Accuracy of the chest x-ray in screening for tuberculosis in Uganda: A cross-sectional study.


 ABSTRACT

BACKGROUND: The WHO END TB strategy requires ≥90% case detection to combat tuberculosis (TB). Increased TB case detection requires a more sensitive and specific screening tool. Currently, the symptoms recommended for screening TB have been found to be sub-optimal. CXR as a screening tool for pulmonary TB was evaluated in this study, as well as factors related to its false positive results. 

METHODS: A cross sectional study of 4441 records of consented/assented participants ≥15 years. Participants with a cough ≥2 weeks and/or any abnormality in the lung on CXR were included in the study. Löwenstein-Jensen (LJ) culture was used as the gold standard. The CXR were categorised as Abnormal meaning presence of any CXR abnormality suggestive of active tuberculosis. Symptoms were categorised as abnormal meaning presence of any of cough ≥ 2 weeks, fever, weightloss or night sweats.

RESULTS: The CXR had sensitivity 93%, specificity 65% compared to culture results while symptoms had sensitivity 76% and specificity 31%. The adjusted prevalence ratio (APR) of a false positive CXR result increased with age categories (years); 45 - 54, APR 1.18 (1.08, 1.29), 55 – 64 APR 1.18 (1.09, 1.29), 65+years APR 1.2 (1.10, 1.30).The APR was 0.93 (0.90, 0.96) among males and 0.86 (0.79, 0.93) among HIV positive individuals.

CONCLUSIONS: The CXR is a fair tuberculosis screening tool and performed better than symptoms in Uganda.


Background
The WHO 2018 estimates show that Tuberculosis (TB) was responsible for 1.2 million deaths among HIV negative persons and 250,000 deaths among HIV positive persons. (1) Uganda has a high TB incidence of 201 per 100,000 population as well as high TB related mortality of 26 per 100,000 population. The TB case detection rate for Uganda is 72%. (2) In order to combat tuberculosis, WHO has proposed three strategies ; intensi ed case nding (ICF), isoniazid preventive therapy (IPT) and infection control (IC). (3) The WHO END TB strategy indicates a need for ≥ 90% TB case detection among others to combat TB. (4) The strategic plan for the Uganda national tuberculosis and leprosy programme (NTLP) seeks to achieve 85% case detection by 2019/2020. However, con rming TB is di cult. Currently TB culture is the gold standard but nancial and logistic challenges make it di cult to scale up. There is need to explore more widely available, low cost screening and diagnostic tools and algorithms to aid TB diagnosis. The CXR is one such potential tool.
The currently employed symptoms screening tool for tuberculosis in Uganda has been observed to have sensitivity 40.7% and speci city of 81.3% according to a survey done in South Africa considering presence of any cough, fever, weightloss or Night sweats. (5) Systematic reviews done to assess the performance of the symptoms in high HIV prevalence regions have shown sensitivity 84% and speci city 74%. (6) The symptoms screening tool runs a risk of missing patients with tuberculosis as the patients do not seek care and do not have symptoms due to early TB disease.

Page 3/19
The chest radiograph (CXR) was commonly used in mass screening campaigns in Europe and the US in the early TB chemotherapy era (1950s), but has not been widely used in screening TB in low-income countries because of high cost and lack of capacity. (7,8) This CXR has been found to have sensitivity of 98% and speci city of 75% according to a systematic review. (6) The recognition that bold new strategies are needed to control TB (END TB Strategy) has led to reconsideration of CXR as a potential tool, as innovative, lower costs strategies for performing CXR (digital radiography) and reading them (computer aided reading) have become available. (9,10) The use of more sensitive tuberculosis screening tools increases the pool of presumptive TB cases hence increasing case detection as more individuals are exposed to the con rmatory test which ultimately leads to reduction in mortality and morbidity. (11) The impact of false positive results in screening of tuberculosis.
The performance of screening tests is affected by the level of false negative and false positive results that are obtained as a result of using the test. The CXR and symptoms have been reported to have a very high number of false positive results hence challenges in their application as many individuals who would have been saved from a con rmatory test will still undergo the con rmatory test only to nd them without the disease. (12,13) The identi cation of the factors responsible for false positive results is important as it guides the application of the test to individuals who are more likely to bene t from the test. A study done in Tennessee found a high rate of normal CXR among PLHIV with culture con rmed tuberculosis. (14) These factors if identi ed help guide the implementation of screening interventions in the various populations and countries.
Uganda carried out a national tuberculosis prevalence survey (UTPS) from 2014-2015 which screened for tuberculosis by collecting data on TB symptoms using questionnaires and performed CXR in all consenting participants. (15) Therefore using the data from UTPS, we sought to evaluate the performance of the Chest radiograph in the screening of tuberculosis and factors associated with a false positive CXR.

Study design
We conducted a cross sectional study by secondary analysis of the data collected during the Uganda National Tuberculosis prevalence survey (UTPS). Study design, setting and sampling method used in the UTPS The Uganda National TB prevalence survey was conducted from October 2014 to July 2015 with the primary objective of determining the prevalence of Tuberculosis in Uganda. Villages (clusters) across the country were sampled using probability proportionate to size (PPS). The starting block was randomly selected by the village leader and then blocks were added in a clockwise manner around the original block until the required cluster size was achieved. Eligible respondents aged ≥ 15 years in selected blocks were invited to the survey (15) .

Study procedures
Screening strategy All eligible consenting/assenting individuals who were ≥ 15 years were screened for tuberculosis by research assistants in the community using a questionnaire which listed the symptoms and a CXR (Fig. 1). Parents/Guardians of individuals < 18 years also offered informed consent.
Participants with a cough ≥ 2 weeks and/or any abnormality in the lung on CXR were considered presumptive for TB and were requested to submit two sputum samples (a spot and an early morning sample).
Respondents who did not have a CXR taken were also eligible to submit sputa, even if asymptomatic.
All respondents eligible for sputum collection had HIV testing.

CXR reading
CXRs were read in the eld by a trained technician and interpreted as normal or abnormal. Two experienced independent radiologists categorised all the abnormal CXRs as normal, suggestive of active TB disease, inactive/healed TB, and extra pulmonary abnormalities. A third radiologist adjudicated in case there were differences and the differences were resolved by consensus. All normal x-rays were read a single time.

The Gold standard
The culture results obtained with Löwenstein-Jensen medium and con rmed as survey TB were considered as the gold standard.

Study participants
Records of participants ≥ 15 years were included in the study. We excluded records that were missing data on culture and CXR results.

Sampling and sample size
All records in the electronic database that met the inclusion criteria for the study were analysed.

Results
Study Pro le A total of 40,539 participants were screened for TB in the UTPS. We excluded 35,690 records of participants that did not submit sputum to the laboratory. We excluded 348 records because they were To estimate the sample size for performance of the CXR and symptoms. The modi ed Kish Leslie formula for diagnostic studies was used. We assumed 95% con dence level (CI), sensitivity 90%, speci city 83% (16) and TB prevalence of 401 per 100,000 population (15) to obtain the desired sample size for assessing the performance of the CXR in screening for tuberculosis.

Data extraction
We used a data extraction form to obtain variables of interest from the electronic database.
The CXR results were considered as abnormal (suggestive of active TB disease) or Normal (healed TB/inactive TB, other TB related lung abnormalities, other non-TB related lung abnormalities and extra pulmonary abnormalities, normal).
The symptoms screen encompassed presence of cough of ≥ 2 weeks, fever, weight loss and night sweats as recommended by Uganda Ministry of Health. Individuals with any of the above symptoms were considered presumptive for tuberculosis.

Data analysis
We used the STARD 2015 guidelines in reporting the results of this study (17) . Data was analysed using Stata version 13.0 (College Station, Texas, USA). In our study, records with any of the above symptoms i.e. cough ≥ 2 weeks, fever, weightloss or night sweats were analysed as positive for symptoms. Records indicating CXR abnormalities suggestive of active TB were analysed as positive by the CXR.
Variables were summarised using percentages and proportions depending on their distribution. Performance of the symptoms screening and CXR were reported as sensitivity, speci city, negative and positive predictive value, negative and positive likelihood ratios.
The chi square was used to determine the association between each predictor and false positive CXR results. These were reported as prevalence ratios due to the high prevalence of both CXR and symptoms false positive results in the population.
The Poisson multivariate regression model with robust standard errors was used to obtain factors that were independently associated with false positive CXR or symptoms. The results were reported as prevalence ratios because the prevalence of false positive results was greater than 10%. All factors with level of signi cance ≤ 5% were considered signi cant. Interaction was assessed using the chunk test and considered present at P-value < 0.05. Confounding was assessed and considered present at ≥ 10% difference in the prevalence ratio. missing data on culture results. 60 records were excluded due to missing data on CXR results. The remaining 4441 records were analysed (Fig. 2).

Discussion
This study is one of the rst studies to compare the performance of the CXR or symptoms against culture con rmed pulmonary tuberculosis by LJ Method in a community tuberculosis screening setting such as the National Tuberculosis prevalence surveys.
Our results show that the CXR performs better than the symptoms in correlating with positive tuberculosis sputum culture results in community settings which agrees with other studies. (3,12,(18)(19)(20)(21) The positive and negative likelihood ratios of the CXR versus symptoms further indicate that the CXR performs better in correlating tuberculosis sputum culture results. The results indicate that the use of the CXR in tuberculosis screening in community settings may hence increase tuberculosis case detection rate in high tuberculosis prevalence settings. This may be due to the increased number of individuals that are taken for the con rmatory tests. This is especially important in the era of the new advanced TB diagnostic tools.
Our results indicate that individuals above 45 years are more likely to have a false positive CXR when screened for tuberculosis. This may be explained by the increase in respiratory pathology associated with increase in age including pneumonia, chronic obstructive pulmonary disease among others. (22) Therefore, the CXR is not a preferred TB screening test among these individuals.
HIV positive individuals and males are more likely to have a true diagnosis of tuberculosis with both the CXR and symptoms. (18) This is important in the diagnosis of tuberculosis since these populations have been shown to be at high risk for tuberculosis. (23) However, other studies have shown that PLHIV have been found to have a lot of abnormalities on CXR and hence the CXR not a good TB diagnostic tool among such populations. (13) Therefore further studies are required to investigate the discrepancy noted between our study and other studies.
The strengths of our study include community screening for tuberculosis as shown by the use of data collected during the UTPS. The study was reported according to the STARD (Standards for reporting of Diagnostics Accuracy Studies) 2015 guidelines. (17) Our study is subject to limitations. The fact that sputum culture was only done for participants with abnormal CXR or tuberculosis symptoms could have led to selection bias hence over estimation of the sensitivity of the CXR. However, the large sample size improves our estimation of the performance of the CXR.
We only looked at culture con rmed pulmonary tuberculosis but not clinically diagnosed and not culture negative or Extra Pulmonary TB. This could greatly impact the performance of symptoms screening and/ or CXR to identify true TB.
The study was performed under community screening conditions and therefore the results may not be applicable to tuberculosis screening in the hospital setting.
This study shows that community tuberculosis screening using the CXR improves tuberculosis case detection. Therefore, the CXR should be considered in community tuberculosis screening programmes.

Conclusion
This study made use of data obtained during the Uganda National Tuberculosis prevalence survey to determine the performance of the CXR in correlating tuberculosis culture results and factors associated with a false negative tuberculosis screening result. The results indicate that the CXR is a good tuberculosis screening tool and may improve TB case detection when used in community tuberculosis screening on addition to symptoms assessment. Individuals with knowledge to observe TB de ning pathology on the CXR are adequate to provide expertise for CXR interpretation.

Competing interests
The authors declare that they have no competing interests Diagrammatic representation of the combined symptom and chest x-ray screening strategy.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. AdditionalFile1.xls