INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Gold miners working in South Africa are screened with miniradiographs (100 × 100 mm) as part of an annual occupational fitness examination. The radiographs, which are mandatory, are primarily intended to detect tuberculosis (TB) but are also assessed for the presence of silicosis. Both TB and higher grades of silicosis are compensatable occupational diseases of gold miners. Since 1997, mining companies have been required to screen their employees with a standard sized radiograph once every 3 yr during employment and on leaving the industry. Miniradiographs were previously the sole screening tool in otherwise fit men. One further method used for detecting and compensating occupational diseases in South African gold miners is autopsy examination of the lungs, which is routinely requested on the death of miners during employment.

Despite the use of miniradiographs for compensation screening and epidemiological studies of the prevalence and effects of silicosis (1), the accuracy with which they are able to detect silicosis has not previously been investigated. This study is a retrospective comparison of radiological and pathological silicosis grades, using annual screening miniradiographs and routine autopsy pathology findings. The main aim was to investigate the sensitivity and specificity of silicosis grading of miniradiographs, using autopsy findings as the "gold standard." The ability to detect silicosis in the presence of post-tuberculous scarring, and the prevalence of radiological and pathological silicosis in different age groups were also investigated.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Identification of Study Subjects

Miners who had died between January 1996 and July 1997 while working for a single company in the Free State Province, South Africa were identified from payroll records, a book containing a list of deaths occurring at the hospital used by mining employees, and records of the histopathology department of the National Center for Occupational Health in Johannesburg. Packets containing the miniradiographs, autopsy report, and employment record for each dead man were stored together. The files were located and the autopsy findings, name of the pathologist, and employment records were entered onto a database.

Pathology

The age, race, employment history, and death certificate opinion as to cause of death were available to the pathologist at the time of autopsy. The number of discrete palpable nodules in the two lungs were routinely recorded as none, occasional (1 to 4 nodules), few (5 to 14), moderate (15 to 30), or marked (more than 30). Sections were taken from each lobe of the lung and any palpable nodules and examined microscopically for the presence of: dust, silicotic nodules, progressive massive fibrosis, small foci of irregular fibrosis associated with silica dust (early silicotic fibrosis), active TB, and other pulmonary diseases. Two of the seven pathologists had used a different grading system from the others, which only recorded the presence of silicosis for compensatable cases (15 or more nodules). All of the men who had been examined by either of these two pathologists were excluded from the study, regardless of the grade given.

Radiology

Miniradiographs, taken annually for screening purposes, were screened and a film taken within 2 yr of death was selected for each man where possible. Serial radiographs were compared so that films of unusually low technical quality or with new focal pathology suggestive of active infection or neoplastic disease could be avoided. An earlier film was selected instead in such cases. In some cases, no recent radiographs were present in the packet, and the most recent available film was then selected. The dates of selected films were recorded and films were independently graded for radiological silicosis by two readers without reference to personal identifiers, age, employment history, cause of death, or autopsy findings. Radiographs were taken using Siemens generators, Odelta 100 XV11S cameras, and an automatic exposure mechanism.

The International Labor Office (ILO) system for grading pneumoconiosis (4) was followed. This system, intended for use with standard sized radiographs, classifies films according to the presence and number of nodules, with grade 0 being none, and grades 1 to 3 representing increasing categories of nodule profusion, so that vascular markings are obscured by grade 3 silicosis. To allow for uncertainty in radiological interpretation, a second possible grade is also given, with the two grades separated by a slash. For example, 0/1 is an intermediate grade where the reader has graded the film as normal (0), but considers that grade 1 changes cannot be excluded (/1). In the current study, 0/1 and 1/0 were the only intermediate grades used. Standard sized reference radiographs are provided for each grade, and were available to the study readers. Decisions on final grades were reached by consensus between the two readers when initial grades differed. One reader (P.C.H.) had been screening miniradiographs and referring men for silicosis compensation since 1974, and both had previously graded several thousand miniradiographs together using the same system. Radiographs were also scored for post-tuberculous scarring, defined as a chronic focal and asymmetrical fibrosis with or without cavitation involving most of at least one of the upper lobes.

Cause of Death

The most likely cause of death was determined from hospital records and autopsy findings but without reference to the pathological silicosis grade. TB was given as the cause of death when both stated as such on the death certificate and present at autopsy, or where active TB was found at autopsy and was considered by the pathologist to be sufficiently extensive as to be the likely cause of death.

Statistical Methods

Data were analyzed with STATA 5.0 software (STATA Corporation, TX). A kappa score, which evaluates the degree of agreement above that expected by chance, was used to assess interrater agreement for the initial radiological grades. A weighted kappa score, which takes account of close as well as perfect agreement, was also calculated. Cell weights for the weighted kappa score were allocated as w_ij = 1  |i  j|/(g  1), where i and j refer to the row and column number respectively for the two raters and g was the number of grades. The sensitivity and specificity, and positive and negative predictive values, of the radiological diagnosis were calculated for condensed 2 by 2 tables using different grades to define radiological and pathological silicosis, and taking pathological silicosis as the gold standard. The association between age and silicosis was tested for significance using the chi-square test.

The study was approved by the ethics committee of the Ernest Oppenheimer Hospital, Welkom, South Africa.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Subject Characteristics

The process of selection, starting with all men initially considered, is summarized in Figure 1. Only men who had been examined by one of the five pathologists using the standard grading system were eligible for inclusion. The age, employment duration, cause of death, and interval between radiological film and death for the 241 men (50% of all those initially considered) who were included in the final analysis are shown in Table 1. All were black African men, and all had worked in dusty jobs.

View larger version (38K): [in this window] [in a new window]   Figure 1.   Selection of study subjects.

Cause of Death

Trauma was the cause of death in 84 men (35%), and TB in 40 (17%). A total of 177 men (73%) had a human immunodeficiency virus (HIV) test before their death, which had been positive in 107 men (44% of all subjects and 60% of those tested). HIV prevalence among men dying from TB was 73%. There were 67 deaths (28% of all subjects) from causes meeting case definitions for acquired immunodeficiency syndrome (AIDS), of which 32 (48% of AIDS deaths) were from cryptococcal disease and 29 (43% of AIDS deaths) were from HIV-associated TB. An additional 11 HIV-positive men died from bacterial pneumonia, increasing the proportion of deaths from HIV-associated causes to 32%. There was only one case of pulmonary Kaposi's sarcoma. This was associated with new focal radiological changes in the year before death, but not in the previous annual film that was used for the study.

Age-specific Prevalence of Radiological and Pathological Silicosis

The proportions of men with any grade of radiological and autopsy silicosis were similar. For both radiology and pathology the prevalence of silicosis increased with increasing age, as shown in Figure 2. There were no cases of pathological silicosis complicated by progressive massive fibrosis. Age was used as a proxy for employment duration because data on employment were missing for 19 men, and thought to be incomplete for some men, who may have worked for other mining companies. Most miners start work when in their early twenties, and so age and employment duration are closely correlated. The associations between age and radiological and pathological silicosis were significant (p < 0.001 in both cases for data shown in Figure 2).

View larger version (36K): [in this window] [in a new window]   View larger version (33K): [in this window] [in a new window]   Figure 2.   The prevalence of different grades of radiological (top panel ) and pathological (bottom panel ) silicosis by age at death. *Includes men with silicotic fibrosis.

Interrater Agreement for Initial Radiological Grades

The interrater agreement for initial silicosis grades given by the two radiological readers is shown in Table 2. For the 6 by 6 table, the kappa score was 0.46, and the weighted kappa was 0.71. 

Intertest Agreement for Pathology and Consensus Radiology Grades

The agreement between radiology and pathology grades is shown in Table 3. Using any radiological abnormality (grade 0/1 or above) to define radiological silicosis, there was agreement that silicosis existed in 54 men, agreement that no silicosis was present in 129, and discordant findings in 58 (32 abnormal radiology, normal pathology; 26 abnormal pathology, normal radiology). The sensitivity for this 2 by 2 comparison was 67.5%; the specificity was 80% and the positive and negative predictive values were 63% and 83%, respectively. Of the 26 men with abnormal pathology and normal radiology, eight (31%) had an interval of more than 2 yr between the radiograph and death, similar to the proportion of all study subjects (28%). Limiting the analysis to men with a short interval between radiograph and death gave very similar results to those obtained when all men were included (data not shown).

Overall pathological concordance with early grades of radiological silicosis (0/1 and 1/0) was poor, as was radiological concordance with early grades of pathological silicosis (silicotic fibrosis and occasional nodules). Of 34 men with a radiological grade of 0/1 and 19 men with a grade of 1/0, only 14 (41%) and 9 (47%) respectively had silicosis detected at autopsy. Similarly, only 23 (51%) of 45 men with silicotic fibrosis or occasional nodules at autopsy were detected as having radiological abnormalities. Undetected pathological abnormalities were present in 26 (17%) of the 155 men with a radiological grade of 0/0. The majority of undetected cases had early pathological disease (22 men with silicotic fibrosis or less than 5 nodules, two men with 5 to 14 nodules, and two men with 15 to 30 nodules). Concordance was better for the higher radiological and pathological grades.

Effect of Using Different Radiological Grades to Define Silicosis

Using different radiological grades to define silicosis had a marked effect on the sensitivity, specificity, and positive and negative predictive values. Increasing the radiological grade used resulted in higher specificity and positive predictive values, but lower sensitivity and negative predictive values.

For detection of any degree of pathological silicosis, the specificity and positive predictive values for different radiological definitions were 99% and 94% when radiological silicosis was defined as grades 1/1 and above, and 93% and 77% when radiological disease was defined as grade 1/0 and above. This compares with specificity and positive predictive values of 80% and 63% for grades 0/1 and above. Sensitivity and negative predictive values were, respectively: 39% and 76% for grades 1/1 and above; 50% and 79% for grades 1/0 and above; and 67.5% and 83% for grades 0/1 and above.

Effect of Increasing the Pathological Grade Used to Define Silicosis

Table 4 shows the sensitivity and specificity, and positive and negative predictive values, when radiological silicosis was defined by different grades and the pathological grade used to define silicosis was increased to that of 5 nodules or more. Although sensitivity and negative predictive values were improved by using the higher pathological grade to define silicosis, the specificity and positive predictive values at each radiological definition of silicosis fell. The highest specificity (96%) and positive predictive values (76%) were found when grades of 1/1 and above were used to define radiological silicosis. Sensitivity was 71% and negative predictive value was 95%. Lowering the radiological grade used to define silicosis increased the sensitivity to a maximum of 89% for the grade 0/1, but at the expense of specificity and positive predictive value, which decreased to 73% and 36%, respectively.

Post-tuberculous TB scarring did not appear to worsen the ability to detect concurrent silicosis; the data from men with post-tuberculous TB scarring using grades of 1/1 and above to define radiological silicosis are shown in Table 4.

Pathological Correlation with Individual Radiological Grades

As a result of initial disagreement, consensus grades differed from the individual reader grades. In reaching consensus, Reader 1 changed his opinion on 40 films. Similarly Reader 2 changed her opinion on 34 films. Despite the number of initial disagreements, there was little difference between the sensitivity and specificity of the two readers or their consensus grade. Sensitivity of the consensus grade was that of the most sensitive reader or slightly above, but specificity was not consistently altered. When 5 nodules or more was used to define pathological silicosis, the sensitivity values for the different radiological definitions for Reader 1, Reader 2, and their consensus grades were, respectively: 69%, 66%, and 71% for grades 1/1 and above; 71%, 74%, and 74% for grades 1/0 and above; and 83%, 83%, and 89% for grades 0/1 and above. Similarly, the respective specificity values were: 93%, 97%, and 96% for grades 1/1 and above; 88%, 90%, and 87% for grades 1/0 and above; and 74%, 73%, and 73% for grades 0/1 and above.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The findings from this study show that it is possible to detect higher grades of pathologically confirmed silicosis with reasonable accuracy from miniradiographs, even in the presence of TB scarring. Most men with any degree of silicosis at autopsy were identified as having abnormal radiology, but concordance was poor for earlier radiological and pathological grades.

Using different pathological and radiological grades to define silicosis had marked effects on the specificity, sensitivity, and positive and negative predictive values of radiological diagnosis. This was because of a combination of both false-negative radiological readings, mainly affecting men with early pathological silicosis, and false-positive radiological readings, mainly affecting the ILO grades of 0/1 and 1/0. Thus, the radiographs of a high proportion of men with early pathological silicosis (less than 5 palpable nodules) were graded as normal. Because of this, definitions of pathological silicosis that included these early grades led to considerably lower sensitivity and negative predictive values for the radiological diagnosis than when 5 or more nodules was used to define pathological disease. Similarly, definition of radiological silicosis that included ILO grades 0/1 and 1/0 had considerably lower specificity and positive predictive values than when ILO grade 1/1 was defined as the lowest silicotic grade.

Using a pathological grade of 5 nodules or more and a radiological grade of ILO 1/1 and above to define silicosis resulted in sensitivity and specificity values of 71% and 96%, and positive and negative predictive values of 76% and 95%. The positive and negative predictive values should be interpreted with some caution, because these values are highly dependent on the prevalence of disease in the population studied, which will vary between work forces and age groups within the same work force. In this context, the current study should not be interpreted as a study of silicosis prevalence, because the risk of death is age-dependent and potentially increased by the presence of silicosis.

Two factors that could have affected the sensitivity and specificity results are, first, the relatively long time interval between the dates of the radiograph and death in a proportion of subjects and, second, the potential for HIV-associated pulmonary disease to have been misinterpreted as radiological silicosis. Limiting the analysis to men with an interval between radiograph and death of less than 2 yr, however, did not alter the sensitivity results. The potential impact of HIV-associated disease was minimized by screening serial radiographs so that films with new focal changes suggestive of infection or neoplastic disease were not used. The main three HIV-associated pulmonary diseases that were found at autopsy were TB, cryptococcal pneumonia, and bacterial pneumonia. These are unlikely to have been mistaken for silicosis, given the above precaution of screening serial films.

The radiological diagnosis of silicosis is known to be imprecise even from standard size radiographs. Previous studies have shown a high degree of interobserver variation, both for distinguishing normal from abnormal films and for grades given to abnormal films (5). Various strategies have been developed to minimize the impact of this effect, such as use of composite or majority grades from several readers (6, 7, 9), although the difference between final and individual grades has not previously been investigated against an independent test. The gain in accuracy from using two readers in the current study was minimal; there was a slight improvement in sensitivity, but no consistent change in specificity, in their consensus compared with individual grades. This may have been affected by previous work that the two readers had carried out together, in that they may have had a closer agreement on interpretation of difficult radiographs and grade transition points than would otherwise have been the case.

Miniradiographs have been used to investigate silicosis as a risk factor for TB in South African miners in epidemiological studies that involved the same work force, type of radiograph, readers, and consensus system as this current study (2, 3). A significant association with TB was found for radiological silicosis of grades 1/0 and above, with a nonsignificant increase in risk for men with grade 0/1 (2, 3). In addition, the incidence of TB was unusually high in older miners with normal radiographs and no risk factors for TB besides their occupation (2). The current findings that radiology is insensitive for detecting early silicosis, and that there is considerable misclassification for ILO grades 0/1 and 1/0, imply that the strength of the effects of early silicosis on risk of TB will have been substantially underestimated. This supports the hypothesis that silica exposure is a considerably more potent risk factor for TB than has previously been recognized, and one that is likely to be exerting a powerful effect at the population level in work forces where exposure to silica is universal. The high susceptibility of South African miners to TB that results from exposure to silica is now being exacerbated by the effects of the HIV epidemic (2). The current finding that 28% of study subjects died from AIDS is an indication of the severity of the impact of the HIV epidemic on the health of South African mine workers.

There has been one previous systematic study involving former South African gold miners that compared radiological and pathological findings, and used the ILO classification system (5). The subjects were retired white mine workers and the radiographs were standard sized films from voluntarily attended medical examinations, rather than mandatory miniradiographs. Radiographs were graded for the purposes of the study by three readers. The pathology data were autopsy findings from the same center as used in the current study. The results, however, were notably different in that the radiological sensitivity of the current study far exceeds that found previously, despite the use of smaller radiographs. For the same pathological definition, sensitivities for radiological grades 1/1 and above were 39%, 37%, and 24% for the previous three readers compared with 69% and 66% for the current two readers. As in the current study, sensitivity increased and specificity decreased when lower ILO grades were used to define radiological silicosis, but for each case the sensitivity remained well below that found in the current study, despite a similar specificity.

Possible explanations for the difference between the two study findings include: differences in radiographic quality; age or age-related differences in the radiological manifestations of silicosis; and the potential for awareness of the poor sensitivity found in the previous study to have influenced radiological interpretation by the current readers. Also, although autopsy reports from the same source were used for both studies, these data were less than ideal because grading was carried out by a single observer for compensation rather than research purposes. It is therefore possible that underreporting of noncompensatable silicosis may have occurred. This could have affected the two studies to a different extent because the earliest compensatable grade for subjects in the previous study was lower (any number of nodules) than for subjects in the current study (15 nodules or more). The finding that sensitivity was markedly different for all radiographic grades, with little difference in specificity, however, argues against explanations that depend on different interpretations of silicosis severity (either radiological or pathological) and favors a more universal factor such as radiographic technique, age, and race.

Most studies of the epidemiology and effects of silicosis rely exclusive on radiological diagnosis. The ILO grading system for pneumoconiosis was introduced to facilitate comparison between different centers by standardizing radiological grading. However, interobserver variation in ILO grading is high (5) and there will be other factors, such as different radiograph techniques, affecting the generalizability of results. The degree of difference between the findings of the current study and those of the previous study discussed previously is disappointing, and indicates that variation attributable to these factors can be considerable. This questions the validity of comparing studies in which radiological grading is critical, such as analyses of the dose-response relationship between silica exposure and silicosis, and may in part explain why results from different centers have differed so greatly (10, 11).

Silicosis research has been greatly hampered by the combination of uncertainties concerning the interpretation of radiological changes in men exposed to silica dust, and the difficulties of calculating accurate individual exposures (10). Because of these problems there is still doubt about the maximum concentration of silica to which workers can be exposed for the duration of their working lives without undue risk of developing silicosis (10, 13). The availability of large numbers of paired radiographs and autopsy specimens from South African miners is a unique resource that allows a more objective assessment of radiological grading than is otherwise possible. The current study was limited to miniradiographs, but the recent changes to South African legislation provide the opportunity to use standard sized radiographs in future. This could provide the basis for a much needed critical reassessment of factors affecting radiological interpretation, potentially leading to improved standardization.