Coyte PC, Croxford R, Asche CV, To T, Feldman W, Friedberg J. Physician and Population Determinants of Rates of Middle-Ear Surgery in Ontario. JAMA. 2001;286(17):2128-2135. doi:10.1001/jama.286.17.2128
Author Affiliations: Departments of Health Policy, Management, and Evaluation (Dr Coyte and Mr Asche), Otolaryngology (Dr Friedberg), Pediatrics (Dr Feldman), and Public Health Sciences (Dr To), and the Institute for Policy Analysis (Dr Coyte), University of Toronto, Toronto, Ontario; Institute for Clinical Evaluative Sciences, Toronto (Drs Coyte and To); Clinical Epidemiology Unit, Sunnybrook and Women's College Health Science Centre, Toronto (Ms Croxford); Department of Otolaryngology (Drs Coyte and Friedberg), Mount Sinai Hospital, Toronto; and the Departments of Otolaryngology (Dr Friedberg) and General Pediatrics (Dr Feldman), Hospital for Sick Children, Toronto. Mr Asche is now with Aventis Pharmaceuticals, Bridgewater, NJ.
Context Small-area variations in surgical rates raise concerns about access
to care, treatment appropriateness, and the quality and cost of care.
Objective To measure small-area variations in rates of myringotomy with insertion
of tympanostomy tubes (TTs) and to identify determinants of rate variation.
Design and Setting Retrospective analyses using hospital discharge data for patients who
had undergone a myringotomy with insertion of TT by county in Ontario between
April 1, 1996, and March 31, 1999. Information on possible determinants was
taken from a survey of otolaryngologists and primary care physicians in 1996
and from the 1996 Canadian census and physician demographic databases for
Participants A total of 75 358 hospitalizations for TT placement of children
and adolescents (aged ≤14 years).
Main Outcome Measure Small-area variation in rates of TT.
Results An almost 10-fold difference between the areas with the highest and
lowest rates was found (extremal quotient, 9.6; 95% confidence interval [CI],
8.2-11.1; P<.001). Higher rates occurred in counties
with higher percentages of high school graduates (parameter estimate, 0.01;
95% CI, 0-0.02; P = .049); and where referring physicians
were more likely to be male (parameter estimate, 0.01; 95% CI, 0-0.02; P = .01), North American–trained (parameter estimate,
0.01; 95% CI, 0.01-0.02; P<.001), and have higher
propensities to refer for surgery (parameter estimate, 0.40; 95% CI, 0.09-0.72; P = .02). Otolaryngologist opinion was not a significant
Conclusion Substantial area variation in TT rates was observed. The opinion of
primary care physicians was the dominant modifiable determinant, suggesting
an area of research that may be important in reducing area variation in TT
Otitis media is a common childhood condition, affecting 65% to 93% of
children by age 7 years.1 Myringotomy with
the insertion of tympanostomy tubes (TTs) is the most common type of pediatric
surgical procedure—more than 1 million operations are performed in the
United States and Canada every year2,3—and
may become more important given concerns about antibiotic-resistant bacteria.4
For many surgical procedures, the probability of having surgery depends
on where one lives, a phenomenon termed small-area variation.5 The implication of variation in rates,
which cannot be explained by differences in disease prevalence, is that patients
in low-use areas may be receiving too little care or patients in high-use
areas may be receiving too much (or inappropriate) care. Furthermore, due
to the high prevalence of otitis media, variations in its treatment have implications
for cost. Estimated direct and indirect costs of treatment in the United States
exceed $5 billion a year.2 In Canada, with
approximately one-tenth the population of the United States, the estimate
is $611 million dollars6 (in 1994, the exchange
rate was $1.00 [Canada] to $0.75 [United States]).
Area variation has been theorized to stem from differences in community
population characteristics and methods of health care delivery, including
the number, type, and opinions of medical practitioners.7
One goal of area variation research is "to identify sources of variation whose
modification is both practical and desirable."8
The purpose of this study was to acquire important descriptive information
to measure and explain variations in TT rates in Ontario.
In Ontario, hospitals are required to report all inpatient and same-day
surgical procedures. Such procedures account for over 96% of all TT operations
(the remainder are performed in private offices).9
Universal public health insurance covers all medically necessary services;
supplementary private insurance for these services is prohibited. Therefore,
surgery is available to all on equal financial terms, unlike in the United
Kingdom, where surgery may be performed outside the National Health Service,
or in the United States, where lack of insurance may limit use.10
We searched hospital discharge records for fiscal years (April 1 through
March 31) 1996 through 1999 for all children and adolescents (aged ≤14
years), accounting for 87.5% of all TT hospitalizations, who had undergone
a myringotomy with insertion of TT (Canadian Classification
of Diagnostic, Therapeutic and Surgical Procedures code 32.01). Records
were excluded if the patient's place of residence could not be identified
(0.5%) or if a valid patient identifier was not available, preventing linkage
over time (1.2%).
Since hospitals vary in their coding protocols, our analysis focused
on hospitalizations rather than procedures. (Some hospitals reported 2 TT
procedure codes when a bilateral TT was performed, other institutions submitted
a single procedure code.) Over the study period, 75 358 TT hospitalizations
The rate of TT hospitalizations was determined for each of the 49 counties
in Ontario. Rate was defined as the number of TT hospitalizations for all
children and adolescents in the county, irrespective of where these hospitalizations
took place, divided by the resident population aged 14 years or younger. County
populations were determined from the 1996 Canadian census. Direct standardization
was used to adjust crude rates for differences in age and sex distributions
among the counties. All analyses were performed on the standardized rates.
Potential overestimates of area variation occur if the population is
small and rates of intervention are low.11
The common occurrence of TT surgery and the pooling of 4 years of data lessen
the importance of these effects. Nonetheless, for robustness and comparability,
we used 3 measures of variation: the extremal quotient (EQ), the weighted
coefficient of variation (CV), and the systematic component of variance (SCV).11,12 The EQ is the ratio of the highest
county TT rate to the lowest rate. The CV measures the ratio of the SD of
49 county TT rates to the mean rate, weighted by the population in each county
to account for their unequal sizes, multiplied by 100. The SCV measures variations
in rates after adjusting for random variations, and was multiplied by 1000.
Measures of area variations in the literature refer to counts of events rather
than to individuals, which has been followed here to permit comparisons.
The EQ is perhaps the most widely reported measure of area variation,
and the easiest to understand. The drawback is that it depends only on the
2 most extreme rates, both of which may be outliers. The lowest rate, in particular,
is likely to be sensitive to random variation and errors in the data. The
CV is based on rates in all areas, but may overestimate variations when rates
are low or when the area population is small.11
The SCV, developed to address these concerns, removes the component of variation
attributed to population size, and provides a measure of variation that is
stable across a range of rates and population sizes.11
Large variations in TT rates might occur by chance due to random fluctuation.
To address this, we conducted a computer simulation12
using the observed distribution of TT hospitalizations per person over the
study period. We determined how many children and adolescents had no TT procedures,
how many had 1 TT procedure, how many had 2 TT procedures, etc, over 4 years.
The resulting population was randomly assigned to the 49 counties, with each
receiving the appropriate number of simulated children. This reflects the
distribution of TT rates expected due to chance, under the null hypothesis
of no differences among the counties. Summary measures of area variation (EQ,
CV, and SCV) were computed for the simulated population. This was repeated
10 000 times so that tests could be conducted to determine whether our
observed measures of variation were due to chance alone.
χ2 Tests were also used to test for variation in TT rates.12 Individual county rates were compared with the rate
for the province as a whole, using a level of significance of .001 to adjust
for multiple comparisons. χ2 Values were further adjusted using
a multiple admissions factor to correct for the occurrence of multiple hospitalizations
for some patients.12
In Ontario, family physicians, general practitioners, and pediatricians
(referring physicians) serve as the "gatekeepers" to the health care system.
Patients can obtain consultations with specialists, such as otolaryngologists,
generally only through a physician referral. We hypothesized that the characteristics
and opinions of both otolaryngologists and referring physicians would affect
The median age and median year of licensing of referring physicians
and the percentage of women physicians and percentage of physicians who graduated
from North American (Canada or United States) medical schools vs non–North
American medical schools in each county in Ontario were obtained from the
National Physician Demographic Databases for 1996 through 1999.
Referring physician opinions regarding TTs were obtained from a survey
of all pediatricians in Ontario and a random sample of general practitioners/family
physicians, mailed in the fall of 1996.13 The
response rate was 52.5% for family physicians and 55.3% for pediatricians.
Respondents tended to be younger, female, to have been trained in Canada,
and to have been licensed more recently. These physician characteristics were
not related to either the propensity to refer or opinions expressed about
It has been suggested that physician enthusiasm is the most important
determinant of small-area variation.14 Two
measures of physician enthusiasm were created. Survey respondents indicated,
in 5 categories ranging from much less likely to much more likely to refer,
the manner in which each of 17 clinical and social factors influenced their
decision to refer a child for consideration of TT surgery. The first measure
of physician enthusiasm, which we called propensity to refer,15 was based on each physician's enthusiasm
for TT surgery, relative to his/her peers. (For example, most referring physicians
indicated that they were more likely or much more likely to refer a child
who showed no response to antibiotics or had >7 episodes of otitis media in
6 months. A physician who said these factors had no effect on the decision
to refer would be said to show a low propensity to refer vis-à-vis
these 2 items. The difference between an individual physician's response and
the median response was calculated and summed over all 17 items to yield that
physician's propensity to refer.) The propensity for TT surgery within a county
was the sum of the propensities of all referring physicians within the county,
adjusted for the survey sampling scheme, and weighted by the number of children
or adolescents with otitis media seen in their practice.
A second measure of physician enthusiasm, which we termed outcome perceptions,15 was based on
physicians' estimates of the proportion of children or adolescents who would
have 9 different surgical outcomes, such as fewer episodes of acute otitis
media, otorrhea, and need for tube reinsertion. Responses were combined in
a fashion similar to the propensity score to create a single county-specific
The survey also assessed the threshold at which a physician would refer
a child for number of acute otitis media episodes over 6 months; number of
months with middle ear effusion; level of bilateral conductive average pure
tone hearing loss; and duration of antibiotic therapy without clinical improvements.
Responses to these questions were combined to create a measure of the physician's threshold for referral.
The opinions of otolaryngologists were evaluated using data from a similar
mail survey of all practicing otolaryngologists in Ontario.16
The response rate was 68.3%. Respondents tended to be younger than nonrespondents.
Otolaryngologists' propensity to perform surgery was calculated in the manner
described above and weighted averages were obtained based on the county in
which the otolaryngologists practiced. Some counties in Ontario do not have
any practicing otolaryngologists; children and adolescents in these counties
received treatment in another county or in several other counties. The otolaryngologist
propensity score for a given county was therefore based on the location of
treatment. If 45% of the residents of county A were treated in county A, and
55% were treated in county B, then the otolaryngologist propensity score for
county A was defined as (0.45 [propensity A] + 0.55 [propensity B]), in which
propensity A is the propensity score reported by otolaryngologists from county
A. County-level otolaryngologist scores for outcome perceptions and thresholds
for performing surgery were calculated in the same way. Details of the calculations
have been described elsewhere.15
Two measures of access to surgery were used. The number of referring
physicians per 100 000 population was calculated as the sum of the number
of family and general practitioners per 100 000 population plus the number
of pediatricians per 100 000 population aged 14 years or younger. The
number of specialists per 100 000 population was based on the number
of otolaryngologists serving each county. If the otolaryngologists in county
A performed all of the TTs for residents of county A and 50% of the TTs for
residents of county B, then the number of otolaryngologists per 100 000
population in county A was calculated as: No. of otolaryngologists practicing
in county A divided by (population of county A + 0.50 × population of
county B). A county with no otolaryngologists was assigned a weighted average
of the number of otolaryngologists per 100 000 population practicing
in the counties in which its residents received their surgery.
County population was characterized using information from the 1996
census. We used the proportion of the population aged 14 years or younger
and 2 socioeconomic measures: the proportion of households with an income
under the low income cutoff (a measure of poverty developed by Statistics
Canada based on the proportion of household income spent on food, clothing,
and shelter); and the proportion of the population (aged ≥15 years) with
at least a high school education. Finally, each county was classified as urban
or rural based on its population density.
Number of hospital beds, generally invoked to explain small-area variation,
was not used in this analysis because most (94.4%) TT procedures are performed
as same-day surgeries. Information on the availability of operating rooms
was not available.
Bivariate correlations between TT rate and each of the explanatory variables
and interrelationships among explanatory variables were determined with Spearman
correlation coefficient or Wilcoxon rank sum tests (for the binary urban/rural
variable). Stepwise multiple linear regression and all possible subset regressions
were used to determine factors associated with county use rates. We tested
for violations of regression assumptions and used influence diagnostics to
check for observations with undue influence on the regression results.17
Statistical analyses were performed using SAS statistical software.18 All P values were 2-tailed.
The study was approved by the University of Toronto Ethics Review Committee.
Figure 1 depicts TT rates
for children and adolscents in Ontario between 1996 and 1999. The overall
rate was 8.35 per 1000 population aged 14 years or younger. The rate for males
was more than 50% greater than that for females, 10.0 vs 6.6 per 1000. Rates
were bimodal, peaking for children aged 1 year (20.4) and 4 years (16.3).
This age-sex distribution is congruent with those reported for children and
adolescents elsewhere in North America.19
Figure 2 depicts the age-sex
standardized TT rates for the 49 counties. χ2 Tests, adjusted
for multiple TT insertions, were used to assign each county to 1 of 3 categories
based on their rates relative to that for the province as a whole. Twelve
predominantly rural areas, with 25.3% of the provincial population aged 14
years or younger, were significantly (P<.001)
above the provincial TT rate of 8.35. In contrast, 6 mainly urban counties,
with 42.7% of the provincial population, were significantly (P<.001) below the provincial rate. The remaining counties, with
rates ranging from 3.83 to 11.97 per 1000 population, were not significantly
different from the provincial average.
Over the study period, the EQ was 9.6 (95% confidence interval, 8.2-11.1; P<.001); the maximum EQ expected by chance alone12 was 2.2. This implies almost a 10-fold difference
in TT rates between extremal counties, even after adjusting for the age-sex
composition of the population and pooling 4 years of data. The CV of 37.2
and the SCV of 267.9 were also significantly larger than can be explained
by chance (P<.001). These results are consistent
with the contention that there were large area variations in the surgical
management of childhood otitis media in Ontario. Table 1 reports the results, with context provided by measures of
rate variation for other common procedures and conditions.
County characteristics are presented in Table 2. Missing values for some counties are due primarily to lack
of survey responses from counties with small numbers of specialists. A wide
range of values are reported for propensity to perform (or refer for) TT,
thresholds for surgery, and outcome perceptions among otolaryngologists and
referring physicians, as well as for a range of physician and population characteristics.
Weak bivariate correlations between referring physician opinion and TT rates
were observed, with higher rates associated with a higher propensity to refer
for surgical consideration and a lower threshold for referral. Place of training
was also weakly correlated with TT rates—counties with a greater proportion
of referring physicians trained in North America tended to have higher rates.
Rates of TT showed stronger bivariate correlations with otolaryngologists'
opinions, particularly their propensity to perform the procedure and the threshold
at which they felt surgery was appropriate. Proportion of female specialists
in the county was the strongest single predictor of TT rates, with higher
rates in counties having relatively few female otolaryngologists.
The results of the multivariable models, using a number of different
multiple linear-regression techniques, consistently identified the same significant
predictors of TT rates. Higher rates were associated with counties with relatively
more North American-trained referring physicians, with more adults with at
least a high school education, and with a higher propensity of county physicians
to refer children or adolescents for TT surgery. Higher proportions of referring
physicians who were female were associated with lower TT rates. Regression
diagnostics identified 2 counties, both with high proportions of female physicians,
1 with the lowest rate of completed high school education and 1 with the highest
rate, as having a large influence on the parameter estimates. Removing these
2 observations did not affect the choice of explanatory variables in the model. Table 3 reports the results of the regression
after removing these 2 counties. The final model explains 34.2% of the total
variation in TT rates. The standardized coefficients indicate that the proportion
of physicians trained in North America is the strongest of the 4 predictors,
followed by the proportion of female physicians. Given this model, which contains
variables related to referring physician characteristics, none of the variables
related to otolaryngologists contributed a significant amount of additional
information. The reverse was not true. Given a model that contained the significant
variables related to specialist characteristics, referring physician propensity
was significant (P = .03), and proportion of female
referring physicians was marginally significant (P
= .08). It appeared that TT rates, at the county level, could be predicted
without reference to the characteristics of otolaryngologists.
As an aid to interpreting the variables included or excluded from the
multivariable regression model, Table 4 presents correlations between those variables selected for the
model and other explanatory variables. (Only correlations significant at the ≤.05
level are shown.) The variables that were included in the model are associated
with availability of referring physicians, with referring physicians' threshold
for referring a patient for surgical consideration, and with urban counties.
Population education and referring physician place of training were also associated
with the proportion of female otolaryngologists, explaining why this variable
did not appear in the multivariable model.
Myringotomy with insertion of TTs is the most common surgical procedure
performed on children and adolescents.2,3
Despite its frequency and cost,2,6
there is an absence of extensive epidemiological and health services research.1 Little is known about the magnitude and reasons for
variations in TT rates.
The substantial TT rate variations reported in this study exceed those
previously published for other jurisdictions20,21
and are larger than those reported for almost all other procedures,1,11,22 raising concerns about
health system costs and treatment appropriateness. With such large regional
variations, there may be some jurisdictions in which children and adolescents
are not receiving the care they need and other regions where TT surgery may
be performed unnecessarily.
The main intent of this study was to identify factors related to small-area
variation that might be amenable to intervention.8
Rates of TT were higher in counties with relatively more primary care physicians
trained in North America, and lower in counties with higher proportions of
female primary care physicians. This may be related to practice patterns as
well as training. Our survey found that female referring physicians are younger,
more likely to work part-time, and to work in group practices, rather than
solo practices (data not shown). The percentage of the population with a high
school education (a socioeconomic variable) was also related to area variation,
with higher rates in counties with a more educated population. This finding
agrees with the results of other studies, which have found that socioeconomic
status may explain a large proportion of rate variation.7,21
Counties with fewer female physicians and more North American–trained
physicians tended to be rural. In contrast, higher levels of educational achievement
tended to occur in urban counties. Thus, the bivariate relationships between
rural/urban location and TT rates appear to be related to physician characteristics
modified by educational achievement. The rural/urban variable was not a significant
predictor in the multivariable regression model.
Some studies have found that physician density is related to area variation.7,23 However, we found no relationship
between density of either referring physicians or otolaryngologists and use
rates. Thus, attempts to change the distribution of surgeons may have little
effect on variation in the use of TT surgery. This finding agrees with an
earlier study that found no relationship between rates of total knee replacements
and physician density in Ontario.24
The 1 factor related to area variation in this study that is potentially
amenable to intervention was physician opinion. We found that the referring
physician's opinion or enthusiasm for TTs was an important determinant of
rate variations, with higher TT rates associated with greater propensity to
refer. Although reduction in variation in the opinions of physicians has the
potential to reduce small-area variation, this does not ensure that the "right"
rate will be achieved.25 However, while clinicians
will always vary somewhat in their enthusiasm for a procedure, if variation
in physicians' opinions represents uncertainty or misperceptions about the
risks and benefits of surgery, addressing opinions that do not match published
literature and evidence-based practice guidelines provides one of the best
opportunities to reduce rate variation.
This study found that it was the opinion of primary care physicians
that predicted rates. This contrasts with the results of an earlier study
that found that the opinions of surgeons, rather than of referring physicians,
predicted rates of total knee replacements.15
The survey results13,16 suggest
that by the time children or adolescents meet the referring physicians' criteria
for referral for TTs, they may meet or exceed the criteria of the otolaryngologist
to perform the procedure. Our results suggest that efforts to reduce uncertainty
among physicians be directed at primary care providers.
There are 2 main study limitations. The first is lack of data about
the prevalence or severity of otitis media in different counties. Health services
literature has shown that while the prevalence of disease may be associated
with area variation in procedure rates, it is not generally the dominant factor.5,19 The second is that we did not have
information about parents' preferences for the treatment of their children.
For many surgical procedures, patients' preferences (or the preferences of
their parents) are an important determinant of decision making.26,27
Accounting for parental preferences may explain at least part of the observed
variation.27 However, parental preferences
may be partially based on information received from clinicians about the risks
and benefits of a procedure. If clinicians are providing variable (or inaccurate)
information, this may lead to misperceptions and ill-informed decision making.
A further limitation is that the study was based on hospital discharge
data acquired from the Canadian Institute for Health Information. The main
methodological concern with the use of administrative data is the validity
and reliability of the data.28,29
Studies that have compared administrative data with hospital chart data have
overwhelmingly concluded that major events, such as surgical procedures and
mortality, patient demographics, and primary diagnosis are coded accurately.29 As well, number of children and adolescents with
otitis media seen in the previous month (the variable used to weight physician
opinions) was based on recall and is subject to memory biases.
Our study was confined to clinical practice in Ontario for the period
1996 to 1999. The results will not necessarily generalize to other jurisdictions,
even within Canada, nor to other periods in Ontario. However, they are worth
consideration in any system in which primary care physicians play an important
role in gatekeeping, or have the opportunity to counsel patients with respect
to treatment and referral.
In conclusion, substantial variations in age-sex standardized TT rates
were observed in Ontario. These variations were exceptionally large compared
with TT rate variations in other jurisdictions, and compared with rate variations
for other medical and surgical procedures. After controlling for population
characteristics and access to care, the enthusiasm of referring physicians
for this procedure was the dominant modifiable determinant of area variation.
Providing referring physicians with evidence-based guidelines addressing areas
of uncertainty they face in managing children and adolescents with otitis
media may help to reduce variation. Future research will be necessary to determine
how patient characteristics should affect the decision to refer and how they
affect the actual surgical outcomes. Furthermore, we need to determine the
reasons for differences in the opinions of the gatekeepers to classify these
reasons into exogenous and modifiable factors, and finally, test strategies
to reduce variation in their opinions.25