively, with no significant differences. Kaplan–Meier curves for (b) the MAC score and (c) the BACES score showed comparable prognostic discrimination. AUC, area under the curve; ROC, receiver operating characteristic.

Full size image

Discussion

In this study, we developed a prognostic scoring model to identify patients predicted to have poor prognosis in patients with MAC-PD. Multivariate analysis revealed that age ≥ 65 years, male sex, IP, albumin < 3.5 g/dL, and cavitary lesions ≥ 2 cm in diameter were independent prognostic factors. The MAC prognostic score demonstrated efficacy in identifying patients with poor prognosis.

BACES score is a prognostic tool for NTM-PD, and a separate prognostic score for MAC-PD has been reported in Japan10,11. The 5-year all-cause mortality rates in the derivation cohorts of these two reports were 21% and 20.5%, respectively, both higher than the 15.7% observed in the present study. The derivation cohorts in these reports were drawn from a single-centre study at a tertiary care hospital, which may have had a more severe case distribution than that in our report. In contrast, our study included 18 facilities with diverse backgrounds, ranging from tertiary care facilities to local hospitals12. A key strength of this study is its multicentre design, incorporating diverse hospitals, which enhances the generalizability of the findings to real-world clinical practice. Furthermore, while previous studies have investigated predictors of all-cause mortality, this study uniquely identified risk factors specifically associated with respiratory infection-related mortality.

When developing the scoring model, the variables were limited to those with a high-hazard ratio, those of clinical significance, or those whose data could be readily obtained in practice. This approach was necessary because of the small number of events in the derivation group (n = 51) and the need to develop a clinically applicable and efficient scoring model. In the present study, age ≥ 65 years, male sex, IP, albumin < 3.5 g/dL, and cavitary lesions ≥ 2 cm in diameter were identified as risk factors. Advanced age, male sex, hypoalbuminemia, and cavitary lesions have been identified as independent prognostic factors in several studies on NTM-PD10,11,15,16. IP has been identified as an independent risk factor for the development of NTM-PD, as demonstrated by a meta-analysis17. However, IP has not been clearly established as a prognostic factor for NTM-PD in large-scale studies. In the present study, we did not collect specific information on the subtypes of IP. However, several studies have suggested that the prognosis of patients with pleuroparenchymal fibroelastosis (PPFE), a distinct subtype of IP, complicated by NTM-PD, is extremely poor. For instance, radiological findings consistent with PPFE have been observed in 11.9% of patients with MAC-PD, and these patients exhibited significantly worse prognoses compared to those without such findings18. Moreover, the prognosis of NTM-PD patients with concomitant PPFE is poorer than that of patients with cavitary NB or FC types of NTM-PD19. In addition, respiratory infections such as NTM-PD have been suggested as potential causative factors in non-idiopathic PPFE20. These observations suggest a close association between PPFE and NTM-PD. Although the presence of PPFE could not be assessed in the present study, it is possible that the poor prognosis observed in patients with IP may have been influenced by factors related to PPFE. Consistent with previous reports, the present study also found that the coinfection of CPA during the disease course was associated with a poorer prognosis21,22,23,24. However, because CPA is generally not present at the time of MAC-PD diagnosis, it was not included as a variable in the multivariable analysis.

Despite the higher mortality risk associated with MAC prognostic scores of 5–6, treatment initiation rates were highest in patients with a score of 0 in both the derivation and validation groups. Multivariable analysis identified older age and dementia as negative predictors of treatment initiation, while connective tissue diseases, positive sputum acid-fast bacilli smear, and cavitary lesions were positive predictors (Table S1). As shown in Figure S1, treatment initiation rates declined with increasing age, suggesting that inclusion of age in the score may account for the lower treatment rates among high-score patients. A study involving older patients with MAC-PD reported lower treatment success rates and reduced prognostic impact of therapeutic interventions in those aged ≥ 80 years25, suggesting that the diminished benefit of treatment in this population may have contributed to the lower treatment initiation rates observed in our study.

The MAC prognostic score demonstrated remarkable prognostic predictive capabilities. Among the 991 patients with complete data, 27.7% (220/795) in the derivation group and 29.6% (58/196) in the validation group had scores ≥ 3, which were associated with higher rates of respiratory infection-related mortality. Compared to the BACES score, the MAC prognostic score exhibited non-inferior performance in predicting poor prognosis in patients with MAC-PD. Importantly, the BACES score is not directly comparable, owing to the inclusion of MABC in addition to MAC. Furthermore, while the BACES score evaluates all-cause mortality, our study assessed respiratory infection-related mortality, which more specifically reflects the prognosis directly attributable to MAC-PD.

This study has some limitations. First, due to the retrospective design, some patients had missing data. To ensure data integrity, these patients were excluded from the analysis, and this exclusion may have introduced selection bias. Second, this study was conducted in the Kyushu region of Japan and has not been validated in other regions. Third, although respiratory medicine specialists at each facility reviewed this multicentre study, no standardised validation of the radiological findings and disease type was performed. Furthermore, institutional variations in treatment regimen selection were not considered. In addition, the attribution of death to respiratory infection or other causes was based on the clinical judgment of the attending physicians, which may have introduced variability. Fourth, drug susceptibility test results were not evaluated, particularly macrolide resistance. Notably, a previous study reported a poor prognosis in patients with macrolide-resistant MAC-PD, with a 5-year mortality rate of 47%26, which was not assessed in the present study. Fifth, the association between treatment regimens and prognosis was not examined. Lastly, the scoring model demonstrated a relatively low PPV, possibly due to the relatively small number of deaths in this cohort compared with previous studies. Importantly, few studies have identified prognostic factors specifically associated with respiratory infection-related mortality in patients with MAC-PD. The findings of this multicentre study, therefore, provide valuable insights into disease-specific prognosis and have important implications for clinical risk stratification.

This study identified the following prognostic factors for MAC-PD: age ≥ 65 years, male sex, interstitial pneumonia, albumin < 3.5 g/dL, and cavitary lesions ≥ 2 cm in diameter. A prognostic model incorporating these factors effectively stratified patients by mortality risk. The strengths of this study include its multicentre design, which reflects diverse clinical settings, and the specific evaluation of respiratory infection-related mortality, offering a more disease-focused assessment. Further large-scale prospective studies are warranted to validate this model in other populations.