Public Health Nursing Vol. 32 No. 6, pp. 731–737 0737-1209/© 2015 Wiley Periodicals, Inc. doi: 10.1111/phn.12200
SPECIAL FEATURES: METHODS
Bed Sharing, SIDS Research, and the Concept of Confounding: A Review for Public Health Nurses Elizabeth M. Keys, PhD Student, BN, RN and James A. Rankin, PhD, NP Faculty of Nursing, University of Calgary, Calgary, Alberta, Canada Correspondence to: Elizabeth Keys, PhD Student, Faculty of Nursing, Professional Faculties Building, 2500 University Drive NW, Calgary, Alberta, Canada T2N 4N1. E-mail: [email protected]
ABSTRACT Confounding is an important concept for public health nurses (PHNs) to understand when considering the results of epidemiological research. The term confounding is derived from Latin, confundere, which means to mix-up or mix together. Epidemiologists attempt to derive a cause and effect relationship between two variables traditionally known as the exposure and disease (e.g., smoking and lung cancer). Confounding occurs when a third factor, known as a confounder, leads to an over- or underestimate of the magnitude of the association between the exposure and disease. An understanding of confounding will facilitate critical appraisal of epidemiological research findings. This knowledge will enable PHNs to strengthen their evidence-based practice and better prepare them for policy development and implementation. In recent years, researchers and clinicians have examined the relationship between bed sharing and sudden infant death syndrome (SIDS). The discussion regarding the risk of bed sharing and SIDS provides ample opportunity to discuss the various aspects of confounding. The purpose of this article is to use the bed sharing and SIDS literature to assist PHNs to understand confounding and to apply this knowledge when appraising epidemiological research. In addition, strategies that are used to control confounding are discussed. Key words: bed sharing, confounding, epidemiological research, evidence-based, nursing practice, public health nursing, sudden infant death.
A major focus for public health nurses (PHNs) is on risk reduction and primary prevention (Hemingway, Aarts, Koskinen, Campbell, & Chasse, 2013). The factors associated with sudden infant death syndrome (SIDS) represent significant health risks to infants. In 2004, an international panel of SIDS experts refined the definition of SIDS as: The sudden unexpected death of an infant less than 1 year of age, with onset of the fatal episode apparently occurring during sleep, that remains unexplained after a thorough investigation, including performance of a complete autopsy and review of the circumstances of death and the clinical history. (Krous et al., 2004, p. 235)
The Triple Risk Hypothesis (Guntheroth & Spiers, 2002; Rognum & Saugstad, 1993) is the cur-
rent guiding framework for SIDS etiology. While a complete explanation of this etiological framework is beyond the scope of this article, in brief, the Triple Risk Hypothesis posits that infants succumb to SIDS when three factors intersect: (a) the infant has inherent risk (nonmodifiable risk factors such as genetic or brainstem abnormalities); (b) the infant is within a vulnerable developmental and age-related period; and (c) modifiable risk factors are present (such as parental smoking or soft bedding; Guntheroth & Spiers, 2002; Rognum & Saugstad, 1993). Early epidemiologic findings indicated the infant’s sleeping position as a significant modifiable risk factor for SIDS (Mitchell, 1993). Using the case-control research method investigators around the world have consistently demonstrated that
731
732
Public Health Nursing
Volume 32
Number 6
infants who are placed in a prone position for sleep are 1.3 to 7.08 times more likely to succumb to SIDS than those placed in a supine position (Blair et al., 2009; Colvin, Collie-Akers, Schunn, & Moon, 2014; Mitchell, Thach, Thompson, & Williams, 1999; Vennemann et al., 2009). In the United States, “Back to Sleep” campaigns have reduced the rate of SIDS by at least 50% (American Academy of Pediatrics Task Force on Sudden Infant Death Syndrome, 2005). Since this dramatic decrease, researchers have focused on identifying additional modifiable risk factors to facilitate further reduction of SIDS. Some of these factors include parental smoking, drug and alcohol use, the infant’s sleep environment and location within the home (Mitchell, 2009; Moon, Horne, & Hauck, 2007). Current infant sleep guidelines recommend that parents avoid the practice of bed sharing with their infant due to an increased SIDS risk (Mitchell, Freemantle, Young, & Byard, 2012; Task Force on Sudden Infant Death Syndrome, 2011). Despite these recommendations, the practice of bed sharing is common (Ateah & Hamelin, 2008; Nie, Bailey, Istre, & Anderson, 2010). Consequently, researchers have examined the relationship between bed sharing and SIDS in more detail. PHNs play a key role in assisting parents to identify and act on modifiable SIDS risk factors. The attention to risk reduction can sometimes appear to conflict with other elements of family wellness and autonomy. Specifically, PHNs may encounter tension between reducing SIDS risk and supporting breastfeeding mothers who often practice bed sharing (Fetherston & Leach, 2012). Understanding and appraising the epidemiological evidence that informs current safe sleep recommendations will assist in navigating these tensions. One of the most important factors in interpreting epidemiological research is understanding the concept of confounding (Gordis, 2014). Confounding is defined as “the distortion of a measure of effect of an exposure on an outcome due to the association of the exposure with other factors that influence the occurrence of the outcome” (Porta, 2014, p. 55). In other words, given a particular exposure, confounding may bias the magnitude of the effect of an association on an outcome. Confounding can cause: (a) an overestimate of an effect which leads researchers to conclude there is a difference when there is none; or (b) an underestimate of an effect which leads
November/December 2015
researchers to conclude there is no difference when there actually is one (Elwood, 2007; Pearce & Greenland, 2014). The purpose of this article is to explain confounding and facilitate PHNs’ application of this knowledge when appraising epidemiological research. Critical appraisal of epidemiological studies will help PHNs assess the applicability of research findings to their own practice populations. This ability will support an evidence-based practice and further PHNs’ capacities for assuming roles in policy development and implementation. The authors use examples from the bed sharing and SIDS research literature to explain confounding. Strategies used to control confounding are explained and the implications of understanding confounding for nursing practice are summarized.
What is Confounding? In 1854, Dr. John Snow demonstrated a classic epidemiological example of exposure and disease outcome in Soho, London. Snow, one of the founders of modern epidemiology, identified the causal relationship between a contaminated water pump in Broad Street (the exposure) and the outbreak of cholera (the outcome; Paneth, 2004). The causal relationship between contaminated water and disease is relatively straightforward in contrast to the many predisposing factors that are cited to be associated with the etiology of “modern” diseases. In SIDS, several predisposing factors have been identified (Task Force on Sudden Infant Death Syndrome, 2011). Present-day epidemiological researchers examine a wide range of biological, social, and economic variables and their association with disease (Brownson & Hoehner, 2006). In its simplest form, confounding arises when a third variable is thought to influence the result of an observed association between an exposure and an outcome (Elwood, 2007; Pearce & Greenland, 2014). The third variable is commonly referred to as a potential confounder until it has been demonstrated that it influences the result of the observed association; at which time it becomes an actual confounder. In order for a variable to be considered a confounder the following criteria must be met: (a) the variable must be associated with the exposure; (b) independent of the association with the exposure, the variable must be associated with the
Keys and Rankin: Confounding Bed Sharing SIDS outcome or disease; and (c) the confounder must not be an intermediate step in the causal pathway between exposure and disease (Olsen, Christensen, Murray, & Ekbom, 2010). Risk factors for SIDS represent different types of exposure, with SIDS as the outcome. Figure 1 shows how confounding occurs when a putative causal relationship between an exposure and an outcome is influenced by a third variable: the confounder (Elwood, 2007). It is important to emphasize that the confounder is not part of the proposed causal pathway (Olsen et al., 2010). Rather, the confounder is independently associated with the exposure and the outcome. Consider the association between infants who bed share and the occurrence of SIDS (Vennemann et al., 2012); the relationship may be influenced by a potential confounder (Figure 2). Maternal age is considered a potential confounder as mothers under the age of 20 years are 1.55 (95% confidence interval [CI] = 1.08–2.24) times more likely to bed share (Broussard, Sappenfield, & Goodman, 2012). Researchers have subsequently demonstrated that mothers less than 18 years increases the risk of SIDS nine-fold (95% CI = 5.9–14.1; Carpenter et al., 2013). Maternal age is not part of the proposed causal path and it has been shown to modify the effect of bed sharing on SIDS; it is therefore no longer considered a potential confounder, rather it is now considered to be a confounder. When the confounder is positively associated with the outcome and the exposure, as in this
Figure 1. The Effect of Confounding on the Association between Exposure and Outcome
733
Figure 2. Maternal Age as a Confounder in the Bed Sharing-SIDS Relationship younger maternal age example, the magnitude of the exposure-outcome relationship is overestimated (Elwood, 2007). In this example, the true risk of SIDS when bed sharing may be artificially amplified by positive associations with younger maternal age. Depending on the strength of these associations, the relationship between bed sharing and SIDS could be partially explained by maternal age. Given that confounding can influence the actual risk of negative health outcomes, it is important for PHNs to consider situations where potential confounders may artificially inflate putative causal relationships. A confounding variable may reduce the magnitude of the association between the exposure and the outcome (Elwood, 2007). As a result, the confounder can reduce the exposure-outcome association with the extent that it may not even be detected (Elwood, 2007). For instance, breastfeeding has been shown to have a protective effect by reducing the risk of SIDS by 45% (Hauck, Thompson, Tanabe, Moon, & Vennemann, 2011). However, mothers who breastfeed are 2.3 (95% CI = 1.87–2.89) times more likely to bed share (Blair, Heron, & Fleming, 2010). In this situation, the magnitude of the protective effect of breastfeeding on SIDS may be sufficiently large to reduce the association between bed sharing and SIDS. Nevertheless, when the confounding (in this case, protective effect) of breastfeeding is controlled in the analysis, there remains a 5.1 (95% CI = 2.3– 11.4) increased risk of SIDS for those that bed share (Carpenter et al., 2013). It is important to be aware that the terms confounding and bias are on occasion used interchangeably (Olsen et al., 2010). A variable can only be truly confounding when it has an independent relationship with the exposure and the outcome (Law, Green, & Ellison, 2012). In other words, if a potentially confounding variable is only associated with either the exposure or outcome but not both, it is not a confounder.
734
Public Health Nursing
Volume 32
Number 6
For instance, maternal smoking is associated with a 2.7 (95% CI = 1.0–6.4) times increase in SIDS risk (Liebrechts-Akkerman et al., 2011). However, no relationship between maternal smoking and bed sharing has been reported. Hence, maternal smoking is not considered a true confounder in the bed sharing-SIDS relationship. Instead, maternal smoking is an effect modifier. An effect modifier changes the risk of a particular outcome (SIDS), but is not associated with the exposure of interest (bed sharing; Olsen et al., 2010). Confusing an effect modifier for a confounding variable can interfere in the development of accurate causal models (Olsen et al., 2010).
Identifying Potential Confounders PHNs should reflect on their practical knowledge of their specific practice populations to assess whether researchers address potential confounders. This appraisal assists PHNs to determine the generalizability of the research findings for the communities with whom they work. A preliminary step in identifying potential confounders is to examine the literature for associations that have been reported for both the exposure and outcome (Polit & Beck, 2012). For example, a PHN may notice that many families who bed share belong to a particular ethnic group. Before deciding that ethnicity is a confounding variable, an association between an ethnic group and SIDS would need to be identified. In fact, it has been demonstrated that ethnicity is a confounding variable in the association between bed sharing and SIDS. Findings from several studies indicate that families belonging to nonCaucasian races often have increased rates of bed sharing. For example, Ball et al. (2012) found that Pakistani-British mothers were 3.02 (95% CI = 1.96–4.66) times more likely to regularly bed share than Caucasian British mothers, while Colson et al. (2013) found that African-American and Hispanic-American mothers were 3.47 (95% CI = 2.97–4.05) and 1.33 (95% CI = 1.10–1.61) times, respectively, more likely to bed share compared to Caucasian American mothers. Meanwhile, Hogan (2014) has reported a statistically significant (p = .001) increased proportion of sleep-related infant deaths in infants of African-American mothers (62.5%), compared to those of Caucasian mothers (25.7%).
November/December 2015
Controlling Confounding There are several ways that researchers can control for confounding either by design or in the statistical analysis. Randomization, which refers to the random allocation of subjects in a study, into different groups is a powerful way to reduce the risk of confounding. Observational epidemiological studies are more vulnerable to the threat of confounding as investigators are unable to randomize retrospectively or cannot randomize due to ethical constraints (Gordis, 2014; Pearce & Greenland, 2014). For obvious ethical reasons researchers would not be permitted to randomly allocate infants to different groups to study SIDS; similarly this is why there has never been a randomized controlled trial (RCT) to examine the relationship between smoking and lung cancer in humans. As may be seen from Figure 3, the strategies used to control for confounding fall into two categories; researchers can control at the design stage or at the analysis stage of the study. Of course, researchers can only control for confounding at the analysis stage providing they have the foresight to collect data on potential confounders during the study (Olsen et al., 2010; Pearce & Greenland, 2014). For example, SIDS researchers are now more aware of collecting data on potential confounders such as socioeconomic status, ethnicity, and type of infant feeding, compared with earlier research (Carpenter et al., 2013; Dukes et al., 2014; Vennemann et al., 2012). Each of the strategies for controlling confounding (see Figure 3) is now briefly described.
Randomization As previously stated, randomization refers to the random allocation of subjects into groups. For example, in the context of a RCT researchers randomly allocate subjects to a treatment or control group. Theoretically, potential confounders have an equal chance of being evenly distributed between the groups (Grimes & Schulz, 2002; Pearce & Greenland, 2014). In other words, randomization into groups is a very powerful way to minimize the influence that confounding variables have on the study results. It is important to emphasize that randomization does not guarantee that confounding will never occur; rather the strategy effectively minimizes the likelihood that the strength of the association between exposure and outcome has not been influenced by confounding variables.
Keys and Rankin: Confounding Bed Sharing SIDS
735
Figure 3. Strategies for Controlling Confounding by Study Design and Data Analysis Randomization is the only strategy that is able to minimize the impact of unknown confounders (Grimes & Schulz, 2002). However, true randomization may not always be accomplished (Pearce & Greenland, 2014). Consequently, even investigators conducting RCTs use additional strategies to control confounding. These strategies are also used in observational studies where confounding is much more of a concern (Pearce & Greenland, 2014).
Matching Matching subjects on potential confounding variables, such as age and sex, occurs during the design phase of a study (Elwood, 2007; Pearce & Greenland, 2014). Maternal age, education level, ethnicity, socioeconomic status, and type of infant feeding are now known to be confounders in the bed sharing and SIDS relationship (Carpenter et al., 2013; Hogan, 2014; Vennemann et al., 2012). It is therefore reasonable to match participants based on these variables; however, matching for more than two or three variables can make it challenging to find appropriate controls (Pearce & Greenland, 2014). Thus, matching in the bed sharing and SIDS research is often limited to matching by age and geographic region (see Carpenter et al., 2004 for example). Furthermore, the process of matching cases for variables that are not true confounders transforms them into artificial confounders (Gordis, 2014; Pearce & Greenland, 2014). This transformation must then be controlled during the data analysis (Elwood, 2007; Pearce & Greenland, 2014). For example, although male infants have a slightly increased risk of SIDS compared to female infants (Moon et al., 2007), there is likely no relationship between the sex of the infant and bed sharing. In
this case, matching subjects based on infant sex would not be helpful in reducing confounding and, furthermore, may complicate the data analysis.
Restriction Restriction is used to control confounding by excluding subjects with potential confounding variables from the study or by excluding them in the analysis (Elwood, 2007). For example, in a casecontrol study of bed sharing and SIDS, Fu, Moon, and Hauck (2010) controlled for confounding due to African-American ethnicity by restricting the study to African-Americans. Restriction is an effective way to control for the effect of confounding, however by excluding subjects certain confounders, such as breastfeeding or low socioeconomic status, severely limits the generalizability of the results (Krickeberg, Pham, & Pham, 2012). Moreover, investigators who apply very restrictive inclusion criteria may have difficulty in recruiting a sufficiently large sample size for the study (Brownson & Hoehner, 2006). Stratification Stratification involves subdividing the subjects into smaller groups or “strata” based on different values of the potential confounder (Elwood, 2007; Parfrey & Barrett, 2009). The analysis of the association between the exposure and the outcome is then computed for each of the smaller groups; the researcher thereby obtains stratum specific results (Elwood, 2007). The ultimate sample size for each of the strata depends on the number of subdivisions made on the potential confounder (Pearce & Greenland, 2014). For example, for the potential confounder of sex there is usually only two subdivisions; however, con-
736
Public Health Nursing
Volume 32
Number 6
founding variables such as age or birth order can be divided into many more categories. In their metaanalysis, Carpenter et al. (2013) stratified the data to assess for confounding by infant feeding, maternal age, parity, ethnicity, and marital status. This technique can be limited by the precision with which the potential confounder is measured (Pearce & Greenland, 2014). In the Carpenter study, this limitation is exemplified in that breastfeeding (yes or no) and ethnicity (Caucasian or non-Caucasian) were measured as binary variables. On the other hand, stratifying a potential confounder into too many categories can result in some strata with no subjects (Olsen et al., 2010). For example, this may occur when maternal age is analyzed using the precise age in years (18, 19, 20 and so on) rather than combining subjects into age groups (under 18 years, 19–20 years, 21–25 years). From a statistical perspective, a more sophisticated approach to deal with confounding is to use stratification and multivariate analysis techniques in combination with one another (Pearce & Greenland, 2014).
Multivariate analysis Public health nurses may be aware of one type of multivariate analysis known as logistic regression; SIDS investigators conducting case-control studies control for confounding by using logistic regression analysis. Logistic regression involves complex statistical modeling that estimates the likelihood of an outcome given a particular exposure and controls for several confounding variables simultaneously (Elwood, 2007). For example, Carpenter et al. (2013) used logistic regression to calculate the risk of SIDS (the outcome) based on bed sharing (the exposure), and simultaneously controlled for variables such as parental smoking, mother’s age, infant feeding method, infant sex, birth weight, and parity. In case-control studies, the magnitude of the risk between an exposure and an outcome is expressed in terms of a statistic known as the odds ratio (OR). The cases have the disease (or outcome of interest) and the controls do not, both groups are then compared on the exposure. If a particular exposure increases the likelihood of an outcome, the OR will be greater than 1.0 (Laing & Rankin, 2011). For example, the OR of SIDS given maternal illicit drug use after birth is 11.5 (Carpenter et al., 2013). This means that, for the cases in the study, infants whose mothers used illicit drugs after birth
November/December 2015
(the exposure) were 11.5 times more likely to succumb to SIDS (the outcome). If a particular exposure reduces the likelihood of an outcome, the OR is less than 1.0 (Laing & Rankin, 2011). For instance, Hauck et al. (2011) found that infants in their study who received any amount of breast milk had an OR of 0.40 for SIDS. This means that these infants were 60% less likely to succumb to SIDS. An OR that is equal to 1.0 is the null value and indicates that there is no difference between the cases and the controls; the risk is the same in both groups (Laing & Rankin, 2011). An understanding of confounding facilitates the PHN’s ability to interpret the epidemiological research literature and to assess if confounding is clinically relevant. PHNs can become more effective practitioners when they develop their capacity to appraise the epidemiological evidence that informs their practice. In their day-to-day practice, PHNs are already adept at considering the interplay of multiple factors with regard to particular health outcomes. Understanding confounding extends this knowledge repertoire and builds capacity to examine the health risks of the communities that they serve. Furthermore, PHNs can use their understanding of confounding, in combination with their observations in practice, to critically reflect on the health risks of their populations. As a result, PHNs can be confident that they are focusing on reducing exposures and health risks that are truly responsible for negative health outcomes.
References American Academy of Pediatrics Task Force on Sudden Infant Death Syndrome (2005). The changing concept of sudden infant death syndrome: Diagnostic coding shifts, controversies regarding the sleeping environment, and new variables to consider in reducing risk. Pediatrics, 116(5), 1245–1255. doi:10.1542/peds.2005-1499. Ateah, C. A., & Hamelin, K. J. (2008). Maternal bedsharing practices, experiences, and awareness of risks. Journal of Obstetric, Gynecologic, & Neonatal Nursing, 37(3), 274– 281. doi:10.1111/j.1552-6909.2008.00242.x. Ball, H. L., Moya, E., Fairley, L., Westman, J., Oddie, S., & Wright, J. (2012). Bed- and sofa-sharing practices in a UK bioethnic population. Pediatrics, 129(3), e673–e681. Retrieved from http://pediatrics.aappublications.org/ Blair, P. S., Heron, J., & Fleming, P. J. (2010). Relationship between bed sharing and breastfeeding: Longitudinal, population-based analysis. Pediatrics, 126(5), e1119–e1126. doi:10.1542/peds.2010-1277. Blair, P. S., Sidebotham, P., Evason-Coombe, C., Edmonds, M., Heckstall-Smith, E. M. A., & Fleming, P. (2009). Hazardous cosleeping environments and risk factors amenable to change: Case-control study of SIDS in south west England. British Medical Journal, 339(7726), 1–11. doi:10.1136/ bmj.b3666.
Keys and Rankin: Confounding Bed Sharing SIDS Broussard, D. L., Sappenfield, W. M., & Goodman, D. A. (2012). The black and white of infant back sleeping and infant bed sharing in Florida, 2004-2005. Maternal and Child Health Journal, 16(3), 713–724. doi:10.1007/s10995-011-0768-y. Brownson, R. C., & Hoehner, C. M. (2006). Epidemiology. A foundation of public health. In R. C. Brownson, & D. B. Petitti (Eds.), Applied epidemiology: Theory to practice (pp. 1–28). Oxford: Oxford University Press. Carpenter, R., Irgens, L., Blair, P., England, P., Fleming, P., Huber, J., et al. (2004). Sudden unexplained infant death in 20 regions in Europe: Case control study. The Lancet, 363 (9404), 185–191. Retrieved from http://www.thelancet.com/ Carpenter, R., McGarvey, C., Mitchell, E. A., Tappin, D. M., Vennemann, M. M., Smuk, M., et al. (2013). Bed sharing when parents do not smoke: Is there a risk of SIDS? An individual level analysis of five major case-control studies. BMJ Open, 3(5), 1–11. Retrieved from http://bmjopen.bmj.com/ Colson, E. R., Willinger, M., Rybin, D., Heeren, T., Smith, L. A., Lister, G., et al. (2013). Trends and factors associated with infant bed sharing, 1993-2010: The national infant sleep position study. JAMA Pediatrics, 167(11), 1032–1037. doi:10.1001/jamapediatrics.2013.2560. Colvin, J. D., Collie-Akers, V., Schunn, C., & Moon, R. Y. (2014). Sleep environment risks for younger and older infants. Pediatrics, 134(2), e406–e412. doi:10.1542/peds.2014-0401. Dukes, K. A., Odendaal, H. J., Signore, C., Sullivan, L. M., Willinger, M., Wright, C., & PASS Research Network (2014). The safe passage study: Design, methods, recruitment, and follow-up approach. Paediatric and Perinatal Epidemiology, 28(5), 455–465. doi:10.1111/ppe.12136. Elwood, J. M. (2007). Critical appraisal of epidemiological studies and clinical trials. Oxford: Oxford University Press. Fetherston, C. M., & Leach, J. S. (2012). Analysis of the ethical issues in the breastfeeding and bedsharing debate. Breastfeeding Review, 20(3), 7–17. Retrieved from https:// www.breastfeeding.asn.au/bfreview Fu, L. Y., Moon, R. Y., & Hauck, F. R. (2010). Bed sharing among black infants and sudden infant death syndrome: Interactions with other known risk factors. Academic Pediatrics, 10(6), 376–382. doi:10.1016/j.acap.2010.09.001. Gordis, L. (2014). Epidemiology (5th ed.). Philadelphia, PA: Elsevier/Saunders. Grimes, D. A., & Schulz, K. F. (2002). An overview of clinical research: The lay of the land. The Lancet, 359(9300), 57– 61. doi:10.1016/S0140-6736(02)07283-5. Guntheroth, W. G., & Spiers, P. S. (2002). The triple risk hypotheses in sudden infant death syndrome. Pediatrics, 110(5), e64–e69. doi:10.1542/peds.110.5.e64. Hauck, F. R., Thompson, J. M., Tanabe, K. O., Moon, R. Y., & Vennemann, M. M. (2011). Breastfeeding and reduced risk of sudden infant death syndrome: A meta-analysis. Pediatrics, 128(1), 103–110. doi:10.1542/peds.2010-3000. Hemingway, A., Aarts, C., Koskinen, L., Campbell, B., & Chasse, F. (2013). A European Union and Canadian review of public health nursing preparation and practice. Public Health Nursing, 30(1), 58–69. doi:10.1111/j.1525-1446.2012.01048.x. Hogan, C. (2014). Socioeconomic factors affecting infant sleeprelated deaths in St. Louis. Public Health Nursing, 31(1), 10–18. doi:10.1111/phn.12052. Krickeberg, K., Pham, V. T., & Pham, T. M. H. (2012). Epidemiol˙ prevention. New ˙ York, NY: Springer New York. ogy: Key to doi:10.1007/978-1-4614-1205-2 Krous, H. F., Beckwith, J. B., Byard, R. W., Rognum, T. O., Bajanowski, T., Corey, T., et al. (2004). Sudden infant death syndrome and unclassified sudden infant deaths: A definitional and diagnostic approach. Pediatrics, 114(1), 234–238. Retrieved from http://pediatrics.aappublications.org/ Laing, C. M., & Rankin, J. A. (2011). Odds ratios and confidence intervals: A review for the pediatric oncology clinician. Journal of Pediatric Oncology Nursing, 28(6), 363–367. doi:10.1177/1043454211426575.
737
Law, G. R., Green, R., & Ellison, T. H. (2012). Confounding and causal path diagrams. In Y. Tu, & D. C. Greenwood (Eds.), Modern methods for epidemiology (pp. 1–13). Dordrecht, Netherlands: Springer Science & Business Media. doi: 10.1007/978-94-007-3024-3 Liebrechts-Akkerman, G., Lao, O., Liu, F., van Sleuwen, B. E., Engelberts, A. C., L’hoir, M. P., et al. (2011). Postnatal parental smoking: An important risk factor for SIDS. European Journal of Pediatrics, 170(10), 1281–1291. doi:10.1007/s00431-011-1433-6. Mitchell, E. (1993). Sleeping position of infants and the sudden infant death syndrome. Acta Paediatrica, 82(s390), 26–30. Retrieved from http://onlinelibrary.wiley.com/journal/ 10.1111/%28ISSN%291651-2227 Mitchell, E., Freemantle, J., Young, J., & Byard, R. W. (2012). Scientific consensus forum to review the evidence underpinning the recommendations of the Australian SIDS and kids safe sleeping health promotion program—October 2010. Journal of Paediatrics and Child Health, 48, 626–633. doi:10.1111/j.1440-1754.2011.02215.x. Mitchell, E. A. (2009). SIDS: past, present and future. Acta Paediatrica, 98(11), 1712–1719. doi: 10.1111/j.1651-2227.2009.01503.x Mitchell, E. A., Thach, B. T., Thompson, J. D., & Williams, S. (1999). Changing infants’ sleep position increases risk of sudden infant death syndrome. Archives of Pediatrics & Adolescent Medicine, 153(11), 1136–1141. doi:10.1001/archpedi.153.11.1136. Moon, R. Y., Horne, R. S., & Hauck, F. R. (2007). Sudden infant death syndrome. The Lancet, 370(9598), 1578–1587. doi:10.1016/S0140-6736(07)61662-6. Nie, C., Bailey, A., Istre, G., & Anderson, R. (2010). Populationbased survey of infant bed sharing. Injury Prevention, 16 (S1), A69. doi:10.1136/ip.2010.029215.252. Olsen, J., Christensen, K., Murray, J., & Ekbom, A. (2010). An introduction to epidemiology for health professionals. New York: Springer. Paneth, N. (2004). Assessing the contributions of John Snow to epidemiology: 150 years after removal of the Broad Street pump handle. Epidemiology, 15(5), 514–516. doi:10.2307/ 20485943. Parfrey, P. S., & Barrett, B. (2009). Clinical epidemiology: Practice and methods (Vol. 473). London: Humana Press. Pearce, N., & Greenland, S. (2014). Confounding and interaction. In W. Ahrens, & I. Pigeot (Eds.), Handbook of epidemiology (2nd ed., pp. 659–684). New York: Springer Science and Buisness Media. Polit, D. F., & Beck, C. T. (2012). Nursing research. Generating and assessing evidence for nursing practice (9th ed.). Philadephia, PA: Lippincott Williams & Williams. Porta, M. S. (2014). Dictionary of epidemiology (6th ed.). New York, NY, USA: Oxford University Press. Retrieved from http:// www.ebrary.com Rognum, T., & Saugstad, O. (1993). Biochemical and immunological studies in SIDS victims. Clues to understanding the death mechanism. Acta Paediatrica, 82(s390), 82–85. doi:10.1111/j.1651-2227.1993.tb12886.x. Task Force on Sudden Infant Death Syndrome (2011). SIDS and other sleep-related infant deaths: Expansion of recommendations for a safe infant sleeping environment. Pediatrics, 128(5), 1030–1039. doi:10.1542/peds.2011-228. Vennemann, M. M., Bajanowski, T., Brinkmann, B., Jorch, G., Sauerland, C., Mitchell, E. A., et al. (2009). Sleep environment risk factors for sudden infant death syndrome: The German Sudden Infant Death Syndrome Study. Pediatrics, 123(4), 1162–1170. doi:10.1542/peds.2008-0505. Vennemann, M. M., Hense, H., Bajanowski, T., Blair, P. S., Complojer, C., Moon, R. Y., et al. (2012). Bed sharing and the risk of sudden infant death syndrome: Can we resolve the debate? The Journal of Pediatrics, 160(1), 44–48.e2. doi:10.1016/j.jpeds.2011.06.052.
SPECIAL FEATURES: METHODS
Bed Sharing, SIDS Research, and the Concept of Confounding: A Review for Public Health Nurses Elizabeth M. Keys, PhD Student, BN, RN and James A. Rankin, PhD, NP Faculty of Nursing, University of Calgary, Calgary, Alberta, Canada Correspondence to: Elizabeth Keys, PhD Student, Faculty of Nursing, Professional Faculties Building, 2500 University Drive NW, Calgary, Alberta, Canada T2N 4N1. E-mail: [email protected]
ABSTRACT Confounding is an important concept for public health nurses (PHNs) to understand when considering the results of epidemiological research. The term confounding is derived from Latin, confundere, which means to mix-up or mix together. Epidemiologists attempt to derive a cause and effect relationship between two variables traditionally known as the exposure and disease (e.g., smoking and lung cancer). Confounding occurs when a third factor, known as a confounder, leads to an over- or underestimate of the magnitude of the association between the exposure and disease. An understanding of confounding will facilitate critical appraisal of epidemiological research findings. This knowledge will enable PHNs to strengthen their evidence-based practice and better prepare them for policy development and implementation. In recent years, researchers and clinicians have examined the relationship between bed sharing and sudden infant death syndrome (SIDS). The discussion regarding the risk of bed sharing and SIDS provides ample opportunity to discuss the various aspects of confounding. The purpose of this article is to use the bed sharing and SIDS literature to assist PHNs to understand confounding and to apply this knowledge when appraising epidemiological research. In addition, strategies that are used to control confounding are discussed. Key words: bed sharing, confounding, epidemiological research, evidence-based, nursing practice, public health nursing, sudden infant death.
A major focus for public health nurses (PHNs) is on risk reduction and primary prevention (Hemingway, Aarts, Koskinen, Campbell, & Chasse, 2013). The factors associated with sudden infant death syndrome (SIDS) represent significant health risks to infants. In 2004, an international panel of SIDS experts refined the definition of SIDS as: The sudden unexpected death of an infant less than 1 year of age, with onset of the fatal episode apparently occurring during sleep, that remains unexplained after a thorough investigation, including performance of a complete autopsy and review of the circumstances of death and the clinical history. (Krous et al., 2004, p. 235)
The Triple Risk Hypothesis (Guntheroth & Spiers, 2002; Rognum & Saugstad, 1993) is the cur-
rent guiding framework for SIDS etiology. While a complete explanation of this etiological framework is beyond the scope of this article, in brief, the Triple Risk Hypothesis posits that infants succumb to SIDS when three factors intersect: (a) the infant has inherent risk (nonmodifiable risk factors such as genetic or brainstem abnormalities); (b) the infant is within a vulnerable developmental and age-related period; and (c) modifiable risk factors are present (such as parental smoking or soft bedding; Guntheroth & Spiers, 2002; Rognum & Saugstad, 1993). Early epidemiologic findings indicated the infant’s sleeping position as a significant modifiable risk factor for SIDS (Mitchell, 1993). Using the case-control research method investigators around the world have consistently demonstrated that
731
732
Public Health Nursing
Volume 32
Number 6
infants who are placed in a prone position for sleep are 1.3 to 7.08 times more likely to succumb to SIDS than those placed in a supine position (Blair et al., 2009; Colvin, Collie-Akers, Schunn, & Moon, 2014; Mitchell, Thach, Thompson, & Williams, 1999; Vennemann et al., 2009). In the United States, “Back to Sleep” campaigns have reduced the rate of SIDS by at least 50% (American Academy of Pediatrics Task Force on Sudden Infant Death Syndrome, 2005). Since this dramatic decrease, researchers have focused on identifying additional modifiable risk factors to facilitate further reduction of SIDS. Some of these factors include parental smoking, drug and alcohol use, the infant’s sleep environment and location within the home (Mitchell, 2009; Moon, Horne, & Hauck, 2007). Current infant sleep guidelines recommend that parents avoid the practice of bed sharing with their infant due to an increased SIDS risk (Mitchell, Freemantle, Young, & Byard, 2012; Task Force on Sudden Infant Death Syndrome, 2011). Despite these recommendations, the practice of bed sharing is common (Ateah & Hamelin, 2008; Nie, Bailey, Istre, & Anderson, 2010). Consequently, researchers have examined the relationship between bed sharing and SIDS in more detail. PHNs play a key role in assisting parents to identify and act on modifiable SIDS risk factors. The attention to risk reduction can sometimes appear to conflict with other elements of family wellness and autonomy. Specifically, PHNs may encounter tension between reducing SIDS risk and supporting breastfeeding mothers who often practice bed sharing (Fetherston & Leach, 2012). Understanding and appraising the epidemiological evidence that informs current safe sleep recommendations will assist in navigating these tensions. One of the most important factors in interpreting epidemiological research is understanding the concept of confounding (Gordis, 2014). Confounding is defined as “the distortion of a measure of effect of an exposure on an outcome due to the association of the exposure with other factors that influence the occurrence of the outcome” (Porta, 2014, p. 55). In other words, given a particular exposure, confounding may bias the magnitude of the effect of an association on an outcome. Confounding can cause: (a) an overestimate of an effect which leads researchers to conclude there is a difference when there is none; or (b) an underestimate of an effect which leads
November/December 2015
researchers to conclude there is no difference when there actually is one (Elwood, 2007; Pearce & Greenland, 2014). The purpose of this article is to explain confounding and facilitate PHNs’ application of this knowledge when appraising epidemiological research. Critical appraisal of epidemiological studies will help PHNs assess the applicability of research findings to their own practice populations. This ability will support an evidence-based practice and further PHNs’ capacities for assuming roles in policy development and implementation. The authors use examples from the bed sharing and SIDS research literature to explain confounding. Strategies used to control confounding are explained and the implications of understanding confounding for nursing practice are summarized.
What is Confounding? In 1854, Dr. John Snow demonstrated a classic epidemiological example of exposure and disease outcome in Soho, London. Snow, one of the founders of modern epidemiology, identified the causal relationship between a contaminated water pump in Broad Street (the exposure) and the outbreak of cholera (the outcome; Paneth, 2004). The causal relationship between contaminated water and disease is relatively straightforward in contrast to the many predisposing factors that are cited to be associated with the etiology of “modern” diseases. In SIDS, several predisposing factors have been identified (Task Force on Sudden Infant Death Syndrome, 2011). Present-day epidemiological researchers examine a wide range of biological, social, and economic variables and their association with disease (Brownson & Hoehner, 2006). In its simplest form, confounding arises when a third variable is thought to influence the result of an observed association between an exposure and an outcome (Elwood, 2007; Pearce & Greenland, 2014). The third variable is commonly referred to as a potential confounder until it has been demonstrated that it influences the result of the observed association; at which time it becomes an actual confounder. In order for a variable to be considered a confounder the following criteria must be met: (a) the variable must be associated with the exposure; (b) independent of the association with the exposure, the variable must be associated with the
Keys and Rankin: Confounding Bed Sharing SIDS outcome or disease; and (c) the confounder must not be an intermediate step in the causal pathway between exposure and disease (Olsen, Christensen, Murray, & Ekbom, 2010). Risk factors for SIDS represent different types of exposure, with SIDS as the outcome. Figure 1 shows how confounding occurs when a putative causal relationship between an exposure and an outcome is influenced by a third variable: the confounder (Elwood, 2007). It is important to emphasize that the confounder is not part of the proposed causal pathway (Olsen et al., 2010). Rather, the confounder is independently associated with the exposure and the outcome. Consider the association between infants who bed share and the occurrence of SIDS (Vennemann et al., 2012); the relationship may be influenced by a potential confounder (Figure 2). Maternal age is considered a potential confounder as mothers under the age of 20 years are 1.55 (95% confidence interval [CI] = 1.08–2.24) times more likely to bed share (Broussard, Sappenfield, & Goodman, 2012). Researchers have subsequently demonstrated that mothers less than 18 years increases the risk of SIDS nine-fold (95% CI = 5.9–14.1; Carpenter et al., 2013). Maternal age is not part of the proposed causal path and it has been shown to modify the effect of bed sharing on SIDS; it is therefore no longer considered a potential confounder, rather it is now considered to be a confounder. When the confounder is positively associated with the outcome and the exposure, as in this
Figure 1. The Effect of Confounding on the Association between Exposure and Outcome
733
Figure 2. Maternal Age as a Confounder in the Bed Sharing-SIDS Relationship younger maternal age example, the magnitude of the exposure-outcome relationship is overestimated (Elwood, 2007). In this example, the true risk of SIDS when bed sharing may be artificially amplified by positive associations with younger maternal age. Depending on the strength of these associations, the relationship between bed sharing and SIDS could be partially explained by maternal age. Given that confounding can influence the actual risk of negative health outcomes, it is important for PHNs to consider situations where potential confounders may artificially inflate putative causal relationships. A confounding variable may reduce the magnitude of the association between the exposure and the outcome (Elwood, 2007). As a result, the confounder can reduce the exposure-outcome association with the extent that it may not even be detected (Elwood, 2007). For instance, breastfeeding has been shown to have a protective effect by reducing the risk of SIDS by 45% (Hauck, Thompson, Tanabe, Moon, & Vennemann, 2011). However, mothers who breastfeed are 2.3 (95% CI = 1.87–2.89) times more likely to bed share (Blair, Heron, & Fleming, 2010). In this situation, the magnitude of the protective effect of breastfeeding on SIDS may be sufficiently large to reduce the association between bed sharing and SIDS. Nevertheless, when the confounding (in this case, protective effect) of breastfeeding is controlled in the analysis, there remains a 5.1 (95% CI = 2.3– 11.4) increased risk of SIDS for those that bed share (Carpenter et al., 2013). It is important to be aware that the terms confounding and bias are on occasion used interchangeably (Olsen et al., 2010). A variable can only be truly confounding when it has an independent relationship with the exposure and the outcome (Law, Green, & Ellison, 2012). In other words, if a potentially confounding variable is only associated with either the exposure or outcome but not both, it is not a confounder.
734
Public Health Nursing
Volume 32
Number 6
For instance, maternal smoking is associated with a 2.7 (95% CI = 1.0–6.4) times increase in SIDS risk (Liebrechts-Akkerman et al., 2011). However, no relationship between maternal smoking and bed sharing has been reported. Hence, maternal smoking is not considered a true confounder in the bed sharing-SIDS relationship. Instead, maternal smoking is an effect modifier. An effect modifier changes the risk of a particular outcome (SIDS), but is not associated with the exposure of interest (bed sharing; Olsen et al., 2010). Confusing an effect modifier for a confounding variable can interfere in the development of accurate causal models (Olsen et al., 2010).
Identifying Potential Confounders PHNs should reflect on their practical knowledge of their specific practice populations to assess whether researchers address potential confounders. This appraisal assists PHNs to determine the generalizability of the research findings for the communities with whom they work. A preliminary step in identifying potential confounders is to examine the literature for associations that have been reported for both the exposure and outcome (Polit & Beck, 2012). For example, a PHN may notice that many families who bed share belong to a particular ethnic group. Before deciding that ethnicity is a confounding variable, an association between an ethnic group and SIDS would need to be identified. In fact, it has been demonstrated that ethnicity is a confounding variable in the association between bed sharing and SIDS. Findings from several studies indicate that families belonging to nonCaucasian races often have increased rates of bed sharing. For example, Ball et al. (2012) found that Pakistani-British mothers were 3.02 (95% CI = 1.96–4.66) times more likely to regularly bed share than Caucasian British mothers, while Colson et al. (2013) found that African-American and Hispanic-American mothers were 3.47 (95% CI = 2.97–4.05) and 1.33 (95% CI = 1.10–1.61) times, respectively, more likely to bed share compared to Caucasian American mothers. Meanwhile, Hogan (2014) has reported a statistically significant (p = .001) increased proportion of sleep-related infant deaths in infants of African-American mothers (62.5%), compared to those of Caucasian mothers (25.7%).
November/December 2015
Controlling Confounding There are several ways that researchers can control for confounding either by design or in the statistical analysis. Randomization, which refers to the random allocation of subjects in a study, into different groups is a powerful way to reduce the risk of confounding. Observational epidemiological studies are more vulnerable to the threat of confounding as investigators are unable to randomize retrospectively or cannot randomize due to ethical constraints (Gordis, 2014; Pearce & Greenland, 2014). For obvious ethical reasons researchers would not be permitted to randomly allocate infants to different groups to study SIDS; similarly this is why there has never been a randomized controlled trial (RCT) to examine the relationship between smoking and lung cancer in humans. As may be seen from Figure 3, the strategies used to control for confounding fall into two categories; researchers can control at the design stage or at the analysis stage of the study. Of course, researchers can only control for confounding at the analysis stage providing they have the foresight to collect data on potential confounders during the study (Olsen et al., 2010; Pearce & Greenland, 2014). For example, SIDS researchers are now more aware of collecting data on potential confounders such as socioeconomic status, ethnicity, and type of infant feeding, compared with earlier research (Carpenter et al., 2013; Dukes et al., 2014; Vennemann et al., 2012). Each of the strategies for controlling confounding (see Figure 3) is now briefly described.
Randomization As previously stated, randomization refers to the random allocation of subjects into groups. For example, in the context of a RCT researchers randomly allocate subjects to a treatment or control group. Theoretically, potential confounders have an equal chance of being evenly distributed between the groups (Grimes & Schulz, 2002; Pearce & Greenland, 2014). In other words, randomization into groups is a very powerful way to minimize the influence that confounding variables have on the study results. It is important to emphasize that randomization does not guarantee that confounding will never occur; rather the strategy effectively minimizes the likelihood that the strength of the association between exposure and outcome has not been influenced by confounding variables.
Keys and Rankin: Confounding Bed Sharing SIDS
735
Figure 3. Strategies for Controlling Confounding by Study Design and Data Analysis Randomization is the only strategy that is able to minimize the impact of unknown confounders (Grimes & Schulz, 2002). However, true randomization may not always be accomplished (Pearce & Greenland, 2014). Consequently, even investigators conducting RCTs use additional strategies to control confounding. These strategies are also used in observational studies where confounding is much more of a concern (Pearce & Greenland, 2014).
Matching Matching subjects on potential confounding variables, such as age and sex, occurs during the design phase of a study (Elwood, 2007; Pearce & Greenland, 2014). Maternal age, education level, ethnicity, socioeconomic status, and type of infant feeding are now known to be confounders in the bed sharing and SIDS relationship (Carpenter et al., 2013; Hogan, 2014; Vennemann et al., 2012). It is therefore reasonable to match participants based on these variables; however, matching for more than two or three variables can make it challenging to find appropriate controls (Pearce & Greenland, 2014). Thus, matching in the bed sharing and SIDS research is often limited to matching by age and geographic region (see Carpenter et al., 2004 for example). Furthermore, the process of matching cases for variables that are not true confounders transforms them into artificial confounders (Gordis, 2014; Pearce & Greenland, 2014). This transformation must then be controlled during the data analysis (Elwood, 2007; Pearce & Greenland, 2014). For example, although male infants have a slightly increased risk of SIDS compared to female infants (Moon et al., 2007), there is likely no relationship between the sex of the infant and bed sharing. In
this case, matching subjects based on infant sex would not be helpful in reducing confounding and, furthermore, may complicate the data analysis.
Restriction Restriction is used to control confounding by excluding subjects with potential confounding variables from the study or by excluding them in the analysis (Elwood, 2007). For example, in a casecontrol study of bed sharing and SIDS, Fu, Moon, and Hauck (2010) controlled for confounding due to African-American ethnicity by restricting the study to African-Americans. Restriction is an effective way to control for the effect of confounding, however by excluding subjects certain confounders, such as breastfeeding or low socioeconomic status, severely limits the generalizability of the results (Krickeberg, Pham, & Pham, 2012). Moreover, investigators who apply very restrictive inclusion criteria may have difficulty in recruiting a sufficiently large sample size for the study (Brownson & Hoehner, 2006). Stratification Stratification involves subdividing the subjects into smaller groups or “strata” based on different values of the potential confounder (Elwood, 2007; Parfrey & Barrett, 2009). The analysis of the association between the exposure and the outcome is then computed for each of the smaller groups; the researcher thereby obtains stratum specific results (Elwood, 2007). The ultimate sample size for each of the strata depends on the number of subdivisions made on the potential confounder (Pearce & Greenland, 2014). For example, for the potential confounder of sex there is usually only two subdivisions; however, con-
736
Public Health Nursing
Volume 32
Number 6
founding variables such as age or birth order can be divided into many more categories. In their metaanalysis, Carpenter et al. (2013) stratified the data to assess for confounding by infant feeding, maternal age, parity, ethnicity, and marital status. This technique can be limited by the precision with which the potential confounder is measured (Pearce & Greenland, 2014). In the Carpenter study, this limitation is exemplified in that breastfeeding (yes or no) and ethnicity (Caucasian or non-Caucasian) were measured as binary variables. On the other hand, stratifying a potential confounder into too many categories can result in some strata with no subjects (Olsen et al., 2010). For example, this may occur when maternal age is analyzed using the precise age in years (18, 19, 20 and so on) rather than combining subjects into age groups (under 18 years, 19–20 years, 21–25 years). From a statistical perspective, a more sophisticated approach to deal with confounding is to use stratification and multivariate analysis techniques in combination with one another (Pearce & Greenland, 2014).
Multivariate analysis Public health nurses may be aware of one type of multivariate analysis known as logistic regression; SIDS investigators conducting case-control studies control for confounding by using logistic regression analysis. Logistic regression involves complex statistical modeling that estimates the likelihood of an outcome given a particular exposure and controls for several confounding variables simultaneously (Elwood, 2007). For example, Carpenter et al. (2013) used logistic regression to calculate the risk of SIDS (the outcome) based on bed sharing (the exposure), and simultaneously controlled for variables such as parental smoking, mother’s age, infant feeding method, infant sex, birth weight, and parity. In case-control studies, the magnitude of the risk between an exposure and an outcome is expressed in terms of a statistic known as the odds ratio (OR). The cases have the disease (or outcome of interest) and the controls do not, both groups are then compared on the exposure. If a particular exposure increases the likelihood of an outcome, the OR will be greater than 1.0 (Laing & Rankin, 2011). For example, the OR of SIDS given maternal illicit drug use after birth is 11.5 (Carpenter et al., 2013). This means that, for the cases in the study, infants whose mothers used illicit drugs after birth
November/December 2015
(the exposure) were 11.5 times more likely to succumb to SIDS (the outcome). If a particular exposure reduces the likelihood of an outcome, the OR is less than 1.0 (Laing & Rankin, 2011). For instance, Hauck et al. (2011) found that infants in their study who received any amount of breast milk had an OR of 0.40 for SIDS. This means that these infants were 60% less likely to succumb to SIDS. An OR that is equal to 1.0 is the null value and indicates that there is no difference between the cases and the controls; the risk is the same in both groups (Laing & Rankin, 2011). An understanding of confounding facilitates the PHN’s ability to interpret the epidemiological research literature and to assess if confounding is clinically relevant. PHNs can become more effective practitioners when they develop their capacity to appraise the epidemiological evidence that informs their practice. In their day-to-day practice, PHNs are already adept at considering the interplay of multiple factors with regard to particular health outcomes. Understanding confounding extends this knowledge repertoire and builds capacity to examine the health risks of the communities that they serve. Furthermore, PHNs can use their understanding of confounding, in combination with their observations in practice, to critically reflect on the health risks of their populations. As a result, PHNs can be confident that they are focusing on reducing exposures and health risks that are truly responsible for negative health outcomes.
References American Academy of Pediatrics Task Force on Sudden Infant Death Syndrome (2005). The changing concept of sudden infant death syndrome: Diagnostic coding shifts, controversies regarding the sleeping environment, and new variables to consider in reducing risk. Pediatrics, 116(5), 1245–1255. doi:10.1542/peds.2005-1499. Ateah, C. A., & Hamelin, K. J. (2008). Maternal bedsharing practices, experiences, and awareness of risks. Journal of Obstetric, Gynecologic, & Neonatal Nursing, 37(3), 274– 281. doi:10.1111/j.1552-6909.2008.00242.x. Ball, H. L., Moya, E., Fairley, L., Westman, J., Oddie, S., & Wright, J. (2012). Bed- and sofa-sharing practices in a UK bioethnic population. Pediatrics, 129(3), e673–e681. Retrieved from http://pediatrics.aappublications.org/ Blair, P. S., Heron, J., & Fleming, P. J. (2010). Relationship between bed sharing and breastfeeding: Longitudinal, population-based analysis. Pediatrics, 126(5), e1119–e1126. doi:10.1542/peds.2010-1277. Blair, P. S., Sidebotham, P., Evason-Coombe, C., Edmonds, M., Heckstall-Smith, E. M. A., & Fleming, P. (2009). Hazardous cosleeping environments and risk factors amenable to change: Case-control study of SIDS in south west England. British Medical Journal, 339(7726), 1–11. doi:10.1136/ bmj.b3666.
Keys and Rankin: Confounding Bed Sharing SIDS Broussard, D. L., Sappenfield, W. M., & Goodman, D. A. (2012). The black and white of infant back sleeping and infant bed sharing in Florida, 2004-2005. Maternal and Child Health Journal, 16(3), 713–724. doi:10.1007/s10995-011-0768-y. Brownson, R. C., & Hoehner, C. M. (2006). Epidemiology. A foundation of public health. In R. C. Brownson, & D. B. Petitti (Eds.), Applied epidemiology: Theory to practice (pp. 1–28). Oxford: Oxford University Press. Carpenter, R., Irgens, L., Blair, P., England, P., Fleming, P., Huber, J., et al. (2004). Sudden unexplained infant death in 20 regions in Europe: Case control study. The Lancet, 363 (9404), 185–191. Retrieved from http://www.thelancet.com/ Carpenter, R., McGarvey, C., Mitchell, E. A., Tappin, D. M., Vennemann, M. M., Smuk, M., et al. (2013). Bed sharing when parents do not smoke: Is there a risk of SIDS? An individual level analysis of five major case-control studies. BMJ Open, 3(5), 1–11. Retrieved from http://bmjopen.bmj.com/ Colson, E. R., Willinger, M., Rybin, D., Heeren, T., Smith, L. A., Lister, G., et al. (2013). Trends and factors associated with infant bed sharing, 1993-2010: The national infant sleep position study. JAMA Pediatrics, 167(11), 1032–1037. doi:10.1001/jamapediatrics.2013.2560. Colvin, J. D., Collie-Akers, V., Schunn, C., & Moon, R. Y. (2014). Sleep environment risks for younger and older infants. Pediatrics, 134(2), e406–e412. doi:10.1542/peds.2014-0401. Dukes, K. A., Odendaal, H. J., Signore, C., Sullivan, L. M., Willinger, M., Wright, C., & PASS Research Network (2014). The safe passage study: Design, methods, recruitment, and follow-up approach. Paediatric and Perinatal Epidemiology, 28(5), 455–465. doi:10.1111/ppe.12136. Elwood, J. M. (2007). Critical appraisal of epidemiological studies and clinical trials. Oxford: Oxford University Press. Fetherston, C. M., & Leach, J. S. (2012). Analysis of the ethical issues in the breastfeeding and bedsharing debate. Breastfeeding Review, 20(3), 7–17. Retrieved from https:// www.breastfeeding.asn.au/bfreview Fu, L. Y., Moon, R. Y., & Hauck, F. R. (2010). Bed sharing among black infants and sudden infant death syndrome: Interactions with other known risk factors. Academic Pediatrics, 10(6), 376–382. doi:10.1016/j.acap.2010.09.001. Gordis, L. (2014). Epidemiology (5th ed.). Philadelphia, PA: Elsevier/Saunders. Grimes, D. A., & Schulz, K. F. (2002). An overview of clinical research: The lay of the land. The Lancet, 359(9300), 57– 61. doi:10.1016/S0140-6736(02)07283-5. Guntheroth, W. G., & Spiers, P. S. (2002). The triple risk hypotheses in sudden infant death syndrome. Pediatrics, 110(5), e64–e69. doi:10.1542/peds.110.5.e64. Hauck, F. R., Thompson, J. M., Tanabe, K. O., Moon, R. Y., & Vennemann, M. M. (2011). Breastfeeding and reduced risk of sudden infant death syndrome: A meta-analysis. Pediatrics, 128(1), 103–110. doi:10.1542/peds.2010-3000. Hemingway, A., Aarts, C., Koskinen, L., Campbell, B., & Chasse, F. (2013). A European Union and Canadian review of public health nursing preparation and practice. Public Health Nursing, 30(1), 58–69. doi:10.1111/j.1525-1446.2012.01048.x. Hogan, C. (2014). Socioeconomic factors affecting infant sleeprelated deaths in St. Louis. Public Health Nursing, 31(1), 10–18. doi:10.1111/phn.12052. Krickeberg, K., Pham, V. T., & Pham, T. M. H. (2012). Epidemiol˙ prevention. New ˙ York, NY: Springer New York. ogy: Key to doi:10.1007/978-1-4614-1205-2 Krous, H. F., Beckwith, J. B., Byard, R. W., Rognum, T. O., Bajanowski, T., Corey, T., et al. (2004). Sudden infant death syndrome and unclassified sudden infant deaths: A definitional and diagnostic approach. Pediatrics, 114(1), 234–238. Retrieved from http://pediatrics.aappublications.org/ Laing, C. M., & Rankin, J. A. (2011). Odds ratios and confidence intervals: A review for the pediatric oncology clinician. Journal of Pediatric Oncology Nursing, 28(6), 363–367. doi:10.1177/1043454211426575.
737
Law, G. R., Green, R., & Ellison, T. H. (2012). Confounding and causal path diagrams. In Y. Tu, & D. C. Greenwood (Eds.), Modern methods for epidemiology (pp. 1–13). Dordrecht, Netherlands: Springer Science & Business Media. doi: 10.1007/978-94-007-3024-3 Liebrechts-Akkerman, G., Lao, O., Liu, F., van Sleuwen, B. E., Engelberts, A. C., L’hoir, M. P., et al. (2011). Postnatal parental smoking: An important risk factor for SIDS. European Journal of Pediatrics, 170(10), 1281–1291. doi:10.1007/s00431-011-1433-6. Mitchell, E. (1993). Sleeping position of infants and the sudden infant death syndrome. Acta Paediatrica, 82(s390), 26–30. Retrieved from http://onlinelibrary.wiley.com/journal/ 10.1111/%28ISSN%291651-2227 Mitchell, E., Freemantle, J., Young, J., & Byard, R. W. (2012). Scientific consensus forum to review the evidence underpinning the recommendations of the Australian SIDS and kids safe sleeping health promotion program—October 2010. Journal of Paediatrics and Child Health, 48, 626–633. doi:10.1111/j.1440-1754.2011.02215.x. Mitchell, E. A. (2009). SIDS: past, present and future. Acta Paediatrica, 98(11), 1712–1719. doi: 10.1111/j.1651-2227.2009.01503.x Mitchell, E. A., Thach, B. T., Thompson, J. D., & Williams, S. (1999). Changing infants’ sleep position increases risk of sudden infant death syndrome. Archives of Pediatrics & Adolescent Medicine, 153(11), 1136–1141. doi:10.1001/archpedi.153.11.1136. Moon, R. Y., Horne, R. S., & Hauck, F. R. (2007). Sudden infant death syndrome. The Lancet, 370(9598), 1578–1587. doi:10.1016/S0140-6736(07)61662-6. Nie, C., Bailey, A., Istre, G., & Anderson, R. (2010). Populationbased survey of infant bed sharing. Injury Prevention, 16 (S1), A69. doi:10.1136/ip.2010.029215.252. Olsen, J., Christensen, K., Murray, J., & Ekbom, A. (2010). An introduction to epidemiology for health professionals. New York: Springer. Paneth, N. (2004). Assessing the contributions of John Snow to epidemiology: 150 years after removal of the Broad Street pump handle. Epidemiology, 15(5), 514–516. doi:10.2307/ 20485943. Parfrey, P. S., & Barrett, B. (2009). Clinical epidemiology: Practice and methods (Vol. 473). London: Humana Press. Pearce, N., & Greenland, S. (2014). Confounding and interaction. In W. Ahrens, & I. Pigeot (Eds.), Handbook of epidemiology (2nd ed., pp. 659–684). New York: Springer Science and Buisness Media. Polit, D. F., & Beck, C. T. (2012). Nursing research. Generating and assessing evidence for nursing practice (9th ed.). Philadephia, PA: Lippincott Williams & Williams. Porta, M. S. (2014). Dictionary of epidemiology (6th ed.). New York, NY, USA: Oxford University Press. Retrieved from http:// www.ebrary.com Rognum, T., & Saugstad, O. (1993). Biochemical and immunological studies in SIDS victims. Clues to understanding the death mechanism. Acta Paediatrica, 82(s390), 82–85. doi:10.1111/j.1651-2227.1993.tb12886.x. Task Force on Sudden Infant Death Syndrome (2011). SIDS and other sleep-related infant deaths: Expansion of recommendations for a safe infant sleeping environment. Pediatrics, 128(5), 1030–1039. doi:10.1542/peds.2011-228. Vennemann, M. M., Bajanowski, T., Brinkmann, B., Jorch, G., Sauerland, C., Mitchell, E. A., et al. (2009). Sleep environment risk factors for sudden infant death syndrome: The German Sudden Infant Death Syndrome Study. Pediatrics, 123(4), 1162–1170. doi:10.1542/peds.2008-0505. Vennemann, M. M., Hense, H., Bajanowski, T., Blair, P. S., Complojer, C., Moon, R. Y., et al. (2012). Bed sharing and the risk of sudden infant death syndrome: Can we resolve the debate? The Journal of Pediatrics, 160(1), 44–48.e2. doi:10.1016/j.jpeds.2011.06.052.
