[Skip to Content]
Sign In
Individual Sign In
Create an Account
Institutional Sign In
OpenAthens Shibboleth
[Skip to Content Landing]
April 2000

Antimicrobial Use in Defined Populations of Infants and Young Children

Author Affiliations

Copyright 2000 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2000

Arch Pediatr Adolesc Med. 2000;154(4):395-400. doi:10.1001/archpedi.154.4.395

Background  Antimicrobial overprescribing contributes to bacterial resistance, but data on use in infants and young children are limited.

Objectives  To assess antimicrobial use in a defined population of infants and young children and to determine diagnosis-specific prescribing rates for common infections.

Design and Setting  Retrospective cohort study of children served by 44 practices affiliated with 2 managed care organizations.

Patients  Children aged 3 months to 72 months enrolled in either health plan between September 1, 1994, and August 31, 1996.

Analysis  Rates of antimicrobial use were calculated as the number of pharmacy dispensings divided by the number of person-years of observation contributed to the cohort in 2 age groups (3 to <36 months and 36 to <72 months). Other outcomes included the distribution of diagnoses associated with antimicrobial dispensing and population-based rates of diagnosis of common acute respiratory tract illnesses.

Results  A total of 46,477 children contributed 59 710 person-years of observation across the 2 health plans. Rates of antimicrobial dispensing for children aged 3 to 36 months were 3.2 and 2.1 dispensings per person-year in the 2 populations. A substantial fraction of younger children (35% in population A and 23% in population B) received 4 or more antimicrobial prescriptions in a single year. For children aged 36 to 72 months, the dispensing rates for the 2 populations were 2.0 and 1.5 antimicrobials per person-year. We found significant differences in rates between the populations studied and a decrease in use at all sites from 1995 to 1996. The diagnosis of otitis media accounted for 56% of antimicrobial drugs dispensed to children aged 3 to 36 months and 40% of those dispensed to children aged 36 to 72 months. Antimicrobial prescribing for colds and upper respiratory tract infections, bronchitis, and sinusitis was less frequent than previously reported but accounted for 10% to 14% of antimicrobial drugs dispensed.

Conclusions  In these populations, otitis media accounted for the largest number of antimicrobial agents dispensed to children younger than 6 years. Clearly inappropriate indications such as cold, upper respiratory tract infection, and bronchitis accounted for smaller fractions of antimicrobial use but may be most amenable to change. However, interventions that encourage use of strict criteria for diagnosis and treatment of otitis media will likely have the greatest impact on overall antimicrobial exposure. Monitoring defined populations longitudinally will allow assessment of the effectiveness of such national and local initiatives.

THE INCREASING prevalence of antimicrobial-resistant microorganisms is widely viewed as a significant emerging threat to public health.1-4 Penicillin resistance in previously sensitive organisms, such as Streptococcus pneumoniae, has been detected in 16% of isolates from 19 states,5 and rates as high as 41% have been reported for children.6,7 Recent data from the SENTRY surveillance program8 showed the percentage of penicillin-resistant pneumococcus from 27 US hospitals to be 44%, with 16% being highly resistant. A core recommendation of the Centers for Disease Control and Prevention for minimizing the impact of drug-resistant S pneumoniae is the reduction of unnecessary antimicrobial use.3,9 Because children have the highest rates of antimicrobial use,10 specific principles for use in pediatric populations have been developed and disseminated.11

The ambitious goal of changing antimicrobial prescribing practices for children requires accurate data on current patterns of use. In some countries, such data are available for identified geographic districts or the nation as a whole.12,13 In the United States, the 1992 National Ambulatory Medical Care Survey (NAMCS) estimated an annual average rate of antimicrobial prescribing of 928 per 1000 children younger than 15 years.10 The 1995-1996 NAMCS estimated an average annual rate of 1150 per 1000 children younger than 5 years (Linda F. McCaig, MPH, e-mail communication, November 24, 1998). The NAMCS sampled physician visits in representative practices and extrapolated the results to the entire US population. For children younger than 18 years, antimicrobial agents were prescribed for colds, upper respiratory tract infections (URIs), and bronchitis in 44%, 46%, and 75%, respectively, and accounted for 21% of all antimicrobials prescribed.14 Although these represent the best US estimates of antimicrobial use to date, data on the number of patients served by these practices are not directly available. Such visit-based studies yield direct estimates of prescription rates per visit but do not allow characterization of antimicrobial use for a cohort of children over time. Finally, no published analyses of the NAMCS specifically address prescribing to infants and young children, the highest users of antimicrobial agents.14 Others15,16 have contributed useful data on the rates and patterns of diagnosis of otitis media and antimicrobial prescribing using Medicaid claims files, but these might not be generalizable to non-Medicaid populations.

Although it is difficult to obtain complete data on geographically defined communities in the United States, managed care organizations (MCOs) care for reasonably large populations in which patterns of antimicrobial use can be studied. Data from MCOs that allow tracking medication use for a specific population base will be useful to gauge the effectiveness of initiatives to decrease antimicrobial exposure. We therefore studied antimicrobial prescribing in children enrolled in MCOs in 2 cities to determine the rates and distribution of antimicrobial use among infants and young children, to assess diagnosis-specific prescribing, and to determine the frequency with which diagnoses of common respiratory tract infections are assigned.

Patients and methods

We performed a retrospective cohort study in defined populations of children served by 2 large MCOs: Harvard Pilgrim Health Care in eastern Massachusetts (population A) and Group Health Cooperative of Puget Sound in northwest Washington State (population B). We included data from pediatric practices caring for more than 100 children aged 3 to 72 months enrolled in the health plans. Population A included patients from 20 network group practices with a total of 86 pediatricians and 23 pediatric nurse practitioners, with 7 family physicians providing a small fraction of pediatric care. Population B included patients cared for in health centers of an integrated staff model organization; the 1995 data are drawn from 24 practice sites, 2 of which were closed during the study period, with their patients absorbed by the remaining practices. Clinicians for population B included 28 pediatricians, 18 nurse practitioners, 181 family physicians, and 68 physician assistants.


Children aged 3 months to 6 years were studied from September 1, 1994, to August 31, 1996. We used enrollment files of the MCOs to calculate the number of days each child was enrolled and then aggregated them as the denominator for population-based rates. Contributed time began on the first day of the study period, at enrollment in the health plan, or at 3 months of age (whichever came last) and ended at the end of the study, at disenrollment, or at their sixth birthday (whichever came first). We specified 2 age groups before the analysis: 3 months up to 36 months and 36 months up to 72 months; published guidelines17 for care of febrile children use 36 months as the cutoff value. Some children contributed less than a full year of observation time and others contributed observation time to both age groups. We examined 2 separate time periods, September 1, 1994, to August 31, 1995, and September 1, 1995, to August 31, 1996. For our analysis of the frequency of antimicrobial use in individual children (Figure 1), we limited the sample to those who were covered by the plan and remained in one age group for an entire year.

Figure 1.
Frequency distribution of antimicrobial drugs dispensed in children aged 3 months to 36 months (left) and those aged 36 months to 72 months (right).

Frequency distribution of antimicrobial drugs dispensed in children aged 3 months to 36 months (left) and those aged 36 months to 72 months (right).

To be included, enrolled children must have received pharmacy benefits through the health plans. We excluded children nominally enrolled in each health plan (4% in population A and 14% in population B) who had no record of any ambulatory visits (including well-child care) and no antimicrobial dispensing during the observation period. We believe that this most often occurred because of dual sources of health insurance.

Data sources and analysis

We used automated ambulatory claims and pharmacy dispensing data routinely collected by MCOs. The claims systems allowed identification of diagnoses using International Classification of Diseases, Ninth Revision (ICD-9), codes.18 We created diagnostic categories that included the following ICD-9 codes and their subcodes: otitis media (381-382); pharyngitis (including tonsillitis and scarlet fever) (462, 463, 034.0, and 034.1); sinusitis (461 and 473); bronchitis (466 and 490); URI and common cold (460 and 465); and a composite group including sepsis, pneumonia, and urinary tract infection (including 030-041, 480-486, 590, 595, and 599). Remaining diagnoses, including well-child care visits, were categorized as "other." In the 17% of encounters for which more than 1 diagnosis was recorded, we assigned a primary diagnosis giving priority to a respiratory tract illness and, when present, to a potential bacterial source (eg, if common cold and otitis media were coded, the latter was designated "primary"). Follow-up visits (eg, for otitis) are also included in visit counts for each diagnosis.

Pharmacy claims records, which include original prescriptions and refills, were linked with the most recent ambulatory encounter if one occurred within the 3 days preceding the dispensing. We excluded antituberculosis drugs, antihelminthics, topical antimicrobials, and other antimicrobial agents rarely used in ambulatory pediatric care.

Data were analyzed using SAS statistical software (SAS version 6.12; SAS Institute Inc, Cary, NC). Confidence intervals for rates of dispensing were calculated using approximate Fisher intervals for Poisson variables in person-time data (Computer Programs for Epidemiological Analysis; USD Inc, Stone Mountain, Ga). All comparisons of person-time incidence rates assumed 2-tailed α=.05. The frequency distributions of antimicrobials dispensed to each child were compared using the Wilcoxon rank sum test for nonparametric data.


The cohort included 46,477 children who were observed for 59,710 person-years. The gender distribution was similar for both populations (population A, 48.5% female and population B, 48.3% female). Dispensing rates were higher for children aged 3 to 36 months than for those aged 36 to 72 months (P<.001) (Table 1). The rates were also significantly higher in 1995 than in 1996 (P<.001). Both these effects were evident in each of the health plans, which also differed significantly from one another; overall, children in population A received 2.56 antimicrobial drugs per person-year and those in population B received 1.82 (P<.001).

Table 1. 
Rates of Antimicrobial Drug Use of the Study Population*
Rates of Antimicrobial Drug Use of the Study Population*

Figure 1 shows the distribution of antimicrobial drug dispensings per year. More than one quarter of the children in both age groups received no antimicrobials. There were also substantial proportions (11%-35% depending on age and site) who received 4 or more dispensings in a year, with such frequent dispensing more common among younger children (P<.001). First-line penicillins accounted for 57% of all antimicrobials dispensed, trimethoprim-sulfamethoxizole for 17%, cephalosporins for 12%, erythromycins for 7%, and amoxicillin-clavulanate for 4%.

Primary diagnoses linked to antimicrobial agents are presented in Figure 2. Approximately 80% of antimicrobial drug dispensings could be linked to an ambulatory visit in the preceding 3 days. Fifty-six percent of all antimicrobials dispensed to younger children were associated with a diagnosis of otitis media. A diagnosis of cold or URI accounted for 3% of antimicrobials dispensed. The combined fraction of antimicrobials dispensed for colds and URIs, bronchitis, and sinusitis (illnesses with overlapping symptoms and signs) was 10% in children aged 3 to 36 months and 14% in those aged 36 to 72 months.

Figure 2.
Distribution of primary diagnoses linked to an antimicrobial drug dispensing in children aged 3 months to 36 months (left) and those aged 36 months to 72 months (right). URI indicates upper respiratory tract infection; asterisk, includes urinary tract infection, pneumonia, and other unspecified bacterial infections.

Distribution of primary diagnoses linked to an antimicrobial drug dispensing in children aged 3 months to 36 months (left) and those aged 36 months to 72 months (right). URI indicates upper respiratory tract infection; asterisk, includes urinary tract infection, pneumonia, and other unspecified bacterial infections.

Diagnosis-specific visit and treatment rates are shown in Table 2. In each age group and each health plan, otitis media was the most common diagnosis. Children younger than 36 months were treated for otitis with an antimicrobial agent 1.68 and 1.24 times per person-year and older children were treated for otitis media 0.80 and 0.61 times per person-year in populations A and B, respectively. It is not surprising that otitis media was associated with antimicrobial prescribing in only 61% to 78% of patients because we included follow-up visits and diagnoses of otitis media with effusion in this category. Otitis media with effusion accounted for 13% of all otitis visits; visits for otitis media with effusion were associated with an antimicrobial in 36% and 40% of patients in populations A and B, respectively. Although visits for colds and URIs were frequent, the proportion treated with antimicrobials was low, from 8% to 17% depending on site and age group. Together, prescriptions for colds and URIs, bronchitis, and sinusitis accounted for fewer than 0.3 antimicrobial drug dispensings per person-year.

Table 2. 
Population-Based Rates of Diagnoses and Their Treatment for 1995 and 1996 Combined*
Population-Based Rates of Diagnoses and Their Treatment for 1995 and 1996 Combined*

Among infants and young children, respiratory tract illness (especially otitis media) is common, diagnosis is challenging, and pressures to treat with antimicrobial agents may be intense. The overall rates of antimicrobial use we report are somewhat higher than those estimated by the 1992 NAMCS.10 However, our rates of antimicrobial use for otitis media of 1.24 to 1.68 dispensings per person-year for children aged 3 to 36 months and 0.61 to 0.80 for children aged 3 to 6 years are lower than those reported by Berman et al15 in a population-based study using Medicaid claims in Colorado. As with other medical practices,19 we can learn a great deal by studying variation among geographic areas and patient groups.

Our data also show that a substantial fraction of children receive no antimicrobials, whereas others receive several prescriptions each year. This likely reflects person-to-person variation in occurrence of disease and differences in care-seeking behavior by parents and responses to perceived parental expectations by physicians. Some investigators20 have suggested that clinicians frequently prescribe antimicrobial agents for viral URI in response to parental demand or to shorten explanation time during office visits. In a recent survey, 48% of physicians reported parental pressure to prescribe unnecessary antimicrobial agents, and one third reported at least occasionally complying.21 Recent identification of predictors of high prescribing rates among individual physicians adds another dimension to our understanding of what drives overuse.16

Colds and bronchitis accounted for fewer than 10% of prescriptions in these health plans, a figure lower than the 21% reported from NAMCS data for children younger than 15 years.14 The proportion of visits for colds and URIs that were treated in these populations was also lower—8% to 17%—compared with the 44% to 46% previously reported.14 Although bronchitis was often treated with antimicrobial drugs, this indication accounted for 5% or less of all prescriptions. Physicians in these 2 MCOs might choose different diagnostic codes for the same illnesses than the broader sample of physicians in the NAMCS. They might also be more attentive to coding diagnoses that are consistent with treatments prescribed. Alternatively, methodological differences in studies of antimicrobial prescribing might account for some differences in reported results. For example, the NAMCS used a visit-based instrument to capture primary and additional diagnoses. If URI was listed first and otitis media second, an antimicrobial, if present, was linked to the diagnosis of URI. In this study, we used an algorithm that gave priority to a presumed bacterial infection as the primary diagnosis when more than 1 was recorded.

Like other investigators,14 we included commonly used ICD-9 codes to focus on specifically diagnosed respiratory tract infections. Among visits linked to an antimicrobial drug and categorized as "other," some carry another diagnosis consistent with viral upper respiratory tract illnesses. To investigate the impact of our a priori selection of ICD-9 codes, we reviewed diagnoses for the 12% of antimicrobial drugs linked to visits categorized as "other" in population A. Of these, 1.8% of all antimicrobials were linked with viral respiratory tract infections, 1.1% with unspecified or nonrespiratory viral illnesses, 2.2% with possible bacterial infections, 3.0% with noninfectious diagnoses, and 3.7% with well-child care. The latter group is likely to represent children who present for well-child care with an intercurrent illness and shows that use of diagnoses from claims will misclassify some of the reasons antimicrobial agents are prescribed. The potential impact of variability in assigning diagnosis codes, although small, makes overall antimicrobial use rates particularly important study outcomes.

In this study, the diagnosis of acute otitis media was the most common reason for antimicrobial drug prescribing. If acute otitis media is being overdiagnosed, it would likely account for the largest fraction of antimicrobial overuse. The literature regarding the reduction of antimicrobial prescribing often ignores diagnostic uncertainty in respiratory tract infections, which may be another important factor in overprescribing; visualization of the tympanic membrane is rarely optimal in a febrile, squirming young child. Interventions that promote the use of strict criteria for diagnosis of otitis media, or even a "watchful waiting" approach in some patients,22,23 may make the greatest impact.

The substantial differences between the 2 populations studied illustrate the importance of understanding local patterns of disease and management in designing interventions to improve care. The antimicrobial dispensing rates reflect differences in rates of presentation for illnesses and smaller differences in diagnosis-specific treatment rates. These 2 plans differ in geographic location (Northeast vs Northwest), delivery system structure (staff vs network model),24 and the frequency of family physicians as providers of care to children. Although there were no concerted efforts to decrease antimicrobial prescribing during the observation period, practice-specific education from the health plan or other sources might have occurred. Demographic differences in the populations served may also be important. Additional work is needed to determine which, if any, of these factors contribute to the differences in results between the 2 sites.

The decrease in antimicrobial use we observed from 1995 to 1996 emphasizes the importance of suitable comparison groups for assessing the impact of interventions to decrease antimicrobial use. It is difficult to rule out random year-to-year variation, but these data might be evidence of a secular trend in the direction of lower antimicrobial treatment rates. This change antedated the Centers for Disease Control and Prevention guidelines promoting judicious antimicrobial use11 but might have occurred in response to increased awareness by professionals and the public regarding the risks of antimicrobial overuse during this period.

As attempts are made to reduce antimicrobial prescribing, we will need to watch carefully for changes in coding among competing diagnoses. For example, if physicians reduce their antimicrobial treatment of bronchitis but increase the diagnosis and treatment of sinusitis, no progress will be made. Conversely, if public education reduces the frequency of physician visits for upper respiratory tract symptoms, overall exposure to antimicrobial drugs might decrease even if the proportion of visits with treatment remains high. Monitoring defined populations over time will allow analysis of complex changes in care-seeking and prescribing behavior.

The generalizability of this study depends on the extent to which physicians and patients in these MCOs are representative of those in their communities. Studies in other settings will help determine the extent of the variation in prescribing practices across the United States. Although administrative pharmacy databases have the advantage of capturing actual medications dispensed, rather than merely prescribed, they have limitations as well. A substantial fraction (19%) of antimicrobials dispensed could not be linked to an office visit in the preceding 3 days; reasons might include providers prescribing (or refilling prescriptions) by telephone or delayed filling of prescriptions by patients. Although we limited our cohorts to patients with prescription drug coverage, it is possible that a small number of medications were purchased "out of pocket" with no claim submitted to the health plan or were submitted to another insurer. Likewise, these databases would not include records of medication samples given directly to patients. Finally, analysis of automated data sources cannot assess the appropriateness of diagnoses made or treatments prescribed.

The public health community, physician groups, and the lay press are increasingly delivering the unified message that major change in the use of antimicrobial agents is warranted. Data from Iceland support the correlation between community antimicrobial use rates and bacterial resistance25 and the premise that decreasing prescribing may have a substantial impact on resistance.13 We do not know what magnitude reduction in antimicrobial prescribing will affect the prevalence of antimicrobial resistance among community pathogens in the United States. Although continued efforts to eliminate antimicrobial prescribing for colds, URIs, and bronchitis are warranted, use of more stringent criteria for the diagnosis of acute otitis media will likely have the greatest impact on overall antimicrobial exposure for infants and young children. Monitoring defined populations during the next several years will help us assess progress toward the intermediate goal of reducing antimicrobial use and the ultimate goal of preventing resistance.

Accepted for publication August 30, 1999.

This study was funded by the Centers for Disease Control and Prevention, Atlanta, Ga, through the Association of Community Health Plans (task order 200-95-0953). Additional support was provided by the Harvard Pilgrim Health Care Foundation, Boston, Mass.

We acknowledge Donald Goldmann, MD, for his helpful comments on early drafts of the manuscript and Irina Miroshnik and David Rabinowice for their help with the analyses.

Corresponding author: Jonathan A. Finkelstein, MD, MPH, Department of Ambulatory Care and Prevention, Harvard Medical School and Harvard Pilgrim Health Care, 126 Brookline Ave, Suite 200, Boston, MA 02215.

Editor's Note: I'm still waiting for the study that will provide clear indication for the use of antimicrobial drugs in otitis media, but I'm not holding my ears.—Catherine D. DeAngelis, MD

Tenover  FCHughes  JM The challenges of emerging infectious diseases.  JAMA. 1996;275300- 304Google ScholarCrossref
Neu  HC The crisis in antibiotic resistance.  Science. 1992;2571064- 1073Google ScholarCrossref
Schwartz  BBell  DMHughes  JM Preventing the emergence of antimicrobial resistance: a call to action by clinicians, public health officials, and patients.  JAMA. 1997;278944- 945Google ScholarCrossref
Williams  RJHeymann  DL Containment of antibiotic resistance.  Science. 1998;2791153- 1154Google ScholarCrossref
Mason  EOLamberth  LLichenstein  RKaplan  SL Distribution of Streptococcus pneumoniae resistant to penicillin in the USA and in-vitro susceptibility.  J Antimicrob Chemother. 1995;361043- 1048Google ScholarCrossref
Hofmann  JCetron  MFarley  MM  et al.  The prevalence of drug-resistant Streptococcus pneumoniae in Atlanta.  N Engl J Med. 1995;333481- 486Google ScholarCrossref
Pikis  AAkram  SDonkersloot  JACampos  JMRodriguez  WJ Penicillin-resistant pneumococci from pediatric patients in the Washington, DC, area.  Arch Pediatr Adolesc Med. 1995;14930- 35Google ScholarCrossref
Doern  GVPfaller  MAKugler  KFreeman  JJones  RN Prevalence of antimicrobial resistance among respiratory tract isolates of Streptococcus pneumoniae in North America: 1997 results from the SENTRY antimicrobial surveillance program.  Clin Infect Dis. 1998;27764- 770Google ScholarCrossref
Jernigan  DBCetron  MSBrieman  RF Minimizing the impact of drug-resistant Streptococcus pneumoniae (DRSP): a strategy from the DRSP Working Group.  JAMA. 1996;275206- 209Google ScholarCrossref
McCaig  LFHughes  JM Trends in antimicrobial drug prescribing among office-based physicians in the United States.  JAMA. 1995;273214- 219Google ScholarCrossref
Dowell  SFMarcy  SMPhillips  WRGerber  MASchwartz  B Principles of judicious use of antimicrobial agents for pediatric upper respiratory tract infections.  Pediatrics. 1998;101163- 165Google Scholar
Straand  JRokstad  KHeggedal  U Drug prescribing for children in general practice: a report from the More and Ramsdal Prescription Study.  Acta Paediatr. 1998;87218- 224Google ScholarCrossref
Stephenson  J Icelandic researchers are showing the way to bring down rates of antibiotic-resistant bacteria.  JAMA. 1996;275175Google ScholarCrossref
Nyquist  AGonzales  RSteiner  JFSande  MA Antibiotic prescribing of children with colds, upper respiratory tract infections, and bronchitis.  JAMA. 1998;279875- 877Google ScholarCrossref
Berman  SByrns  PJBondy  JSmith  PJLezotte  D Otitis media–related antibiotic prescribing patterns, outcomes, and expenditures in a pediatric Medicaid population.  Pediatrics. 1997;100585- 592Google ScholarCrossref
Mainous  AGHueston  WJLove  MM Antiobiotics for colds in children.  Arch Pediatr Adolesc Med. 1998;152349- 352Google Scholar
Baraff  LJBass  JWFleisher  GR  et al.  Practice guideline for the management of infants and children 0 to 36 months of age with fever without source.  Ann Emerg Med. 1993;221198- 1210Google ScholarCrossref
World Health Organization, International Classification of Diseases, Ninth Revision (ICD-9).  Geneva, Switzerland World Health Organization1977;
Wennberg  JGittelsohn  AM Variations in medical care among small areas.  Sci Am. 1982;246120- 134Google ScholarCrossref
Hamm  RMHicks  RJBemben  DA Antibiotics and respiratory infections: are patients more satisfied when expectations are met?  J Fam Pract. 1996;4356- 62Google Scholar
Bauchner  HPelton  SIKlein  JO Parents, physicians, and antibiotic use.  Pediatrics. 1999;103395- 401Google ScholarCrossref
van Buchem  FLPeeters  MFvan't Hof  MA Acute otitis media: a new treatment strategy.  Br Med J (Clin Res Ed). 1985;2901033- 1037Google ScholarCrossref
Froom  JCulpepper  LGrob  P  et al.  Diagnosis and antibiotic treatment of acute otitis media: report from International Primary Care Network.  BMJ. 1990;300582- 586Google ScholarCrossref
Lambert  BLSalmon  JWStubbings  JGilomen-Study  GValuck  RJKezlarian  K Factors associated with antibiotic prescribing in a managed care setting: an exploratory investigation.  Soc Sci Med. 1997;451767- 1779Google ScholarCrossref
Arason  VAKristinsson  KGSigurdsson  JAStefansdottir  GMolstad  SGudmundsson  S Do antimicrobials increase the carriage rate of penicillin resistant pneumococci in children? cross sectional prevalence study.  BMJ. 1996;313387- 391Google ScholarCrossref