| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Commentary: A debt of gratitude to J. Alison Glover
The Dartmouth Institute for Health Policy and Clinical Practice, 35 Centerra Parkway, Suite 300, Lebanon, NH 03766, USA.
E-mail: John.wennberg{at}dartmouth.edu
Accepted 4 December 2007
A surgical intervention is a dramatic event in the life of the patient. Few surgeons are hesitant believers in the efficacy of the operations they perform, nor do they doubt their clinical necessity. Most patients are convinced that the benefits of surgery exceed the risks by a wide margin. Yet in the face of such certainty and conviction, it is remarkable how much medical opinion on the need for surgery can vary from place to place, and how much uncertainty there is concerning what the risks and benefits are and what treatments patients actually want.
The work of J. Alison Glover has been a guiding light to many investigators interested in knowing why surgical practice varies so much from one community to another. He was the first to understand the significance of medical opinion in influencing the rate of surgery. He uncovered a more than four-fold variation in the incidence of tonsillectomy among British school districts.1 He took pains to rule out the possibility that the differences in rates (numbers of procedures per 10 000 school children) were explained by factors on the demand side of the utilization equation. Because the child health services were provided free as part of attending school, economic factors and access to care were not a factor. He could find no evidence of an association between surgery rates and ‘any impersonal factor’ predictive of illness such as overcrowding, poverty, bad housing or climate.
Although Glover gave a plausible account for the lack of importance of illness, his approach was based on an argument of exclusion. Unbeknownst to Glover, the natural experiment that was needed to demonstrate directly that patient characteristics and illness rates did not drive the tonsillectomy rates had already occurred. In 1934, the American Child Health Association, like many volunteer health associations of the day, viewed tonsillectomy as a public health good. The association wanted to make certain that no school child in New York City who needed a tonsillectomy had been overlooked. To find out the unmet need, they performed a sophisticated study that, ironically, not only provided direct evidence for the extraordinary variability in professional judgment, but also led to considerable doubt about the notion of unmet need.
The research design required random sampling of 1000 school children. Upon examination, 60% were found to have already undergone tonsillectomy. The remaining 40% were examined by the school physicians, who selected 45% in need of an operation. To make sure that no one in need of a tonsillectomy was left out, the Association arranged for the children not selected for tonsillectomy to be re-examined by another group of physicians. Perhaps to everyone's surprise, the second wave of physicians recommended that 40% of these have the operation. Still not content that unmet need had been adequately detected, the Association then arranged yet a third examination of the twice-rejected children by another group of physicians. On the third try, the physicians produced recommendations for the operation on 44% of the children. By the end of the three-examination process, only 65 children of the original 1000 had not been recommended for tonsillectomy.2
Glover took advantage of another natural experiment to further document the importance of medical opinion. He monitored the changes in tonsillectomy rates that followed a change in the school health officer—the physician responsible for diagnosis and referral of children for tonsillectomy. His most famous case involved Dr Garrow, who replaced an unnamed predecessor as school health officer in the Hornsey Borough school district. Following Garrow's recruitment, the rates of tonsillectomy fell dramatically and remained at a rate 
10% of that previously experienced.3
In Vermont, where I and Alan Gittelsohn had developed a population-based method for monitoring variation in health care delivery among local medical areas, we noted similar variation as well as a rapid drop in risk of tonsillectomy in one region. The drop, we subsequently learned, occurred because of conscious change in treatment policy among local physicians, not because of changes in personnel.4 The change was initiated by a report we provided to the Vermont State Medical Society detailing the rates for each Vermont area. Dr Roy Buttles, the Society's president, circulated the information among Vermont physicians. Prior to the release of the report, the rate in the Morrisville area was such that the pre-adult risk of tonsillectomy was 65%. Upon learning of the high rate in their area, two physicians, Dr Lewis Blowers, a general surgeon, and Robert Parker, a pediatrician, undertook an active review process that led to the rapid decline of their use of the procedure to 10% of children, a change in rate similar to that documented by Glover.
The most direct clinical evidence supporting the medical opinion hypothesis was provided some 20 years after Glover's original report. Compelled by the logic of Glover's explanation, Michael Bloor, who later became professor of sociology at Aberdeen University, and his colleagues G. A. Venters and M. L. Samphier set out to document the role of medical opinion by directly observing physicians as they interacted with their patients to reach their treatment decision.5 They took as their laboratory two health districts in Scotland with substantially different tonsillectomy rates. The high rate district, they showed, had higher rates of referral from general practitioners to surgeons as well as higher rates for performing surgery on referred patients, but with considerable differences among individual physicians within each area. With the permission of the surgeons, they then sat in on the clinical sessions to observe and document variation in the decision rules and practice patterns and to correlate the differences with the physician's propensity to operate.
The physicians' practice styles were found to differ on the basis of the relative importance they gave to the history vs the physical examination. Surgeons with a high proclivity to operate tended to stress the importance of the physical examination. For example, one surgeon felt that three physical signs—purulent material in the tonsil, reddened anterior pillars and palpable cervical nodes—were decisive, and his rule of thumb was to operate on any child with two or more of these signs. Among low operators, the reverse was the case; much more stress was put on the history. Practice style differences were also found in the details elicited from the history and the interpretation of the meaning of referral. One high rate surgeon felt that the mere fact of referral implied an extensive history of morbidity, and in most cases his decision making strategy combined an examination of the child with a simple check on his assumption of morbidity by asking the parent if the child suffered a ‘lot of trouble’. In direct contrast, a low rate physician acted as an independent assessor seeking to reconstruct the clinical history in highly specific terms. Low rate surgeons were also characterized as having a high tendency to wait and see after deciding that an operation was not an immediate necessity and that an intermediate approach (such as antibiotic treatment) could be used.
Bloor's detailed observations of the actual behaviour of clinicians in reaching their tonsillectomy decisions convinced him that Glover was right. To use Bloor's words, the differences in rates between the regions: "can be attributed to differences between specialists in their assessment practices: local differences in the nature of specialist practice ‘create’ local differences in surgical incidence ... (The findings) amount to a detailed vindication of Glover's conviction that variations in the incidence of surgery are largely the product of medical opinion rather than the product of the differential distribution of morbidity."6
 |
Scientific uncertainty and the pattern of variation |
|---|
 Top Scientific uncertainty and the... The surgical signature...References |
|---|
Glover followed yet another line of reasoning in making his case for medical opinion. He documented the professional controversies concerning the value of tonsillectomy—the disputes over medical theory and fact—that made it impossible for the profession to reach consensus on the ‘best practice’ for dealing with chronic tonsillitis. Tonsillectomy was not a life-saving, emergency operation. Some physicians, Glover among them, were sceptical of its value as a preventive measure, believing the operation to be effective only in properly selected cases with demonstrated morbidity (such as those with ‘frequently repeated attacks of acute tonsillitis which cannot be explained by extraneous infections’). But in Glover's view, this restricted view of the procedure's value provided no justification for the use of tonsillectomy as a public health intervention where the substantial majority of children were subjected to tonsillectomy, as was the case in some school districts (and as was still the case in the 1960's in Morrisville, Vermont). In addition to the highly dubious belief that school children had a high risk of developing serious illness in the future merely because they had a tonsil, it was Glover's judgment that such widespread use of tonsillectomy as a public health strategy denied the ‘probability that the tonsil serves some useful purpose, its tendency for spontaneous involution, and the success of non-operative methods of treatment that are often likely overlooked ...’ Moreover, it ignored the fact that tonsillectomy was a risky procedure; he reported that 424 British school children had died following the surgery between 1931–35.
In the 1970's, we were able to extend Glover's epidemiology of surgical practice to other common surgical procedures and to add medical areas in Maine and Rhode Island to the database. We noted that the pattern of variation for common surgical procedures varied according to the procedure, and that the pattern for a given procedure was similar from state to state. The surgical repair of an inguinal hernia showed little variation; surgeries to remove gallstones and the appendix were moderately variable; hysterectomy (removal of the uterus) and prostatectomy for non-cancerous enlargement of the prostate were quite variable from area to area; and tonsillectomy was the most variable. In the tradition of Glover, we undertook a literature review to document controversies for these procedures, seeking to make a link between the degree of controversy as manifested in the literature and the degree of variation in the procedure. Through the literature review we found a striking correlation between the degree of variation in the rates for a procedure and the extent of professional disagreement and controversy on the preferred ways for treating the underlying condition.7,8
|
The surgical signature phenomenon |
|---|
|
TopScientific uncertainty and the...The surgical signature...References |
|---|
Small area variation studies have sometimes been characterized as the ‘view from 30 000 feet’—a snapshot of comparative performance at a highly aggregated level. Yet as Glover showed, the variations in surgery have their origins in the microcosm of the doctor–patient relationship—in the one-on-one exchange between an individual physician and a patient. In the case of Glover's epidemiology, there was a one-to-one correspondence between the principle decision maker (the school health officer) and the population served (the children attending the health officer's school), such that the risks for surgery of that group are directly measured in the population-based rates.
In the United States and other countries in which a natural relationship does not exist between administrative boundaries and the providers of care, small area analysis has depended on patient origin studies at the postal code or minor civil division level of aggregation to create larger markets such as a hospital service area. In these situations, the surgery in an area is commonly done by a small group of physicians and because the rates in an area represent the weighted average of the contributing physicians, the variations in rates represent an underestimate of the ‘true’ variation that results from clinical decision making. Nevertheless, as we have learned, the small area model shows wide variation in rates with, as Glover noted, ‘the extremes often being in adjacent areas’. The variations also tend to persist over time: at least over an intermediate period of time, the workforce located in a region tends to be stable; the clinical rules of thumb clinicians use remain unchallenged; and the workload commitments of the individual physicians remain more or less the same. The resulting sharp differences in the risk of surgery, attributable primarily to variation in local medical opinion, give rise to what we came to call the surgical signature, so named because one can recognize a community through the characteristic pattern of the risk for surgery.
Figure 1, adapted from an article in the Journal of the Maine Medical Association,9 shows the surgical signature in the mid-1970s for the five most populated hospital service areas in Maine. For each area, the relative rates for five procedures are displayed. The rate at which a specific procedure was performed in these neighbouring hospital service areas varies markedly. In each area, a different procedure was performed most often. In Portland, prostatectomy for an enlarged prostate was the most preferred operation, exceeding the state average by 40% and the lowest region (Bangor) by 
100%. In Lewiston, hysterectomy exceeded the state average by 60% and the lowest area (neighbouring Augusta) by 125%. Augusta, however, excelled in surgery for varicose veins, with rates almost 90% higher than the state average and 2.5 times that of the Portland area. Surgeons practicing in the Waterville area were particularly prone to operate on haemorrhoids, with a rate 2.8 times greater than the state average and 4.6 times greater than the surgeons serving Portland. In Bangor, surgery rates overall were lower than in the other regions; however, the area still exceeded the other areas in surgery rates for tonsillectomy.
|
The consistency in rates from year to year means that the numbers of individuals exposed to the risk of surgery varies remarkably from region to region. For example, over the four-year period from 1973–76, hysterectomy rates in Lewiston (with a population of about 20 000 eligible women) were such that over 800 more women were operated upon than would have experienced surgery, had the average rate for the state applied. (Had the rate for Augusta applied, 1600 fewer operations would have been done.) Two very enthusiastic surgeons undertook most of the procedures performed on women in this area. The rates were such that about 70% were estimated to undergo the operation by age 70.
The surgical signature phenomenon is ubiquitous, seen in all regions, regardless of their credentials for excellence in medicine. Over the years, we have obtained data on communities served by famous academic medical centres. The situation in Boston and New Haven is an example. Residents of these two communities receive most of their care from physicians who are faculty members of some of the nation's most prestigious medical schools. One might assume that because of these credentials, the care received by these two populations would be of the highest quality. Yet how different the quality: for some operations the risk of surgery was much higher in New Haven than in Boston; for others, Bostonians were much more likely to undergo surgery. During the 1980s, the risk for a hysterectomy (to treat symptoms of menopause) and for bypass surgery (to treat coronary artery disease) was nearly twice as high for New Haven residents as for Bostonians; while for carotid artery surgery (to prevent stroke) and hip replacement operations (for arthritis), the risk was 2.4 and 1.6 times greater, respectively, for residents of Boston.
The common cause was traced to Glover's medical opinion—theories strongly held by small groups of physicians concerning medical efficacy or value I learned through interviews with physicians in both communities that the differences in carotid endarterectomy rates could be attributed to a group of skeptical neurologists who simply didn't believe in the procedure, preferring aspirin to surgery for any New Haven patient that came to them for advice. By contrast (although relative to many other parts of the country the rates were low) the physicians in Boston had, on average, greater ‘faith’ in carotid artery surgery. The conservative medical management of coronary artery disease and symptoms of menopause was more popular in Boston, while clinicians in New Haven more often preferred surgical management. On the other hand, New Haven physicians were more conservative in the management of arthritis of the hip.
The influence of Glover continues to be felt today. The study of variations leads naturally to questions of outcomes and the value of health care underlines the need to evaluate the common practices of medicine and emphasizes the importance of respecting patient opinion or preferences in the decision to utilize discretionary medical care.
|
References |
|---|
|
TopScientific uncertainty and the...The surgical signature...References |
|---|
1 Alison Glover J. The incidence of tonsillectomy in school children. (1938) 31. Proceedings of the Royal Society of Medicine. 1219–36. (Reprinted Int J Epidemiol) 2008;37:09–19.
2 American Child Health Association. The Pathway to Correction in Physical Defects. New York, 1934, p. 80.
3 Garrow RP. Report of School Medical Officer. (1930) Borough of Hornsey, Ministry of Health, The Royal Society of Medicine.
4 Wennberg J, Blowers L, Parker R, Gittelsohn AM. Changes in Tonsillectomy Rates Associated with Feedback and Review. Pediatrics (1977) 59:821–26.
5 Bloor MJ, Venters GA, Samphier ML. Geographical variation in the incidence of operations on the tonsils and adenoids. Part I. J Laryngol Otol (1978) 92:791–801.[Web of Science][Medline]
6 Bloor MJ, Venters GA, Samphier ML. Geographical variation in the incidence of operations on the tonsils and adenoids. Part II. J Laryngol Otol (1978) 92:883–95.[Web of Science][Medline]
7 Wennberg J, Gittelsohn A. Small Area Variations in Health Care Delivery: A Population-Based Health Information System Can Guide Planning and Regulatory Decision-Making. Science (1973) 182:1102–8.
8 Wennberg J, Gittelsohn A. Variations in Medical Care Among Small Areas. Sci Amer (1982) 246:120–34.[Web of Science][Medline]
9 Wennberg J, Gittelsohn A. Health Care Delivery in Maine I: Patterns of Use of Common Surgical Procedures. J Maine Med Assoc (1975) 66:123–30.[Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  G. Davey Smith Big business, big science? Int. J. Epidemiol., February 1, 2008; 37(1): 1 - 3. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||