Diagnosis, Microbial Epidemiology, and Antibiotic Treatment of Acute Otitis Media in Children:  A Systematic Review FREE

Quiz Ref IDAcute otitis media (AOM) is the most common childhood infection for which antibiotics are prescribed in the United States.^1- 3 A study using 2006 Medical Expenditure Panel Survey data demonstrated an average expenditure of $350 per child with AOM, totaling $2.8 billion.⁴ Timely and accurate diagnosis and appropriate management of AOM may have significant consequences for ambulatory health care utilization and expenditures.

Multiple systematic reviews on AOM diagnosis and management have been conducted,^5- 10 including our 2001 study,¹¹ which was the basis for the 2004 American Academy of Pediatrics and American Academy of Family Physicians joint practice guidelines.¹² Since then, new trials have been published, the heptavalent pneumococcal conjugate vaccine (PCV7) has become widely used, and clinician practice has changed regarding antibiotic choice for AOM.¹³

In light of these additional studies and practice changes, we conducted a systematic review to support the new AOM practice guidelines (currently in preparation) from the American Academy of Pediatrics. We report on the evidence for (1) the precision and accuracy of AOM diagnosis, (2) the association of PCV7 use with changes in AOM microbial epidemiology, (3) the decision about whether to treat with antibiotics, and (4) the comparative effectiveness of different antibiotics for uncomplicated AOM in average-risk children and associated antibiotic-related adverse events.

We searched PubMed, the Cochrane Controlled Clinical Trials Register Database, the Cochrane Database of Reviews of Effectiveness, and the Web of Science for articles published January 1999 through July 2010 on AOM diagnosis, treatment outcomes, and association of PCV7 use with changes in AOM microbiology using Medical Subject Headings terms (eg, otitis media, vaccines), key words (eg, diagnostic, microbiology, therapy), and individual antibiotic terms. This search supplemented a previous January 1966 through March 1999 search with additional key words for PCV7 and newer antibiotics.¹¹ We performed reference mining of relevant systematic reviews.

We included articles in any language that studied children aged 4 weeks to 18 years. We excluded studies on children with immunodeficiencies and craniofacial anomalies. Systematic reviews, randomized controlled trials (RCTs), controlled clinical trials, and observational studies were included in the initial search; case reports, clinical overviews, editorials, and practice guidelines were excluded.

Observational studies were considered for the PCV7 and diagnostic questions but excluded for the treatment question. For the PCV7 question, only articles that assessed AOM microbiology (using middle ear fluid) both before and after PCV7 implementation were included. For the diagnostic question, we considered studies of children that performed independent comparisons of signs or symptoms with a clear criterion standard; studies using clinicians in training were excluded. For the antibiotic comparative effectiveness question, only studies that examined clinical improvement as an outcome (not just microbiologic findings) were included. The search strategy and inclusion/exclusion criteria are detailed elsewhere.¹⁴

Two investigators (T.R.C., M.A.L.) independently reviewed titles and abstracts for potentially relevant articles. They then independently abstracted data from the full-text articles using structured review forms that included inclusion/exclusion criteria, outcome measures, and study quality. Disagreements were resolved by consensus; the principal investigators (P.G.S., G.S.T.) resolved remaining disagreements. The study biostatistician abstracted data (verified by a clinician investigator) for pooled analyses. One investigator independently abstracted treatment-related adverse event data.

We used the Jadad criteria to assess RCT quality,¹⁵ AMSTAR¹⁶ to assess systematic review quality, and QUADAS¹⁷ to assess diagnostic study quality.

For diagnostic studies, we report summary data, including sensitivities and specificities, when available. The number of studies was insufficient to allow pooling of data across studies. Furthermore, the criterion standards for the diagnostic studies varied widely.

For studies examining the association between PCV7 use and changes in AOM microbial epidemiology, we report summary data; the studies were too few in number and lacked enough consistency across study design and population for pooled analysis.

For treatment studies, an adequate number of articles was identified for pooled analyses of some comparisons. Comparisons were grouped by individual antibiotics rather than by antibiotic class to maximize the clinical relevance of our findings. The only a priori exception was to group ampicillin with amoxicillin because of similarity. When 3 or more articles examined the same comparison, we used the DerSimonian and Laird random-effects model to pool rate differences across studies.¹⁸ Sensitivity analysis was performed for pooled significant findings.

For pooled estimates, we report the I² statistic and P value for the χ² test of heterogeneity, which tests the null hypothesis that individual study results are homogeneous.^19- 20I² values near 100% represent high degrees of heterogeneity. For assessment of publication bias in our pooled analyses, we report the Egger asymmetry test.

We used Stata version 10.0 to perform the meta-analyses.²¹ The study received a waiver of institutional review board review from the RAND Human Subjects Protection Committee.

The literature searches and reference mining yielded 8945 titles. After removal of duplicates and clearly irrelevant titles, 738 went for further review. After 2 rounds of screening, 55 articles were accepted and combined with 80 articles identified from the 2001 systematic review.¹¹ These included 4 articles (3 research articles plus 1 systematic review) on diagnosis, 6 on PCV7-microbiology, and 125 on antibiotic treatment (Figure 1).

In clinical practice, 3 criteria are used to diagnose AOM: (1) acute symptoms of infection, (2) evidence of middle ear inflammation, and (3) presence of middle ear effusion (MEE).¹² Published research focuses on what constitutes acute symptoms of infection and what physical findings are associated with middle ear inflammation or effusion. A challenge with interpreting this research is the lack of a consistent gold standard, which varied from otolaryngologist-made diagnosis to tympanocentesis.

We identified 1 systematic review⁵ and 3 additional studies^22- 24 that addressed the question of diagnostic accuracy and precision in identifying any or all of the 3 criteria. Detailed data on these studies are available in our evidence report¹⁴; findings suggest that certain otoscopic signs are strongly associated with AOM, while data on the importance of symptoms as a predictor of AOM are less convincing.

Symptoms. A 2003 review by Rothman et al⁵ found that ear pain (sensitivities: 54%, 60%, 100%; specificities: 82% and 92%; positive likelihood ratio [LR], 3.0 [95% confidence interval {CI}, 2.1 to 4.3]; positive LR, 7.3 [95% CI, 4.4 to 12.1]) and ear rubbing (sensitivity: 42%; specificity: 87%; positive LR, 3.3 [95% CI, 2.1 to 5.1]) were modestly associated with AOM diagnosis. The review by Rothman et al included 4 studies examining specific symptoms among 965 total participants.^25- 28 In 2 of the studies, participants were recruited from otolaryngology practices and may not be representative of the general population of children with AOM. A more recent single study found that among 469 children aged 6 to 36 months presenting to primary care offices with parent-suspected AOM, AOM diagnosis was not associated with occurrence, duration, or severity of parent-reported symptoms (eg, ear rubbing, ear pain, fever).²⁴

Quiz Ref IDOtoscopic Signs. One study²⁹ examined in the review by Rothman et al assessed the association of otoscopic findings of middle ear inflammation (redness: positive LR, 8.4 [95% CI, 7 to 11]) and effusion (cloudy: positive LR, 34 [95% CI, 28 to 42]; bulging: positive LR, 51 [95% CI, 36 to 73]; immobile: positive LR, 31 [95% CI, 26 to 37]) with AOM (determined by clinical symptoms and the presence of MEE on tympanocentesis).⁵

A study published subsequently to the review by Rothman et al examined the accuracy of tympanometric (evaluation of middle ear function by measurement of acoustic impedence) and otoscopic findings compared with tympanocentesis as the criterion standard to determine the presence of MEE. Among children with MEE on tympanocentesis, 97% had type B (abnormal) tympanometry results, and 100% had otoscopic examination findings consistent with AOM. These results may overestimate the accuracy of tympanometry, because the investigators performing otoscopy were not blinded to the tympanometry results, and the criterion standard of tympanocentesis was performed only when otoscopic or tympanometric findings suggested MEE.²²

In another study subsequent to the review by Rothman et al, 22% of AOM cases diagnosed by a general practitioner were concurrently diagnosed by an otolaryngologist as otitis media with effusion, viral otitis, or a normal tympanic membrane.²³

Six studies examined the association between PCV7 use and changes in AOM microbial epidemiology (Table). These studies fit into 2 categories: observational studies of AOM isolates both before and after the 2000 licensure of PCV7 and PCV7 efficacy RCTs examining AOM-related organisms.

Most studies found that Haemophilus influenzae became more prevalent as an AOM isolate and that Streptococcus pneumoniae became less prevalent although it remained important.^30- 31,33 In an observational study of children with persistent AOM or AOM with treatment failure, the proportion of S pneumoniae MEE isolates decreased (from 44% in 1998-2000 to 31% in 2001-2003, P = .02), while the proportion of H influenzae isolates increased (from 43% in 1998-2000 to 57% in 2001-2003, P = .01).³⁰ Another study found an increase (as a proportion of all S pneumoniae isolates) in nonvaccine serotype S pneumoniae and a decrease in vaccine serotype S pneumoniae (non-PCV7 S pneumoniae: from 12% in 1999 to 32% in 2002, P < .01).³² In a vaccine-efficacy RCT, investigators found a greater proportion of S pneumoniae isolates in the control group (33%) than in the PCV7 group (23%) (P < .001).³⁴ It is important to note that study findings did not always reach statistical significance, and most studies focused on patients with severe or persistent AOM.

One hundred twenty-five articles compared the effectiveness of antibiotic treatment options in uncomplicated AOM. Older articles that examined antibiotics no longer typically used for AOM are not discussed here but are included in the evidence reports.^14,36

Evidence about the benefits of treating with antibiotics comes from 2 types of studies: placebo-controlled studies of immediate use of antibiotics and studies comparing immediate use of antibiotics with a strategy of observation with possible delayed treatment (“wait-and-see” or “prescription-to-hold”).

We identified 8 studies that compared ampicillin or amoxicillin with placebo. One did not report clinical success (only pain resolution) and was not included in the pooled analysis.³⁷

Quiz Ref IDIn pooled analysis of the remaining 7 RCTs, the random-effects pooled rate difference for success by day 14 was estimated at 12% (95% CI, 5% to 18%), with a 73% success rate for ampicillin/amoxicillin and a 60% success rate for placebo. The number needed to treat (NNT) for clinical success was 9 (95% CI, 6 to 20) (eTable 1 and Figure 2). The more recent, higher-quality studies reported smaller benefits. The I² statistic was 69% (P = .04 by χ² test of heterogeneity), indicating the presence of unexplained heterogeneity, which could be attributable to differences in the populations studied, research methods used, or both. The Egger test did not suggest publication bias (P = .77).

In sensitivity analyses, we excluded an outlier³⁸ because its 95% CI favored placebo far more strongly than any other individual study. The pooled analysis with the remaining 6 articles yielded rates of 70% vs 54% (pooled rate difference, 13% [95% CI, 8% to 19%]), with 7 children (95% CI, 5 to 12) needing treatment with ampicillin/amoxicillin to gain a case of clinical success (I² = 62%, P = .09; P = .18 by Egger test).^39- 44 When we pooled the 4 studies with a quality score of 3 or more (of 5), excluding the outlier, the rates were 76% vs 67% (pooled rate difference, 10% [95% CI, 6% to 14%]), and the NNT was 10 (95% CI, 7 to 18), without evidence of heterogeneity (I² = 0.0%, P = .48) or publication bias (P = .26).^41,43- 44

We identified 5 studies that compared other antibiotics with placebo (eTable 1), but they are not included in pooled analysis because we examined the overall benefit of antibiotics more commonly prescribed for AOM (ie, amoxicillin) over placebo.

We identified 4 studies of delayed treatment approaches; 2 reported higher rates of clinical success with immediate compared with delayed use of antibiotics,^45- 46 and 2 found no difference.^47- 48 One article reported rates of 95% vs 80% (rate difference, 15% [95% CI, 6% to 24%]; NNT, 7 [95% CI, 4 to 17]) favoring amoxicillin over the wait-and-see approach for parent-perceived success at day 12,⁴⁹ whereas the other reported rates of 86% vs 70% (rate difference, 16% [95% CI, 6% to 26%]; NNT, 6 [95% CI, 4 to 17]), also favoring amoxicillin over the prescription-to-hold approach for parent-perceived clinical success at day 3.⁴⁵ Thirty-four percent⁴⁶ and 24%⁴⁶ of participants in the delayed antibiotic groups in these studies received delayed antibiotics, respectively.

The risk of harms from antibiotic treatment for AOM has been less well studied than the benefits. Four of the 7 placebo-controlled studies reported on harms. One reported the counterintuitive, although not statistically significant, result of more cases of rash and diarrhea in placebo-treated patients than in amoxicillin-treated patients.⁴⁴ Pooled analysis of the other 3 trials^39,41,43 yielded rates of 13% vs 8% for diarrhea (pooled rate difference, 5% [95% CI, 0% to 10%]; I² = 23%; P = .30), while 2 individual studies had a rate difference of 4% (4% vs 0%) and 3% (8% vs 5%) for rash^39,41; these differences did not reach statistical significance. These point estimates are compatible with published estimates of the rate of rash (3%-10%) and diarrhea (5%-10%).^50- 53 In the studies by Little et al⁴⁵ and Spiro et al,⁴⁷ the rate of diarrhea was higher for the antibiotic group than for the prescription-to-hold group (19% vs 9%; rate difference, 10% [95% CI, 2% to 18%] and 23% vs 8%; rate difference, 14% [95% CI, 6% to 22%] for the 2 studies, respectively), with a number needed to harm (NNH) of 10 (95% CI, 6 to 50) and 7 (95% CI, 5 to 17), respectively.^45,47

McCormick et al⁴⁹ reported no difference in the rate of antibiotic-related adverse events, and Neumark et al⁴⁸ did not examine adverse events. In RCTs comparing amoxicillin with other antibiotics, the proportion of amoxicillin-treated children reporting rash ranged from 2% to 11% and the proportion reporting diarrhea ranged from 3% to 16%.^54- 60

None of the studies evaluated the rates of longer-term adverse effects of immediate antibiotic treatment, including antibiotic resistance.

eTable 2 describes selected antibiotic comparative effectiveness studies and pooled analyses for comparisons examining 3 or more studies. The Egger test was not suggestive of publication bias for any of the pooled analyses.

The success rate differences were statistically nonsignificant in the pooled analyses comparing ampicillin/amoxicillin vs ceftriaxone (4 trials, I² = 50.7%), ampicillin/amoxicillin vs cefixime (4 trials, I² = 22.9%), ampicillin/amoxicillin vs cefaclor (4 trials, I² = 13.0%), amoxicillin-clavulanate vs ceftriaxone (5 trials, I² = 22.9%), cefaclor vs azithromycin (3 trials, I² = 0%), and amoxicillin-clavulanate vs 5 days of azithromycin (5 trials, I² = 62.2%) and vs 3 or fewer days of azithromycin (7 trials, I² = 84.1%).

Statistically significant differences between treatment regimens were found in a few individual studies. Amoxicillin-clavulanate was superior to cefaclor (97% vs 84%; rate difference, 13% [95% CI, 5% to 21%])⁶¹; 10 days of amoxicillin-clavulanate was superior to 5 days of azithromycin (86% vs 70%; rate difference, 16% [95% CI, 2% to 30%])⁶²; 5 days of amoxicillin-clavulanate was not as effective as 7 to 10 days (77% vs 88%; rate difference, −11% [95% CI, −20% to −3%] in the study by Cohen et al⁶³ and 71% vs 87%; rate difference, −16% [95% CI, −24% to −8%] in the study by Hoberman et al⁶⁴); and 5 days of ceftibuten was not as effective as 10 days of ceftibuten (78% vs 98%; rate difference, −20% [95% CI, −28% to −12% ]).⁶⁵

In the pooled comparisons, use of ampicillin/amoxicillin resulted in a lower rate of diarrhea than cefixime (14% vs 21%; rate difference, −8% [95% CI, −13% to −4%]; NNH, 12 [95% CI, 8 to 25]; I² = 0%), and use of amoxicillin-clavulanate resulted in a higher rate of diarrhea than 1 dose of ceftriaxone (20% vs 9%; rate difference, 11% [95% CI, 7% to 16%]; NNH, 9 [95% CI, 6 to 15]; I² = 10.8%) and higher rates of any adverse event compared with 5 days of azithromycin (26% vs 9%; rate difference, 16% [95% CI, 7% to 25%]; NNH, 6 [95% CI, 4 to 14]; I² = 81.9%).

We identified several important findings for AOM diagnosis, microbiology, and antibiotic management.

Acute otitis media is a clinical diagnosis with 3 components: acute signs of infection, evidence of middle ear inflammation, and effusion.¹² Evidence suggests that certain otoscopic findings (ie, a red and immobile or bulging tympanic membrane) predict AOM, but the accuracy or precision of a clinical diagnosis has not been determined. A major limitation to the evidence regarding diagnosis is the lack of a gold standard. The diagnostic tools studied (eg, otoscopy) are often part of the only available gold standard—a clinical diagnosis. Perhaps the most important way to improve diagnosis is to increase clinicians' ability to recognize and rely on key otoscopic findings.

Since the introduction of PCV7, there have been significant shifts in AOM microbiology, with S pneumoniae becoming less prevalent and H influenzae becoming more prevalent. A recent study of a single pediatric practice found evidence suggesting that this balance may be shifting again because of an increase in the proportion of AOM with nonvaccine S pneumoniae serotypes.⁶⁶ These data and the introduction of PCV13 support the need for ongoing surveillance of AOM isolates.

Immediate ampicillin/amoxicillin treatment has a modest benefit compared with placebo or delayed antibiotics but also may be associated with more diarrhea and rash. Quiz Ref IDOf 100 average-risk children with AOM, approximately 80 would likely get better within about 3 days without antibiotics.⁶⁷ If all were treated with immediate ampicillin/amoxicillin, an additional 12 would likely improve, but 3 to 10 children would develop rash and 5 to 10 would develop diarrhea. Clinicians need to weigh these risks (including possible long-term effects on antibiotic resistance) and benefits before prescribing immediate antibiotics for uncomplicated AOM.

Most antibiotics used to treat uncomplicated AOM in children at normal risk have similar rates of clinical success; we found no evidence of the superiority of any other antibiotic over amoxicillin.

Quiz Ref IDIn most cases of uncomplicated AOM when amoxicillin is appropriate (eg, excluding children with penicillin allergy and those who previously did not improve after a course of amoxicillin), there is no evidence for first-line use of higher-cost antibiotics (eg, cefdinir, cefixime). For a 20-kg child with AOM, a 7-day course of cefdinir costs approximately $96, compared with $34 for an equivalent course of amoxicillin (pricing information available at http://www.drugstore.com). In an analysis of data from the National Ambulatory Medical Care Survey, among visits for AOM (visits for a new problem without additional diagnoses requiring antibiotic therapy), amoxicillin was prescribed in 49%, amoxicillin-clavulanate in 16%, cefdinir in 14%, and other cephalosporins in 6%.¹³ If just half of the 14% of the estimated 8 million children who visit a physician for AOM annually⁴ were to receive amoxicillin instead of cefdinir (assuming the other half were appropriately prescribed cefdinir because of a non–type-1 penicillin allergy), the estimated annual savings would exceed $34 million. This estimate does not account for potential additional savings from adopting a less aggressive approach to antibiotic prescribing that might avoid a certain number of prescriptions altogether.

This review has several limitations that must be considered. First, article screening and data abstraction were not blinded, which may potentially introduce bias. However, there is evidence that blinding does not alter the results of meta-analyses.⁶⁸ Second, we may not have identified some relevant evidence. For example, we did not search EMBASE or seek unpublished data. We used statistical tools to detect publication bias but found no evidence of it in our pooled analyses. Additionally, our findings on diagnosis and microbiology are greatly limited by the small number of studies; thus, caution should be used in interpreting our findings for these topics. To account for variation in study quality, we performed sensitivity analyses that pooled only high-quality studies. Third, the studies varied widely in their definitions of clinical success and in AOM diagnostic criteria. Some studies that did not use all 3 AOM diagnostic criteria may have included participants without AOM but with otitis media with effusion or no middle-ear abnormality at all.^69- 70 Lastly, our pooled analyses included studies completed before and after the licensing of PCV7. It is not clear how the changing microbiology of AOM may have influenced study findings; the heterogeneity of AOM over the past 20 years might favor an analysis that does not include pooling data from studies before and after 2000.⁶⁹

One remaining question is what new evidence about antibiotic comparative effectiveness is needed. It is not enough to show statistical significance or lack thereof; the clinical importance of any difference must also be considered. This requires knowing the minimal clinically important difference (MCID) for treatment of AOM. Although there currently is no agreed-on value for the MCID, assuming an MCID of 5% (representing a “small” effect size, according to Cohen's classification⁷¹) means that when existing evidence falls entirely within or outside of this MCID, equivalence or significance can be concluded; when it does not, it can be concluded that more information is needed. Using this definition, we can conclude equivalence for 2 of the 8 pooled analyses in eTable 2 and that effects are inconclusive for the remaining 6. The MCID has important implications for our conclusions; for example, in contrast to a previous systematic review,⁷ we are unable to make definitive conclusions regarding the equivalency of short- vs long-term regimens analyzed by antibiotic when considering an MCID of 5%, except for 7 to 10 days of cefaclor vs 3 days of azithromycin.

To account for both statistical and clinical significance, sample sizes for AOM comparative effectiveness studies need to be large. Because approximately 80% of AOM cases resolve spontaneously,⁶⁷ most RCTs will be able to test superiority of different antibiotics with only the remaining 20%. If the success rate is 88% for the treatment group and 80% for the control group, a sample size of 1150 per group would provide a 95% CI of the difference of 5% to 11%, which is outside the ±5% MCID; this sample size is much larger than that of any published AOM comparative effectiveness study.

We found evidence to guide the diagnosis and management of AOM in children; however, further research is needed that (1) examines clinicians' diagnostic accuracy and precision using the 3 AOM diagnostic criteria; (2) continues surveillance of AOM microbiology, especially in view of the newly approved PCV13; and (3) produces more high-quality studies on AOM management that include clear diagnostic criteria, a better-defined menu of clinical success measures that are universally applied, and more investigation into the comparative antibiotic-related adverse event rates that assesses whether any antibiotic regimen is superior to amoxicillin.