Tuberculosis case finding in a population with high HIV prevalence in western
Kenya
van 't Hoog, A.H.
Publication date
2012
Link to publication
Citation for published version (APA):
van 't Hoog, A. H. (2012). Tuberculosis case finding in a population with high HIV prevalence
in western Kenya.
General rights
It is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), other than for strictly personal, individual use, unless the work is under an open content license (like Creative Commons).
Disclaimer/Complaints regulations
If you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please let the Library know, stating your reasons. In case of a legitimate complaint, the Library will make the material inaccessible and/or remove it from the website. Please Ask the Library: https://uba.uva.nl/en/contact, or a letter to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam, The Netherlands. You will be contacted as soon as possible.
High sensitivity of Chest Radiograph Reading by
Clinical Officers in a Tuberculosis Prevalence Survey
Int J Tuberc Lung Dis. 2011; 15(10): 1308–14.
A.H. van’t Hoog1,2, H.K. Meme3, H. van Deutekom4, A.M. Mithika1, C.Olunga1, F.Onyino1†, M.W. Borgdorff4,2.
1 Kenya Medical Research Institute, KEMRI/CDC Research and Public Health Collaboration, Kisumu, Kenya
2 Academic Medical Centre, University of Amsterdam, The Netherlands
3 Kenya Medical Research Institute, Centre for Respiratory Diseases Research, Nairobi, Kenya
4 Municipal Health Service, Department of Tuberculosis Control, Amsterdam, The Netherlands
SuMMARy
background
Chest radiographs (CXRs) are used in tuberculosis (TB) prevalence surveys to identify participants for bacteriological examination. Expert readers are rare in most African countries. In our survey, clinical officers scored CXRs of 19 216 participants once. We assessed to what extent missed CXR abnormalities affected our TB prevalence estimate. Methods
Two experts, a radiologist and pulmonologist independently reviewed 1031 randomly selected CXRs, mixed with films of confirmed TB cases. CXRs with disagreement on ‘any abnormality’ or ‘abnormality consistent with TB’ were jointly reviewed during a consensus panel. We compared the final expert and clinical officer classifications with bacteriologically confirmed TB as the gold standard.
Results
After the panel, 199 (19%) randomly selected CXRs were considered abnormal by both expert reviewers and another 82 (8%) by one reviewer. Agreement was good among the experts (κ 0.78, 95%CI 0.73-0.82) and moderate between the clinical officers and experts (κ range 0.50-0.62). The sensitivity of ‘any abnormality’ was 95% for the clinical officers, 83% and 81% for the respective experts. The specificities were respectively 73%, 74% and 80%. TB prevalence was underestimated by 1.5%-5.0%.
Conclusions
Acceptable CXR screening can be achieved with clinical officers. Reviewing a sample of CXRs by two experts allows an assessment of prevalence underestimation.
5
bACKGROunD
Tuberculosis (TB) prevalence surveys are recommended in selected high TB burden countries to measure progress towards the Millennium Development Goals and Stop TB Partnership targets, which recommend a 50% reduction in TB prevalence by 2015 compared to the 1990 levels.1, 2 Prevalence surveys require sample sizes ranging from 20 000 to over 200 000, to estimate the disease burden with sufficient precision and to detect significant reductions 10 years later. Most surveys aim to measure the burden of bacteriologically confirmed pulmonary TB. In settings where performing sputum culture examination on all survey participants is not feasible, a screening strategy including chest radiography (CXR) is recommended to select participants for further bacteriological examination of sputum (usually culture and smear).3, 4 The recommended practice is that the films be examined in the field by a medical officer5, as soon as the CXR has been obtained, and at a later time at central level by two independent expert readers, a radiologist and a chest physician, and by a third in case of discrepancies.3, 4
In sub-Saharan Africa, although TB rates are high, but professionals qualified to read and interpret CXRs are rare. For logistical and cost purposes it would be useful if other medical personnel could be trained to read the screening CXRs in prevalence surveys with sufficient accuracy so that expert interpretation of the presence of radiographic abnormalities consistent with present or past TB could be limited to a smaller number of CXRs.
We conducted a TB prevalence survey in a rural area in western Kenya.6 Participants were screened by CXR, symptom questionnaire and sputum microscopy. For CXR screening, clinical officers rather than medical officers were trained by a radiologist to read CXRs at the study enrolment sites and classify those as either normal or abnormal. We wanted to compare clinical officer and expert classifications, and assess to what extent CXR abnormalities that were missed by the clinical officers may have contributed to underestimation of the reported prevalence6.
METHODS
Survey
A TB prevalence survey was conducted between August 2006 and December 2007 in the Asembo (Rarieda District) and Gem District areas in western Kenya, where human immunodeficiency virus (HIV) prevalence is high.7 The final analysis included 20,566 residents aged ≥15 years from 40 sampled clusters (Figure 1). A posterior-anterior chest
radiograph (CXR) was taken of 19 216 (93%) participants, in a mobile unit at a nearby location, using a 400 mA / 40 KW X-ray machine, 35 x 43 cm films and automatic film processing. Two sputum samples for microscopy were requested from all participants, and an additional sputum sample for mycobacterial culture was requested from suspects (see definition below). The prevalence of bacteriologically confirmed TB was 6.0 per 1000 (95% confidence interval (CI) 4.6-7.4), and of smear-positive TB 2.5 per 1000 (95%CI 1.6-3.4). HIV testing was requested of prevalent TB cases. Of 101 patients tested, 52 (51%) were HIV-infected.6
Definitions and classifications
Suspect: survey participant with either a CXR abnormality and/or symptoms suggestive
of TB on interview and/or a positive sputum smear microscopy result.
Confirmed TB case: survey participant with bacteriologically confirmed pulmonary TB:
either one culture sample positive for Mycobacterium tuberculosis, or two sputum smears positive for acid-fast bacilli not explained by positive culture for nontuberculous mycobacteria.
CXR quality rating (applied by experts): 1 = optimal; 2 = suboptimal, but no technical
defect likely to impair classification; 3 = suboptimal with some technical defect but still adequate; or 4 = unreadable.8
Readers and training
Three clinical officers with an education level comparable to paramedical and nursing professions, were trained for 2 weeks by a radiologist in assessing CXRs for quality and presence of common abnormalities, and to identify which abnormal conditions required further care. The CXR of a survey participant was reviewed on site on the same day by one clinical officer blinded to participant information other than age and sex. For CXRs classified as abnormal, the clinical officer indicated whether the abnormality was pulmonary, cardiac, pleural effusion and/or ‘other’. The clinical officers rotated, and alternated CXR reading with clinical tasks on site.
Quality assurance and control
The radiographers applied standard quality procedures.9, 10 After assessment, the CXRs were transported every day to the Kenya Medical Research Institute/Centers for Disease Control and Prevention offices in Kisumu and stored in an air-conditioned
5
room. Procedures to maintain CXR and clinical officer reading quality included 2-daysupervision visits by the radiologist approximately every 6-8 weeks, and a daily ‘warm-up’ whereby the two or three clinical officers present at the field-site jointly reviewed five CXRs. To assess inter- and intra-reader variation, the clinical officers also re-read a random sample of 780 CXRs during the survey. These were selected separate from the sample for expert reading and not shown in Figure 1.
Sample for expert review
From earlier analysis of a non-random sample read by the radiologist (data not shown), we calculated that to estimate the proportion of abnormalities consistent with TB possibly missed by the clinical officers and with a 95%CI of ±10% of the point estimate, 1000 expert readings would be necessary. To allow for non-retrievable or poor quality films, we randomly selected 1050 (5.4%) films from the database with 19,299 CXRs with reports available and mixed these with the CXRs of the confirmed TB cases (Figure 1). One radiologist (expert1, HM) and a pulmonologist (expert2, HD) reviewed the CXRs in March and April 2009. Both were blinded to any participant information, the reason for selection and the other reader’s results. The expert readers used a standard CXR reading and recording system (CRRS) form11, to note the quality rating and to record details on each abnormality and whether the abnormality was considered consistent with TB. During a joint review and consensus panel both expert readers reviewed the CXRs with discrepant results together, blinded to their earlier report and in presence of a rapporteur (AH), who wrote the new results on pre-printed forms. Disagreements were discussed and recorded, including the possible clinical relevance of the abnormality for TB. Reaching consensus was the aim, but was not mandatory.
Data collection and statistical analysis
Field-readings were recorded on scannable forms (Cardiff Teleforms®, Autonomy Cardiff, San Diego, CA, USA), while the experts recorded their results in electronic forms. Results from the consensus panel were manually entered. Data were analyzed with SAS 9.1 (SAS Institute Inc., Cary, NC, USA). To assess inter-observer agreements, we calculated the percentage agreement, and the kappa (κ) with 95%CIs12: κ 0.21-0.40 represented fair agreement, 0.41-0.60 moderate, 0.61-0.80 substantial and 0.81-1.00 almost perfect agreement.13 We calculated the sensitivity and specificity of the clinical officers and experts in identifying TB cases by CXR, using the CXRs of bacteriologically confirmed TB cases as the gold standard positives, and the randomly selected CXRs of participants without TB as the negatives (derived at as shown in Figure 1).
estimating possibly missed cases
Among the 12 041 non-suspects in the survey who did not have symptoms suggestive of TB, had negative sputum smears and a normal CXR according to the clinical officers, and who were therefore not considered for sputum culture examination, some cases of prevalent TB may have been missed due to CXR abnormalities that were missed by the clinical officers. Additional missed cases due to missing data (sputum samples, CXRs) were not considered for this calculation.
We compared the clinical officer and expert reports of non-suspects among the randomly selected CXRs, and calculated the proportion of CXRs that were classifi ed as normal by the clinical officers, but as having an abnormality consistent with TB according to one of the experts or both experts. We multiplied the proportions of missed abnormalities consistent with TB by the prevalence of smear-negative, culture-positive TB among participants who did not report symptoms suggestive of TB but whose CXR was classified as ‘abnormal’ by the clinical officers, assuming that participants for whom CXR abnormalities were missed by the clinical offi cers had a TB prevalence similar to that of participants whose abnormalities were not missed. The 95%CI of the product of the two proportions was calculated according to the Delta method.14 We then applied these proportions to the 12 041 non-suspects to obtain the number of missed prevalent cases.
Ethical approval
The prevalence survey protocol was approved by the scientific and ethical steering committees of the Kenya Medical Research Institute (protocol number 943) and the institutional review board of the US Centers for Disease Control and Prevention (protocol number 4712).
RESulTS
The analysis included 1143 CXRs, 1031 were randomly selected among participants without TB, while 112 were of persons identified with bacteriologically confirmed prevalent TB (Figure 1). The distribution of CXR abnormalities, symptoms suggestive of TB, suspects and prevalent cases was similar in the random sample and the survey (Table 1). Expert 1 rated the quality of 1105 (97%) as suitable for classification: 393 (34%) as optimal, 712 (62%) as ‘2’, and 38 (3%) as ‘3’. Expert 2 rated 1115 CXRs (98%) as suitable for classification: 1024 (90%) as optimal, 91 (8%) as ‘2’, 27 (2%) as ‘3’ and one as unreadable.
5
Mi ss ing inte rv ie w 14 4 (0.7 %) * wi th CX R s (n = 83 ) Parti ci pa nts w ith c om pl ete inte rv ie w da ta inc lu de d i n an al ys is : ( n = 2 0 56 6) Spo t s pu tum for F M: 20 4 09 (99%) † O verni gh t s putu m for F M: 19 78 8 (96%) † Ches t rad io graph y: 19 21 6 (93 %) † N o S us pe ct (n = 1 3 2 66 ) S ym pt om CX R - ( n = 12 0 43 ) S ym pt om CX R m is si ng * (n = 1 18 3) Sus pe ct n = 7 346 (3 6%) CX R+ S ym pto m + ( n = 14 92 ) CX R+ S ym pto m - ( n = 38 50 ) CX R - S ym pto m + ( n = 18 31 ) CX R m is si ng S ym pto m + ( n = 1 67 ) Cul tures res ul t av ai la bl e n = 6 80 8 (9 3%) † Bac teri ol og ic al ly c onf irm ed P TB ( 12 0 wi th CX R) (n = 1 23 ) CX R+ S ym pto m + : 6 5 (36 s m +) CX R+ S ym pto m - : 4 8 (11 s m +) CX R - S ym pto m + : 7 (1 s m +) CX R m is si ng S ym pto m + : 3 (3 s m +) N o PT B (n = 2 0,4 33 ) Surv ey proc ed ures 6 CX R retri ev ed an d a va ila bl e f or an al ys is : 11 2 ( 93 %) CX Rs inc lu de d in an al ys is (n = 11 43): Parti ci pa nts w itho ut P TB (‘ ran do m ne ga tiv es ’) ( n = 10 31) Bac teri ol og ic al ly c onf irm ed c as es (n = 1 12 ) CX Rs of bac terio log ic al ly c on fir m ed cas es in b oth s ets (n = 3) *In cl ud es 1 s m ea r n eg ati ve (M G IT ) c ul tu re p os iti ve c as e † Sp utu m re su lts o f p er so ns w ho w ere u na bl e t o p ro vi de a s pu tu m s am pl e f or m ic ro sc op y or c ul tu re , a nd o f c on ta m in at ed c ul tu re s, w ere c on si de re d n eg ati ve . 1 57 pa rti ci pa nts d id n ot p ro vi de a ny s pu tu m s am pl e. 16 2 cu ltu re s am pl es w ere c on ta m in ate d on b oth m ed ia . CX R+/ - = c he st ra di og ra ph a bn orm al /n or m al (f ie ld re ad in g); Sy m pto m +/ - = Sy m pto m s su gg es tiv e / n ot su gg es tiv e of pu lm on ary tu be rc ul os is ; s m + = sm ea r p os iti ve FM = fl uo re sc en ce m ic ro sc op y; PT B = p ul m on ary tu be rc ul os is ; M G IT = M yc ob ac te ria l Grow th In di ca to r T ub e CX R s el ec tion for ex pe rt re ad ing Rand om s et CX Rs ran do m ly s el ec ted : 1 05 0 (5. 4%) Coul d n ot be retr iev ed : 11 (1%) Ex cl ud ed b ec au se of m is si ng s urv ey da ta: 5 (0. 5%) CX Rs inc lu de d: 10 34 (9 8% ) Regi st ered pa rti ci pa nts (n = 20 7 10 ) CX R s av ai lab le (n = 1 9 29 9 ) Figur e 1. Sur ve y pr ocedur es and selection of r adiogr aphs f or e xpert r eadingTable 1. Characteristics of all chest radiographs of the survey analysis and of the random
sample selected for expert review
All CXRs in survey Random sample
n % n %
Total 19 216 1 034
Abnormal by CO 5 342 27.8 283 27.4 Type of Abnormality (% of those with any abnormality)*
Pulmonary 4 793 89.7 250 88.3 Cardiac 883 16.5 50 17.7 Other 1 169 21.9 71 25.1 Pleural effusion 400 7.5 20 7.1 Read by† Clinical Officer 1 7 151 37.2 383 37.0 Abnormal 2 019 28.2 109 28.5 Clinical Officer 2 7 199 37.5 387 37.4 Abnormal 2 207 30.7 114 29.5 Clinical Officer3 4 813 25.0 261 25.2 Abnormal 1 100 22.9 60 23.0 Symptoms suggestive of TB 3 323 17.3 196 19.0 Non-suspect (Normal CXR, no symptoms suggestive of TB) 12 041 62.7 652 63.1 Bacteriologically confirmed PTB Smear Positive 48 0.2 2 0.2 Smear or Culture Positive 120 0.6 3 0.3
*More than one abnormality is possible
†Missing CO code in 53 (0.3%) of all CXRs and 3 (0.3%) of the random sample. CXR = Chest X-ray; CO = clinical officer; TB = tuberculosis; PTB = pulmonary TB
Agreement
Of the 1031 ‘random negatives’, 199 (19%) were classified after the consensus panel as abnormal by both experts, and another 82 (8%) by one of the experts. They agreed on the presence of an abnormality consistent with TB in 77 (8%; Table 2). The κ for inter-expert reader agreement on the presence of any abnormality increased from 0.52 (95%CI 0.46-0.58) before, to 0.78 (95%CI 0.73-0.82) after the consensus panel, and for abnormalities consistent with TB from 0.40 (95%CI 0.32-0.49) to 0.79 (95%CI 0.72-0.85). Of the 112 radiographs of confirmed TB cases, the κ for inter-expert reader agreement
5
on any abnormality was almost perfect before and after the consensus panel, and forabnormalities consistent with TB increased from 0.74 (95%CI 0.59-0.88) to 0.93 (95%CI 0.84-1.00; Table 2). Table 2. Agr eemen t be tw een experts on an y abnormality and abnormality consis ten t with PTB in the randomly select ed radiogr
aphs* and in the r
adiogr aphs of per sons with P TB Be for e or a fter consensus panel Expert 1 v er sus e xpert 2
Mean pair agreemen
t (%) n ormal - normal n ormal - abnormal Abnormal - normal Abnormal - abnormal κ (95% C i) Randomly select ed r adiogr aphs ( n = 1031)* An y abnormality? Be for e 690 50 134 157 0.52 (0.46-0.58) 82 Aft er 750 12 70 199 0.78 (0.73-0.82) 92 Abnormality c onsis -ten t with TB? Be for e 865 33 83 50 0.40 (0.32-0.49) 89 Aft er 917 9 28 77 0.79 (0.72-0.85) 96 Radiogr aphs of per sons with P TB ( n = 112)* An y abnormality? Be for e 17 2 4 89 0.82 (0.68-0.96) 95 Aft er 19 0 2 91 0.94 (0.86-1.00) 98 Abnormality c onsis -ten t with TB? Be for e 22 4 7 79 0.74 (0.59-0.88) 90 Aft er 25 0 3 84 0.93 (0.84-1.00) 97 *3 per sons with M. tuberculosis ha ve been ex cluded fr om the rando m sample and ar e only include d in the radiogr aphs of per sons with P TB PTB = pulmonar y TB; CI = Con fidence In ter val; TB = tuber culosis
Among the 1031 ‘random negatives’ agreement between the classifications of each of the three clinical officers and ‘any abnormality’ classified by at least one of the experts was moderate, with small differences for each individual clinical officer (Table 3).
Table 3. Agr eemen t in the random sample on an y abnormality be tw een CXR classific ations of each clinic al officer and an abnormality classified b y a t leas t one of the e xperts (a fter the c onsensus panel) Clinic al officer v er sus e xpert* Mean pa ir agr ee -men t (%) Clinical officer n n ormal - normal n ormal - abnormal Abnormal - normal Abnormal - abnormal κ (95% CI) 1 424 245 30 42 107 0.62 (0.54-0.70) 83 2 432 240 35 52 105 0.55 (0.47-0.64) 80 3 284 165 39 23 57 0.49 (0.38-0.60) 78 All 3 1031 648 103 102 178 0.50 (0.44-0.56) 80 * Expert classific ation is c onsider
ed abnormal if either one or both e
xperts classified the CXR as abnormal.
CXR = Ches t r adiogr aph; CI = Con fidence in ter val
5
Of 780 CXRs selected for quality control of the clinical officer field reading, 655 (84%)had rereading results of all three clinical officers, of which 649 (99%) could be linked to the field-reading results. The κ for inter-reader agreement between each pair of the three clinical officers ranged from 0.57 to 0.63 (Table 4).
Table 4. In ter -r eader agr eemen t be tw een the clinic al officer s and in tr a-r eader agr eemen t for each clinic al officer on an y abnormality ’. Comparison Clinic al Officer Ma tched n n ormal - normal n ormal - ab no rm al Abnormal - normal Abnormal - abnormal κ (95% CI) Mean pair agr eemen t (%) Inter-reader agreement 1 v s. 2 655 371 64 46 174 0.63 (0.57-0.69) 83 1 v s. 3 655 383 52 70 150 0.57 (0.51-0.65) 81 2 v s. 3 655 377 40 76 162 0.60 (0.54-0.67) 82 Intra-reader agreement 1 226 122 16 26 62 0.60 (0.49-0.71) 81 2 210 117 18 22 53 0.58 (0.46-0.70) 81 3 197 112 16 33 36 0.43 (0.30-0.56) 75 CI=Con fidence In ter val
Sensitivity and specificity
The sensitivity of the clinical officer readings (all three combined) in classifying CXRs of bacteriologically confirmed TB cases as abnormal was 95% (95%CI 89-98). For the expert readings prior to the consensus panel these were 83% (95%CI 75-89) and 81% (95% CI 73-88) respectively when any abnormality was considered. Considering only abnormalities consistent with TB further reduced the sensitivity, but had the highest specificity: respectively 87% (95%CI 85-89) and 92% (95%CI 90-94; Table 5). The sensitivity differed slightly by HIV status but the 95%CIs overlapped. The sensitivity of the clinical officers’ readings was 94% (95%CI 82-99) in HIV-infected and 100% (95%CI 92-100) in non-HIV-infected TB cases. The lowest expert values for any abnormality were respectively 83% (95%CI 69-92) and 80% (95%CI 66-91).
Table 5. Sensitivity and specificity of clinical officers and of each expert’s CXR classifications for
‘any abnormality’ and ‘abnormality consistent with TB’ (before consensus panel) in identifying bacteriologically confirmed TB
before consensus panel Sensitivity %
(95% CI) Specificity % (95% CI)
Any abnormality
Clinical officers (all 3) field reading 95 (89-98) 73 (70-76) Expert 1 83 (75-89) 72 (69-75) Expert 2 81 (73-88) 80 (77-82) Abnormality consistent with TB
Expert 1 77 (68-84) 87 (85-89) Expert 2 74 (65-82) 92 (90-94)
CXR = Chest radiograph; TB = tuberculosis ; CI = Confidence interval .
estimating missed cases by clinical officers
Of the 652 CXRs of non-suspects in the randomly selected CXRs, 24 (3.7%; 95%CI 2.2-5.1) were classified by at least one of the experts as having an abnormality consistent with TB and 13 (2.0%; 95% CI 0.9-3.1) by both experts, after the consensus panel. Among 3850 suspects in the total survey who did not report symptoms suggestive of TB, had negative sputum smears, and whose CXR was classified as abnormal by the clinical officers, 43 had a positive sputum culture (prevalence 1.12/100, 95%CI 0.67-1.56). Among the 12 041 non-suspects in whom CXR abnormalities might have been missed, this would
5
translate to (24/652) × (43/3 850) × 12 041 = 5.0 possibly missed cases (95%CI 3.5-6.4)considering CXRs with an abnormality consistent with TB according to at least one expert and thus a prevalence estimate under-reported by 3.9% (95%CI 2.8-5.0). Considering CXRs classified by both experts as having an abnormality consistent with TB, (13/652) × (43/3 850) × 12 041 = 2.7 prevalent cases would have been missed (95%CI 1.9-3.5), reflecting underreporting by 2.1% (95%CI 1.5-2.8). If expert classifications before the consensus panel were used, the estimated prevalence underestimation would range between 7.7% (95%CI 5.6-9.8) and 0.5% (95%CI 0.4-0.6, calculations not shown). Among the 112 CXRs of confirmed TB cases, the clinical officers classified six as normal, and missed one (0.9%) abnormality consistent with TB (according to both experts) in a smear-positive case, while one CXR showed a non-TB related abnormality according to one expert in a smear-negative case. Thirteen (11%) CXRs were classified as normal by both experts, and as abnormal by the clinical officers. Of these TB cases, five were HIV-infected, seven were non-HIV-infected two were smear-positive. Ten reported ≥1 TB symptom while three smear-negative non-HIV-infected cases did not report TB symptoms. Four CXRs were considered normal by all readers. All four TB cases had ≥2 TB symptoms, two were HIV-infected with low CD4 cell counts, and HIV status was unavailable for two.
DISCuSSIOn
The sensitivity of the clinical officers’ detection of abnormalities on CXRs of bacteriologically confirmed cases was 95%. The underestimation of prevalence due to CXR screening by clinical officers was in the range of 1.5-5.0%, based on a review of a random sample of CXRs by two experts followed by a consensus debate.
The uncertainty range of the prevalence underestimation decreased after the consensus panel, and could have been further reduced by a third expert referee reader and by a larger random sample of CXRs for expert reading. However the uncertainty range of 1.5%-5.0% is small compared to the 95% CI of the overall prevalence estimate of the survey, determined by the overall survey design, which was ±23% of the point estimate. The high sensitivity of the clinical officers in classifying CXRs of confirmed TB cases as abnormal was reassuring, as were the levels of inter-reader agreement.11, 15 The lower sensitivity of the experts is in line with other reports, also considering an expected reduced sensitivity in this high HIV prevalence population.16-19 The clinical officers were
trained to recognize common abnormalities, consider a CXR abnormal in case of doubt, and to consider all abnormalities regardless of their possible relevance for TB. This resulted in low specificity (71%), also reported in mass screening campaigns and in the clinical TB diagnostic process. 16-19 Consequently, a large proportion of survey participants requires culture, which has considerable logistical and cost implications. From our data, CXR screening by experts would have increased specificity above 90% if only those abnormalities consistent with TB were considered. However, sensitivity would decrease to below 80%, resulting in a larger underestimation of prevalence.
As the availability of physicians for TB diagnosis in clinical settings is limited in several African countries, CXRs reading by clinical officers or other cadres of health worker could also be applied to select suspects for further bacteriological examination with newer rapid, sensitive, but still expensive diagnostic test like GenXpert (Cepheid, Sunnyvale, CA, USA).20 A potential concern is confusion between sensitive reading for screening purposes with the more specific approach for clinical diagnosis of smear-negative TB in absence of additional bacteriological testing, to limit false-positive diagnoses; this limitation could, however, be addressed by training, supervision, quality assurance and operations research.
Our method of estimating missed cases has not been evaluated in a survey where all participants had sputum culture and CXRs read by multiple experts and clinical officers. We may have overestimated sensitivity and underestimated missed cases, due either to cases being missed by the study design, as smear-negative culture-positive TB cases with negative symptoms and CXR on screening were not captured by our survey, or to false-positive TB diagnosis in asymptomatic participants with an ‘over-read’ CXR abnormality. There were few of the latter, however. CXR reading by experts had 97% sensitivity (95%CI 90–100) in a survey conducted in Cape Town11, where all participants were eligible for CXR and sputum culture. Conversely, our assumption that, among participants with a negative symptom screen, the prevalence of smear-negative culture-positive TB was similar in participants whose abnormalities were missed and in those whose abnormalities were not missed may overstate the estimate of missed cases. Missed abnormalities may be less severe and less likely be associated with bacteriologically active TB than more pronounced abnormalties.21, 22 The interpretation of the reported κ statistic requires some caution, as it is influenced by the prevalence of the assessed condition.23
5
Other aspects that could be improved include reading of a standard set of CXRs byexperts, as is done in the case of the International Labour Organization.8 The clinical officers were trained with a set of CXRs prepared by the radiologist, but the experts did not review the same set. The optimal reporting form also remains to be determined. The CRRS form, developed for community surveys,11 allows for detailed reporting, which may increase specificity, but also requires more time than the simpler classification used by the clinical officers. We used conventional radiography in our survey. Digital equipment could make training and supervision easier, in particular if communication with off-site experts is available. Digital equipment also allows the evaluation of computerised systems to assess CXR abnormalities24 for their usefulness in prevalence surveys. Ultimately, a simple diagnostic test would eliminate the need for screening CXRs. However, in the interim, an acceptable level of CXR screening for ‘any abnormality’ can be achieved by clinical officers, with sufficient training and standard operating procedures to maintain quality of CXR reading. The degree to which prevalence is underestimated can be assessed by review of a sample of CXRs by two experts. We recommend this approach for TB prevalence surveys planned in similar settings and further evaluation of clinical applicability in combination with new bacteriological tests.
ACKnOWlEDGEMEnTS
We acknowledge the staff of the KEMRI/CDC Research and Public Health Collaboration, the KEMRI Centre for Respiratory Diseases Research, the study participants and the study team in general for their contributions. In particular we thank the radiographers Moses Kojo and Agneta Monari for taking the CXRs, the data staff Lazarus Odeny, Elizabeth Kendi and John Hongo and study coordinator Janet Agaya for facilitating the expert reading exercise. This paper is published with the approval of the Director, KEMRI. Sources of support
This assessment was part of the operations of the TB prevalence survey, supported by US President’s Emergency Plan for AIDS Relief and by the United States Agency for International Development through John’s Hopkins University.
Disclaimer
The views expressed in this manuscript do not necessarily reflect the official policies of the Department of Health and Human Services; nor does mention of trade names, commercial practices or organizations imply endorsement by the US government.
5
REFEREnCES
1. Glaziou P, van der Werf MJ, Onozaki I, Dye C, Borgdorff MW, Chiang CY, et al. Tuberculosis prevalence surveys: rationale and cost. Int J Tuberc Lung Dis. 2008; 12(9): 1003-8.
2. Dye C, Bassili A, Bierrenbach AL, Broekmans JF, Chadha VK, Glaziou P, et al. Measuring tuberculosis burden, trends, and the impact of control programmes. Lancet Infect Dis. 2008; 8(4): 233-43. 3. van der Werf MJ, Enarson DA, Borgdorff
MW. How to identify tuberculosis cases in a prevalence survey. Int J Tuberc Lung Dis. 2008; 12(11): 1255-60.
4. World Health Organization. Assessing tuberculosis prevalence through population-based surveys. Manila, Philippines: World Health Organization Western Pacific Region; 2007.
5. World Health Organization. Tuberculosis Prevalence Surveys: a handbook, second edition (pre-publication copy). World Health Organization 2010.
6. van’t Hoog AH, Laserson KF, Githui WA, Meme HK, Agaya JA, Odeny LO, et al. High Prevalence of Pulmonary Tuberculosis and Inadequate Case Finding in Rural Western Kenya Am J Respir Crit Care Med. 2011; epub ahead of print 14 January 2011. 7. Amornkul PN, Vandenhoudt H, Nasokho
P, Odhiambo F, Mwaengo D, Hightower A, et al. HIV prevalence and associated risk factors among individuals aged 13-34 years in Rural Western Kenya. PLoS One. 2009; 4(7): e6470.
8. ILO. Guidelines for the use of the ILO International Classification of Radiographs of Pneumoconiosis 2000 edition. Geneva: International Labour Office; 2002. 9. Chiang CY, editor. Handbook for District
Hospitals in Resource Constrained Settings on Quality Assurance of Chest Radiography: JATA and TBCAP; 2008.
10. Ellis SM, Flower C. The WHO manual of diagnostic imaging Radiographic Anatomy and Interpretation of the Chest and the Pulmonary System. Geneva: The World Health Organization in collaboration with the International Society of Radiology; 2006.
11. Den Boon S, Bateman ED, Enarson DA, Borgdorff MW, Verver S, Lombard CJ, et al. Development and evaluation of a new chest radiograph reading and recording system for epidemiological surveys of tuberculosis and lung disease. Int J Tuberc Lung Dis. 2005; 9(10): 1088-96.
12. Sim J, Wright CC. The kappa statistic in reliability studies: use, interpretation, and sample size requirements. Phys Ther. 2005; 85(3): 257-68.
13. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics. 1977; 33(1): 159-74.
14. Mackinnon DP, Lockwood CM, Williams J. Confidence Limits for the Indirect Effect: Distribution of the Product and Resampling Methods. Multivariate Behav Res. 2004; 39(1): 99.
15. Graham S, Das GK, Hidvegi RJ, Hanson R, Kosiuk J, Al ZK, et al. Chest radiograph abnormalities associated with tuberculosis: reproducibility and yield of active cases. Int J Tuberc Lung Dis. 2002; 6(2): 137-42.
16. van Cleeff MR, Kivihya-Ndugga LE, Meme H, Odhiambo JA, Klatser PR. The role and performance of chest X-ray for the diagnosis of tuberculosis: a cost-effectiveness analysis in Nairobi, Kenya. BMC Infect Dis. 2005; 5: 111.
17. Dawson R, Masuka P, Edwards DJ, Bateman ED, Bekker LG, Wood R, et al. Chest radiograph reading and recording system: evaluation for tuberculosis screening in patients with advanced HIV. Int. 2010; 14(1): 52-8.
18. Frieden T, editor. Toman’s Tuberculosis Case detection, treatment, and monitoring – questions and answers. second edition ed. Geneva: World Health Organization; 2004.
19. Kwong JS, Carignan S, Kang EY, Muller NL, FitzGerald JM. Miliary tuberculosis. Diagnostic accuracy of chest radiography. Chest. 1996; 110(2): 339-42.
20. Boehme CC, Nabeta P, Hillemann D, Nicol MP, Shenai S, Krapp F, et al. Rapid molecular detection of tuberculosis and rifampin resistance. N Engl J Med. 2010; 363(11): 1005-15.
21. Gatner EM, Burkhardt KR. Correlation of the results of X-ray and sputum culture in tuberculosis prevalence surveys. Tubercle. 1980; 61(1): 27-31.
22. Barnes PF VT, Vachon LA, Leedom JM, Overturf GD. Chest roentgenogram in pulmonary tuberculosis. New data on an old test. Chest. 1988; 94(2): 316-20. 23. Feinstein AR, Cicchetti DV. High agreement
but low kappa: I. The problems of two paradoxes. J Clin Epidemiol. 1990; 43(6): 543-9.
24. De Boo DW, Prokop M, Uffmann M, van Ginneken B, Schaefer-Prokop CM. Computer-aided detection (CAD) of lung nodules and small tumours on chest radiographs. Eur J Radiol. 2009; 72(2): 218-25. Epub 2009 Sep 10.
