2. US Preventive Services Task Force. Guide to Clinical Preventive Services. 2nd ed. Washington, DC; Office of Disease Prevention and Health Promotion: 1996

3. Harris RP, Helfand M, Woolf SH, et al. Current methods of the US Preventive Services Task Force: a review of the process. Am J Prev Med. 2001;20(3 suppl):21–35

doi:10.1542/peds.2005-2783

To the Editor.—

We would like to comment on the meta-analysis1 re- cently published in Pediatrics that focuses on the relation- ship between the use of a pacifier and sudden infant death syndrome (SIDS), as well as the American Acad- emy of Pediatrics policy statement2 that recommends the use of a pacifier as a possible preventive factor of sudden death.

As previously highlighted,3 methodologic reasons should advise against giving any value to comparisons based on “last sleep.”

Indeed, last sleep is an exclusive feature of a dead infant. An infant who doesn’t die cannot experience the “last sleep before dying!” It seems to be quite clumsy to try to surrogate the absence of last sleep in the control group with a poorly convincing “reference sleep” (that is only 1 particular sleep among many and, therefore, should be considered “usual sleep”).

In contrast, it would be reasonable to make compar- isons based on common exposures (eg, gender, bottle or breastfeeding, parent’s smoking, usual sleeping position, bed sharing, pacifier use, etc).

Taking this approach as our inclusion criterion, we found only 5 studies that compared the use of a pacifier in cases versus controls (search was limited to PubMed and English-language articles).

The effect of pacifier use measured in terms of odds ratio (OR) and estimated by using both univariate and multivariate analyses (4 studies; OR was adjusted for different variables across studies) is smaller than that observed for the last sleep, thus suggesting caution when taking last sleep as a valid indicator.

Others elements of concern regard the reliability of the estimates obtained in the primary studies, such as the adjustment for several variables in relatively few subjects (the study with the biggest effect had 73 cases of SIDS) and the recruitment of several controls for each case, which tend to amplify the effect. The lack of an assessment of methodologic quality and the use of fixed models to combine results that do not take into account the high level of observed variability among results should suggest the use of these results with extreme caution.4

In the absence of clear and strong evidence in support of the use of pacifiers (which could have some adverse effects on breastfeeding), we think it would be more appropriate not to provide any recommendations.

Roberto Buzzetti, MD Roberto D’Amico, PhD Policlinico di Modena University of Modena and Reggio Emilia 41100 Modena, Italy

1850 LETTERS TO THE EDITOR

Downloaded from pediatrics.aappublications.org by guest on May 10, 2012

REFERENCES

1. Hauck FR, Omojokun OO, Siadaty MS. Do pacifiers reduce the risk of sudden infant death syndrome? A meta-analysis. Pediat- rics. 2005;116(5). Available at: www.pediatrics.org/cgi/content/ full/116/5/e716

2. American Academy of Pediatrics, Task Force on Sudden Infant Death Syndrome. The changing concept of sudden infant death syndrome: diagnostic coding shifts, controversies regarding the sleeping environment, and new variables to consider in reduc- ing risk. Pediatrics. 2005;116:1245–1255

3. Buzzetti R, D’Amico R, Liberati A. Sudden infant death syn- drome and sleeping position [letter]. Pediatrics. 2001;108:211 4. Dwyer T, Couper D, Walter SD. Sources of heterogeneity in the

meta-analysis of observational studies: the example of SIDS and sleeping position. J Clin Epidemiol. 2001;54:440–447

doi:10.1542/peds.2005-3035

In Reply.—

We thank the authors for their interest in our article and their thoughtful comments.

We appreciate the updated US Preventive Services Task Force (USPSTF) grading scale and matrix provided by Drs Calonge and Petitti.1 We had not been aware of the newer scale when we conducted our meta-analysis, and thus our use of the 1996 methodology was an unintentional oversight. However, after carefully re- viewing the methodology used currently by the USPSTF, we believe that our pacifier recommendation2 as a po- tential preventive strategy against sudden infant death syndrome (SIDS) would retain the B grading: fair to good strength of evidence of moderate net benefit. As stated in our article, we used the USPSTF taxonomy to classify our recommendation; the task force was in no way involved in the study or its conclusions.

We feel that our review of the evidence was “com- prehensive, objective, and attentive to quality.”1 We conducted a comprehensive literature search beginning at 1966 and used an analytic framework for evaluating the quality of each study; we excluded from the meta- analysis any studies that were of lower quality. We included studies that measured the associations between pacifier use (the preventive service) and the health out- come of interest (ie, occurrence of SIDS) rather than intermediate outcomes. Because the best studies avail- able used case-control designs, these would be classified under the hierarchy of research design as level II-2. Using the criteria for grading internal validity of individ- ual studies would qualify them as having “good” internal validity.

Next, we examined the quality of the full body of evidence. Here, we found that the evidence for pacifier use at last sleep is highly consistent and internally and

externally valid (generalizable). We also believe that the evidence is coherent, that is, makes sense on the basis of hypothesized mechanisms that may explain the appar- ent protective effect of pacifiers.3,4 However, researchers still do not fully understand the cause of SIDS; thus, the mechanism to explain the observed association between pacifiers and reduced risk of SIDS is currently unknown.

It is difficult to fully assess the existence and magni- tude of a causal connection between the preventive service and outcome, because the best evidence available is from observational studies. As noted in our article, “causality” is supported by fulfillment of several of the criteria proposed by Sir Bradford Hill.5 It is unlikely that a randomized, controlled trial of pacifier use and SIDS would or could be conducted.

Finally, the USPSTF methodology calls for assessing the magnitude of the net benefit, that is, weighing the potential benefits against the potential harms of the preventive service. We carefully reviewed the literature that addressed all the potential adverse effects of pacifi- ers in infants, including breastfeeding duration, otitis media, other infections, and dental malocclusions. Al- though there has been considerable debate about the potential negative consequences of pacifier use, we feel that, overall, the current literature does not support these fears for pacifiers to be used as we recommended and believe that the benefit may be substantial: “a large impact on an infrequent condition that poses a signifi- cant burden at the individual patient level.”1

Drs Buzzetti and D’Amico advise against placing value on comparisons based on last sleep, indicating that it is “only 1 particular sleep among many.” However, we would posit that last sleep is the more appropriate sleep period to use. SIDS is believed to be a multifactorial disorder requiring a set of circumstances to be present. Various factors contribute to the risk of dying as a result of SIDS; therefore, it is logical that some of these factors may change during different sleep periods on the basis of location of sleep, recent illness, items in the bed, use of pacifiers, sleep position, etc. Thus, proximal or “trigger” factors are likely to be more important in the causal pathway.

Although it is only the case infant in case-control studies who experienced a “last sleep,” matching to a similar sleep period for the control infant provides a random sleep period (for that infant), which should appropriately represent a comparison period. “Usual” sleep does not necessarily describe accurately what hap- pened during the sleep period in which the SIDS death occurred. As an example to illustrate this point, if some- one is injured in a car accident, it is important to know if that person was wearing his or her seat belt for that car trip, not for “usual” car trips.

As noted in the article that is referenced by Buzzetti and D’Amico in which Dwyer et al analyze possible sources of heterogeneity in the meta-analysis of obser-

vational studies, “studies using ‘usual’ sleep position to define exposure had low effect sizes [odds ratios] com- pared with studies using ‘found’ or ‘placed/put’ posi- tions. This is the pattern of results that would be ex- pected if one sleeping position (eg, the prone position) in fact conveyed higher risk, but it was not used all the time. If the sleeping period varies for a given child, then adopting a ‘usual’ definition of sleeping position in a study would attenuate any observable effect associated with the particular sleeping position that elevates risk.”6 In our analysis of the association between pacifiers and SIDS, we examined both usual and last sleep periods and found that the effect size was lower when used for usual sleep compared with last sleep, but it was still present. We believe this result was a result of the variable nature of “usual” practice across studies as well as among indi- vidual infants.

Their other concern relates to the reliability of esti- mates in the primary studies, such as multiple adjust- ments in “relatively few subjects” and the recruitment of several controls for each case. In calculating the odds ratios in our meta-analysis, the sample sizes of the stud- ies were taken into consideration through weighting on the basis of the confidence intervals. Using multivariate odds ratios provides a more accurate picture, because it takes into account, to the extent possible, other poten- tially confounding factors. The use of multiple controls for each case, particularly in smaller studies, is a well- accepted methodology in case-control study design spe- cifically intended to increase the precision with which the odds ratio is estimated.7 It will not amplify the effect but, rather, will reduce the chance of a type II error. Some of medicine’s greatest discoveries have been based on extremely small case-control studies.8 This would assume, however, that the estimate is correct, and we believe this to be the case on the basis of the following: (1) we included studies in our analysis that met 5 or 6 of the 6 evaluation criteria and passed peer review; and (2) the odds ratios reported, especially for last sleep, were quite consistent.

Finally, it would be helpful to investigate the reasons for heterogeneity across studies. We attempted to do this through an improved analysis technique using the pro- portional odds ratio model.9 In contrast to the fixed- effects model (which pools odds ratios), the proportional odds ratio model estimates effect sizes that are more generalizable. Unfortunately, the necessary data for var- ious covariates were not reported consistently in all the articles, making it impossible to conduct this analysis. In the absence of these data, the fixed-effect model is a valid method.

Drs Stuebe and Lee express several concerns about the relationship between pacifiers and SIDS protection. Some of the same case-control studies that served as the basis for our recommendation have served as the basis for the well-accepted recommendation to place infants

PEDIATRICS Volume 117, Number 5, May 2006 1851 Downloaded from pediatrics.aappublications.org by guest on May 10, 2012

supine for sleep.10 This recommendation has proven to be highly successful.11,12 To conduct prospective cohort studies or clinical trials on any of the SIDS risk factors would be highly infeasible, because the occurrence of SIDS is relatively rare (