When the Chest X-ray Does Not Tell the Whole Story
A Tale of Miners, Selection Bias, and the Healthy Worker Effect

MARGARET R. BECKLAKE

Departments of Medicine, Epidemiology and Biostatistics, and Occupational Health, McGill University, Montreal, PQ, Canada

	ARTICLE

TOP ARTICLE

As the subject of this essay, I have chosen two papers, published within a year of each other in the 1950s. Both addressed the question of whether occupational exposure to mining dust is a cause of COPD, but used different study approaches. The first paper supported the premise, the second did not. My essay describes how I came to investigate this question, and the reasons, in retrospect, for this very disconcerting experience for an aspirant researcher. Did this, in fact, diminish the adventure of discovery, however modest? I think notit may even have enhanced it.

By way of background, I graduated in medicine at the University of the Witwatersrand, Johannesburg, South Africa. Then, shortly after the end of World War II, I was fortunate enough to spend two years at the Postgraduate Medical School at Hammersmith Hospital in London, England, then attracting demobilized service men and women from around the world. Their experience and maturity, and their desire to catch up with, and to reshape their post-war world, created a very stimulating environment. At that time, pathology was the final word in diagnosis, autopsy the rule, and diagnosis by biopsy or tests of organ function rare. Lunch hour pathology rounds were attended by staff from all departments. Debate between the great (the pathologists, I. Doniath and Earl King, and the clinicians, Guy Scadding, John Crofton, John McMichael, Paul Wood, Sharpey-Shafer and Sheila Sherlock) was merciless. Many of us, my husband, Maurice McGregor, and myself among them, embraced this interdisciplinary ferment as a model to carry home and try to emulate.

At Hammersmith, I worked in the laboratory set up by Prof. McMichael to explore the role of lung function tests in the elucidation of a variety of clinical conditions. Our equipment was primitive, comprising a 100 L Tissot spirometer and several heavy, high resistance former basal metabolic rate spirometers. Other apparatus was built in the hospital workshop. We measured total lung capacity by helium dilution, using the original apparatus assembled by McMichael. We used the same apparatus to measure maximum breathing capacity (MBC) in L/minute, extrapolated from the maximum ventilation that the subject could sustain over 15 seconds by fast deep breathing. Then, in 1950 my husband and I returned to South Africa to appointments at our alma mater, and I set about establishing a clinical lung function laboratory, in itself an adventure. In true Hammersmith tradition, most of our equipment was built in the medical school workshop. My husband, who grew up on a farm with a good mechanical workshop, had also done a course in glass blowing (then an essential research skill), and so was able to assemble a lung volume machine to McMichael's design. To minimize the effect of high mass and instrumental resistance (important at 5,700 feet above sea level, the altitude of Johannesburg), he also designed and built a light weight spirometer and a low resistance respiratory valve made in plastic for measuring MBC.

In 1955, I was offered the challenging position of Physiologist to the Pneumoconiosis Unit of the Council for Scientific and Industrial Research, located in the Miners' Medical Bureau in Johannesburg. At that time the gold mining industry was the driving force behind the country's economy, and the Bureau was the statutory body set up in the 1920s to carry out medical examinations for compensation purposes, and for the compulsory pre-employment and annual examinations certifying a miner fit for underground work. What a remarkable research opportunity this offered was pointed out to me in 1955 when I visited Dr. John Gilson, Director of the British MRC Pneumoconiosis Research Unit in Cardiff, Wales. When I told him that in the course of a year all active miners would pass through the Bureau for their annual medical examination, he pointed out that this allowed me to sample from their number at random, an advantage never enjoyed by his Unit. This casual conversation was my first lesson in epidemiology, namely that case series could only quantitate health problems if they were related to a denominator, the number of persons at risk.

I first set out to describe the lung function changes in Witwatersrand gold miners with and without roentgenological changes of silicosis (Becklake MR, Dupreez L, Lutz W. Lung function in silicosis of the Witwatersrand gold miner [American Review of Tuberculosis and Lung Disease 1958:77:400- 412]). My co-authors were the first technician I had trained in this laboratory and a professor of mathematical statistics at the University of the Witwatersrand. The paper also addressed, albeit indirectly, an issue of major concern to the miners at that time. Traditionally, compensation had only been awarded on the basis of roentgenological evidence of pneumoconiosis. However, the miners were convinced that the chest X-ray did not tell the whole story. There were men among their number without roentgenographic evidence of silicosis who, although too breathless to work, received no compensation for what the miners believed was a work-related disease. By contrast, there were men with roentgenographic evidence of silicosis who received compensation even though they felt well and were able to continue working. Our study compared random samples of miners with and without roentgenographic silicosis, matched for age, with a minimum of 7 years underground service. Though modest in numbers (altogether only 40 miners were studied), each man underwent comprehensive testing (including measurements of lung volumes and mechanics, MBC, and exercise testing with ear oximetry). Because we found no single physiological test that distinguished the miners with and without roentgenographic evidence of silicosis, Lutz used a statistical approach, discriminant analysis, to construct a comprehensive physiologic score, based on the eight physiologic measurements, which when combined into an index, best separated men with and without roentogenological silicosis. This innovative approach had been used by Gilson in studies of Welsh coal miners. We concluded that lung function scores were shown to correlate with the length of rock-breaking service, irrespective of the roentgenographic findings. This suggests that disability due to mining may be found in men in whom there is no frank roentgenographic evidence of dust disease. Our findings were thus consistent with the miners' premise and our study hypothesis.

Our second study, already in planning by the time the data for the first paper were gathered, had as its objective to strengthen the evidence on an obviously controversial issue. In this study, Saul Zwi (a Research Fellow in the Pneumoconiosis Research Unit) and I used what we thought was a stronger study design. We compared the lung function of the same 20 miners without silicosis (reported in the first paper) with that of a representative sample of 20 non-miners (railway workers) tested in our laboratory using identical methodology (Zwi S, Becklake MR. Respiratory function of Witwatersrand gold-miners [British Journal of Industrial Medicine 1959,15: 258- 261]). However, to our bewilderment, in this cross-sectional study, we found no evidence, clinical or physiologic, of an increased respiratory disability among the mining group. Reluctantly, and in retrospect I think naively, I accepted these results as proof that the study hypothesis was untenable, and that this line of research was a dead end. Remember that in those days smoking had yet to be recognized as a cause of both lung cancer and chronic bronchitis or COPD (eloquently evoked by Sir Richard Doll in his essay in the How It Really Happened series). Then, once the link between smoking and COPD was recognized, and given the universality of the smoking habit at the time, COPD became defined as a disease of smoking and no one (except the miners) was prepared to recognize other inhaled pollutants as a cause of COPD.

In 1957 our family moved to Montreal, Canada, and my husband and I joined the CardioRespiratory Service of the Royal Victoria Hospital, McGill University, where for several years my research was clinically oriented. But in 1967, I was asked to develop a lung function laboratory to complement Dr. Corbett McDonald's epidemiological research program among Quebec chrysotile miners and millers. These studies provided indirect evidence for the Witwatersrand miners' premise in that obstructive lung function profiles (suggesting COPD) were not only common in these Quebec workers but that they were also related to their dust exposure, and of course to their smoking. Another ten years elapsed before an unexpected turn of events took me back to Johannesburg for a sabbatic leave in 1984/5. Here, with colleagues from the National Centre for Occupational Health, I carried out a case control study exploiting their remarkable autopsy database on miners maintained since 1975. Les Irwig of the Institute of Biostatistics used the now more sophisticated statistical methods available for multivariable analysis to show that both high dust exposure and smoking were strong independent predictors of the presence of emphysema at autopsy.

Meanwhile, the reason our second study and other cross-sectional studies had failed to show a relationship between mining exposure and COPD was slowly taking shape in my mind. In a cross-sectional study of young Montrealers, we had found that those who reported smoking had on average higher (not lower) values for FVC and FEV₁ than those who did not, in other words a healthy smoker effect. Could it be that, like the young Montreal smokers, miners with higher values for FVC and FEV₁ also self-selected into dusty jobs, in other words a healthy worker effect? The answer was in the literature. In a non-systematic review with Umesh Lalloo from the University of Natal, Durban, we found evidence for a healthy worker and a healthy smoker effect, often together in dust-exposed populations although, for the most part, the authors of these papers had commented on neither effect. Furthermore, on re-reading our 1959 paper in preparation for this essay, I was chagrined to discover that I too had missed evidence of a healthy worker effect in our own data. The average vital capacity of the miners aged 31 to 40 years was higher than that of the railway workers of the same age, (5.29 L versus 4.87 L). The data would probably also have shown a healthy smoker effect had the smoking habits been recorded, but this was not part of the respiratory history in 1959.

A second reason why the causal relationship between occupational exposure and COPD was so slow in gaining acceptance is that, before the introduction of modern statistical methods for multivariable analyses, establishing causality for diseases of multifactorial etiology such as COPD was not easy. Only the use of what was, for that time, sophisticated statistical modeling by Lutz, made successful identification of the relationship between mining exposure and COPD possible in our 1958 paper. If my story has a lesson, it is not to despair when your research appears to have come to a dead end. If you wait long enough, time, circumstance, other studies, and some hard thinking aided by new technology may eventually show that your research was on the right track. As for the miners, their original premise did eventually achieve scientific recognition.

	Footnotes

Correspondence and requests for reprints should be addressed to Margaret R. Becklake M.D., Respiratory Epidemiology Unit, McGill University, 1110 Pine Avenue West, Montreal, PQ, Canada H2W 1S6. E-mail: margaret.becklake{at}mcgill.ca