We identified 192 articles for title and abstract screening and 8 articles through additional sources (Fig. 1). Thirteen articles met inclusion criteria [13,14,15,16,17,18,19,20,21,22, 27,28,29], all observational studies. Two studies were secondary analyses of randomized trials but were analyzed as observational studies [13, 20]. Study characteristics including author and year, country, number of participants, description of the population (sex, current smoking status, race), FEV1, baseline serum bilirubin levels, upper limit of bilirubin levels, outcomes of interest, and study level risk of bias are listed in Additional file 1: Table S1.
Studies excluded after full-text review are listed in Additional file 1: Appendix 4.
Four [13, 16, 19, 21] of the 13 studies [13,14,15,16,17,18,19,20,21,22, 27,28,29] were from North America (United States and Canada), four from Europe [14, 17, 18, 29], four from East Asia (China or South Korea) [15, 22, 27, 28], and one was from a cohort of participants in 20 countries  (Additional file 1: Table S1). Five studies were rated low risk of bias [13, 15, 17, 19, 20], three moderate risk of bias [14, 18, 27], and five high risk of bias [16, 21, 22, 28, 29] (Additional file 1: Table S2). Sample size ranged from 131 to 504,206 [17, 22]. The mean age was 54.4 years (10 studies reporting) [13,14,15,16,17,18,19, 21, 27, 28] and 58.8% were men (12 studies reporting) [13,14,15,16,17,18,19,20,21,22, 27, 29]. All studies reported on current smoking status, and the weighted proportion of current smokers was 22.8%; two studies excluded smokers [21, 22]. The mean bilirubin level was 0.62 mg/dL (range 0.37–0.76 mg/dL) (8 studies reporting) [13,14,15,16,17, 19, 21, 27]. Three studies [13, 17, 27] excluded participants with a bilirubin level above 1.75 mg/dL for women and 2.34 mg/dL for men, and one study  excluded participants with a bilirubin level > 0.99 mg/dL. Five studies [13, 16, 19,20,21] reported race; one study (38.2% Black, 17.5% Latino/Hispanic, 9.6% Asian, 33.7% White, and 1.0% other) had broad representation , one was 40.4% non-Hispanic White , and the remainder were predominantly White (range 77–100%) [13, 19, 21].
Seven studies evaluated the association between serum bilirubin and clinical outcomes of interest [13, 17, 19, 22, 27,28,29]. Eight studies [13,14,15,16, 18, 20, 21, 27] reported the association between serum bilirubin and FEV1 and four studies [15, 18, 20, 27] evaluated the association between serum bilirubin and FEV1/FVC ratio.
Figure 2 illustrates the summary of the measured outcomes, stratified by statistical significance. Additional file 1: Table S3 provides the overall assessment of strength of evidence for all outcomes evaluated.
Seven studies evaluated clinically important outcomes and are shown in Table 1 [13, 17, 19, 22, 27,28,29].
Three studies evaluated mortality [13, 17, 27]. One large (n = 504,206) primary care based medical record analysis with low risk of bias found a statistically significant adjusted incidence rate ratio (aIRR) for both men (0.97, 95% CI 0.97–0.98) per 0.1 mg/dL increase in serum bilirubin] and women (0.97, 95% CI 0.96–0.98) per 0.1 mg/dL increase in serum bilirubin] . The other two studies, both in cohorts of persons with obstructive lung disease, found no association between serum bilirubin and all-cause mortality [13, 27]. One of these studies, in a North American longitudinal cohort of smokers with obstructive spirometry, found no association with all-cause mortality [odds ratio (OR) in quintile 1 vs 5 (reference) 1.15, 95% CI 0.80–1.65] or respiratory disease-related mortality [OR in quintile 1 vs 5 (reference) 1.31, 95% CI 0.42–4.09], but there was an association with coronary heart disease-related mortality (OR 2.20 in quintile 1 vs. quintile 5, 95% CI 0.95 to 5.14; p for trend = 0.03) .
We found an inverse association of bilirubin levels with all-cause mortality (“low” strength of evidence). Though the largest study showed a statistically significant association between higher serum bilirubin and lower mortality , the overall body of evidence lacked precision and consistency.
Acute exacerbations of COPD
Two studies evaluated the association between serum bilirubin and AECOPD [19, 27]. Both found higher serum bilirubin was associated with significantly lower rates of AECOPD [19, 27]. One low ROB study was a secondary analysis of two randomized trials of participants with COPD enriched for participants prone to exacerbations. The model was developed in one cohort (n = 853), where bilirubin was not a significant predictor of COPD exacerbation (adjusted hazard ratio per log10 increase in bilirubin = 0.89; 95% CI 0.74–1.09, p = 0.26) but in the validation cohort (n = 1018) the relationship between higher serum bilirubin and time to first AECOPD was significant (adjusted hazard ratio per log10 increase in bilirubin = 0.80; 95% CI 0.67–0.94, p = 0.008) . The other study (n = 535), which had moderate risk of bias, was in a cohort of patients with obstructive lung disease and found that higher serum bilirubin was associated with lower rates of AECOPD [estimated mean from regression 0.62 (standard error 0.18; p = 0.001)] .
We found an inverse association between bilirubin levels and AECOPD (“low” strength of evidence). Though both studies found a statistically significant association between higher serum bilirubin and lower rates of AECOPD, one study found a statistically significant relationship only in the validation cohort, not the development cohort and the other had a moderate risk of bias [19, 27]. There were no data specific to the association between serum bilirubin and risk of hospitalization for AECOPD.
Respiratory health status and exercise capacity
One moderate ROB study evaluated respiratory health status (CAT and SGRQ) and exercise capacity (6-min walk test) and found no significant association between serum bilirubin and any of these measures . We determined that there was insufficient evidence on the association of bilirubin with respiratory health status and exercise capacity.
Four studies evaluated the association between serum bilirubin and COPD diagnosis [15, 17, 20, 27]. One large (n = 504,206) longitudinal cohort study with low risk of bias found a small but statistically significant decrease in the aIRR of COPD diagnosis based on EMR coding with increasing levels of bilirubin [for men aIRR 0.94 (0.93, 0.95) per 0.1 mg/dL increase in bilirubin; for women aIRR 0.94 (0.92, 0.95) per 0.1 mg/dL increase in bilirubin] . Three studies were small, cross-sectional, and had high risk of bias; none found a significant relationship [22, 28, 29]. We found an inverse association between bilirubin levels and COPD diagnosis. We assigned “low strength evidence” based upon the moderate overall bias, inconsistent effects, and imprecision of reported outcomes.
Lung function (FEV1)
Results for studies evaluating the association between serum bilirubin and FEV1 are shown in Table 2. Among the three studies (all low ROB) [13, 15, 20] that evaluated longitudinal change in FEV1, two found a statistically significant relationship. One was a general population cohort with a mean follow-up of 5.4 years (FEV1 declined 13.09 mL/yr slower for every 1 unit increase in natural log of bilirubin) and the other was a longitudinal cohort of smokers with a mean FEV1/FVC ratio of < 0.7 with up to 9 years of follow-up (FEV1 declined 57.0 mL/yr in the highest quintile of bilirubin and 66.3 mL/year in the lowest) [13, 15]. The study that did not find a statistically significant relationship was from a smaller cohort of young, HIV positive adults who were followed for a median of 3.9 years .
Five studies [14, 16, 18, 21, 27], 3 moderate ROB [14, 18, 27] and 2 high ROB [16, 21] evaluated cross-sectional associations between bilirubin and FEV1. Two studies found a statistically significant relationship between higher FEV1 and higher serum bilirubin [14, 16]. One low risk of bias cohort sample from England, Wales, and Scotland  evaluated 5,362 people born in the same week in 1946 and the other high risk of bias study  was from a nationally representative cohort in the United States. Three found no significant relationship between FEV1 and serum bilirubin levels [18, 21, 27].
We found an association between higher bilirubin levels and improved lung function (FEV1). We assigned a “low” strength of evidence due to the moderate overall risk of bias of the eight studies evaluated, clinically small effect, inconsistent effects, and imprecision of the outcome (unreported or wide confidence intervals).
Airflow obstruction (FEV1/FVC ratio)
Four studies evaluated the association between serum bilirubin and FEV1/FVC ratio, a spirometric marker of obstructive lung disease [15, 18, 20, 27] (Table 3). Three of the studies were longitudinal [15, 20, 27]. One longitudinal study with low risk of bias found an association between higher baseline bilirubin levels and faster decline in FEV1/FVC ratio [ß for % decline in FEV1/FVC per 1 unit change in natural log of bilirubin (mg/dL) = 1.69; p < 0.001] ; two found no significant relationship [20, 27]. A cross-sectional population-based analysis with moderate risk of bias found a significant association between higher bilirubin levels and less airflow obstruction [ß for % decline in FEV1/FVC per 1 unit increase in natural log in bilirubin (μmol/L) 0.5; 95% CI 0.1–1.0, p = 0.012] . There was insufficient evidence to evaluate the relationship between airflow obstruction and serum bilirubin.