Rats sniff out pulmonary tuberculosis from sputum: a diagnostic accuracy meta-analysis

Abstract

In Sub-Saharan Africa, African giant pouched rats (Cricetomys gambianus) are trained to identify TB patients by smelling sputum. We conducted a systematic review and meta-analysis of the data to see if this novel method is comparable to traditional laboratory screening and detection methods like Ziehl-Neelsen stain-based assays (ZN) and bacterial culture. The search and data processing strategy is registered at PROSPERO (CRD42019123629). Medline via PubMed, EMBASE, Web of Science, and Cochrane Library databases were systematically searched for the keywords "pouched rat" and "tuberculosis". Data from 53,181 samples obtained from 24,600 patients were extracted from seven studies. Using sample-wise detection, the sensitivity of the studies was 86.7% [95% CI 80.4-91.2%], while the specificity was 88.4% [95% CI 79.7-93.7%]. For patient-wise detection, the sensitivity was 81.3% [95% CI 64.0-91.4%], while the specificity was 73.4% [95% CI 62.8-81.9%]. Good and excellent classification was assessed by hierarchical summary receiver-operating characteristic analysis for patient-wise and sample-wise detections, respectively. Our study is the first systematic review and meta-analysis of the above relatively inexpensive and rapid screening method. The results indicate that African giant pouched rats can discriminate healthy controls from TB individuals by sniffing sputum with even a higher accuracy than a single ZN screening.

Figure 1

Study selection. PRISMA 2019 flow diagram. Database searches identified 84 records. Of the overall obtained studies, 38 were duplicates. From the 46 remaining studies, 30 were excluded either by examining the title or by examining the abstract. Of the remaining 16 studies, full-text examination resulted in identifying seven publications that are eligible for meta-analysis inclusion criteria.

Figure 2

Sensitivity (A) and specificity (B) of individual studies are displayed by squares. The overall values are displayed by rhombus. Error bars indicate confidence interval of 95% [95% CI]. The reference method was Ziehl–Neelsen stain-based, direct microscopy, except for other method is indicated in brackets.

Figure 2

Sensitivity (A) and specificity (B) of individual studies are displayed by squares. The overall values are displayed by rhombus. Error bars indicate confidence interval of 95% [95% CI]. The reference method was Ziehl–Neelsen stain-based, direct microscopy, except for other method is indicated in brackets.

Figure 3

The size of circles indicates the number of patients in a single study. The dark red square shows the sensitivity and specificity summary. The dashed line indicates the 95% confidence region. A meta-analysis of all seven studies revealed that the summary of sensitivity was 81.3% [95% CI 64.0–91.4%] and the specificity was 73.4% [95% CI 62.8–81.9%]. The diagnostic odds ratio was 12.0 [95% CI 3.58–39.9]. According to the AUC 0.82 [95% CI: 0.79–0.86] value, the test was classified as good. HSROC, hierarchical summary receiver-operating characteristic; AUC, area under the curve.

Figure 4

Sensitivity (A) and specificity (B) of individual studies are displayed by squares. The overall values are displayed by rhombus. Error bars indicate confidence interval of 95% [95% CI]. The reference method was Ziehl–Neelsen stain-based, direct microscopy, except for other method is indicated in brackets.

Figure 4

Sensitivity (A) and specificity (B) of individual studies are displayed by squares. The overall values are displayed by rhombus. Error bars indicate confidence interval of 95% [95% CI]. The reference method was Ziehl–Neelsen stain-based, direct microscopy, except for other method is indicated in brackets.

Figure 5

Diagnostic performance of screening using rats compared to ZN microscopic analysis, LED FM, and culture. The size of circles indicates the number of patients in a single study. The dark red square shows the sensitivity and specificity summary. The dashed line indicates the 95% confidence region. Meta-analysis summary of the six studies (five circles are seen, because two studies are exactly overlapping) shows that the sensitivity of rat screening was 86.7% [95% CI 80.4–91.2%] and specificity was 88.4% [95% CI 79.7–93.7%]. The summary for positive LR + and LR − was 7.47 [95% CI 4.05–13.8] and 0.15 [95% CI 0.1–0.23], respectively. The diagnostic odds ratio was 49.8 [95% CI 19.5–127]. According to the AUC 0.93 [95% CI: 0.91–0.95] value the test was classified as excellent. HSROC, hierarchical summary receiver-operating characteristic; AUC, area under the curve.

Figure 6

The quality of each study was assessed by answering questions concerning the clarity of research, blindedness, representative and adequate samples, control of confounding variables, research design suitable to answer the research question, ethical clearance, reporting overall sensitivity percentage, reporting overall specificity percentage, and reporting limitations. The presence of non-mycobacterial species in addition to failure to report HIV status were deemed confounding factors. Green = yes; yellow = unclear; red = no.