Ambient Heat and Sudden Infant Death
Ambient Heat and Sudden Infant Death
We undertook a case-crossover analysis of all SIDS cases before 1 year of age in metropolitan Montreal, Quebec, Canada, from April through October for the years 1981–2010, the 30-year period available to us. Montreal has a continental climate with hot summers and cold winters. Deaths during November through March were not considered because high temperatures were not encountered. In addition, SIDS is common during winter (Mage 2005), and elevated thermal stress from excessive clothing in cold weather may inadvertently bias or mask associations with high outdoor temperatures (Ponsonby et al. 1992b). To increase statistical power, we did not exclude bridge months that reached relatively high maximum daily temperatures, including April (29.4°C) and October (26.6°C).
SIDS cases were identified in vital statistics records of the Quebec health ministry using International Classification of Diseases, 9th Revision (ICD-9) and 10th Revision (ICD-10) codes for principal cause of death (798.0, R95.0). There were 196 cases of SIDS during the time span covered, and 736,916 live births. For comparison, there were 3,869 infant deaths overall during the same period, including 1,009 deaths after 27 days of age.
We hypothesized that high heat exposures could lead to SIDS, and therefore used the maximum outdoor temperature recorded on the same day and day before a SIDS day or control day as two main exposures. We used both days because the exact time of death was unknown, and we could not capture the exact temperature at the time of the event. Maximum temperatures the preceding day are certain to have occurred before death, but temperatures on the same day may have been reached only after death for a proportion of cases. To explore delayed impacts, we used the maximum temperature 2 days before a SIDS day or control day as a secondary exposure. Hourly temperature data were provided by Environment Canada by 24-hr block for the meteorological center situated approximately 20 km from Montreal's core (Smargiassi et al. 2009). Maximum daily temperature was modeled as a continuous variable, and odds ratios were estimated relative to a referent value of 20°C. This referent was selected because 20°C is a comfortable temperature generally not associated with thermal stress in Montreal, and has been used in previous research (Auger et al. 2014; Smargiassi et al. 2009).
Humidity was considered a potential confounder of the association between temperature and SIDS. Humidity data were collected at the same meteorological center as temperature and were measured using mean daily percent relative humidity (continuous). We did not adjust for air pollution, a potential causal intermediate in the pathway between temperature and mortality, because this could bias estimates of the total impact of temperature on SIDS (Buckley et al. 2014).
We computed descriptive statistics and compared the proportion of SIDS deaths that occurred after days with maximum temperature ≥ 28°C with the proportion on the same days for all other causes.
We used the case-crossover design for its strength in assessing associations between transient exposures such as temperature and acute outcomes that are rare, such as SIDS (Maclure and Mittleman 2000). Case-crossover analysis is increasingly used to estimate impacts of temperature on mortality (Basagaña et al. 2011; Basu and Ostro 2008). In this design, each SIDS case serves as its own control, and the statistical analysis consists of comparing temperature exposure at the time of the event to temperatures during a short interval around the time of the case. Because each case is its own control, case-crossover designs inherently adjust for potential individual confounders that vary little over time, such as socioeconomic status, smoking, co-morbidities, and year of birth. In addition, this design accounts for potential bias from seasonal variation in conception and birth (Basso et al. 1995). Case-crossover studies are not ecologic, but use individual rather than aggregate data as the unit of analysis (Künzli and Tager 1997).
To select control days, we used an ambidirectional time-stratified approach where the referent period was the calendar month; we matched days on which SIDS occurred to control days consisting of the same weekdays of the month of death (Levy et al. 2001). If a death, for example, occurred on a Saturday in July 2000, control days comprised all remaining Saturdays in that month. Thus, any potential confounders that were stable during the month, such as socioeconomic status, were automatically adjusted for despite lack of data on such characteristics. We selected the same weekday as controls, thereby automatically adjusting for mortality that might vary by day of week. Because SIDS is rare, bias due to the use of control days that occur during the same month, but after the day of death, will be negligible (Lumley and Levy 2000).
We used conditional logistic regression to calculate odds ratios and 95% confidence intervals (CIs) for the association between maximum temperature of event days relative to temperatures of control days (each temperature variable was modeled separately), and included spline terms with knots at the 10th, 50th, and 90th percentiles (Durrleman and Simon 1989). We verified that use of knots located at the 5th, 50th, and 95th percentiles, and a greater quantity of knots, did not affect the shape of the curves (data not shown). All models were adjusted for relative humidity.
To estimate age-specific effects of temperature, we stratified SIDS cases into two postneonatal periods (1–2 vs. 3–12 months) for separate analyses, based on research suggesting increased susceptibility of infants older than 2 months to thermal stress (Fleming et al. 1992). In addition we performed a secondary analysis of associations at 3–6 months of age, but we did not have sufficient numbers of cases to estimate associations separately for SIDS deaths before 1 month of age (n = 13) or for 7–12 months of age (n = 14).
In sensitivity analyses, we ran models that excluded humidity, and examined associations for data restricted to summer months only (defined as June–August). We used the restricted cubic spline (RCS) macro in SAS version 9.2 (SAS Institute Inc., Cary, NC, USA) for statistical analyses (Heinzl and Kaider 1997). Data were anonymized, and the institutional review board of the University of Montreal Hospital Centre waived the requirement for formal ethics review.
Methods
Study Design and Population
We undertook a case-crossover analysis of all SIDS cases before 1 year of age in metropolitan Montreal, Quebec, Canada, from April through October for the years 1981–2010, the 30-year period available to us. Montreal has a continental climate with hot summers and cold winters. Deaths during November through March were not considered because high temperatures were not encountered. In addition, SIDS is common during winter (Mage 2005), and elevated thermal stress from excessive clothing in cold weather may inadvertently bias or mask associations with high outdoor temperatures (Ponsonby et al. 1992b). To increase statistical power, we did not exclude bridge months that reached relatively high maximum daily temperatures, including April (29.4°C) and October (26.6°C).
SIDS cases were identified in vital statistics records of the Quebec health ministry using International Classification of Diseases, 9th Revision (ICD-9) and 10th Revision (ICD-10) codes for principal cause of death (798.0, R95.0). There were 196 cases of SIDS during the time span covered, and 736,916 live births. For comparison, there were 3,869 infant deaths overall during the same period, including 1,009 deaths after 27 days of age.
Measures of Exposure
We hypothesized that high heat exposures could lead to SIDS, and therefore used the maximum outdoor temperature recorded on the same day and day before a SIDS day or control day as two main exposures. We used both days because the exact time of death was unknown, and we could not capture the exact temperature at the time of the event. Maximum temperatures the preceding day are certain to have occurred before death, but temperatures on the same day may have been reached only after death for a proportion of cases. To explore delayed impacts, we used the maximum temperature 2 days before a SIDS day or control day as a secondary exposure. Hourly temperature data were provided by Environment Canada by 24-hr block for the meteorological center situated approximately 20 km from Montreal's core (Smargiassi et al. 2009). Maximum daily temperature was modeled as a continuous variable, and odds ratios were estimated relative to a referent value of 20°C. This referent was selected because 20°C is a comfortable temperature generally not associated with thermal stress in Montreal, and has been used in previous research (Auger et al. 2014; Smargiassi et al. 2009).
Humidity was considered a potential confounder of the association between temperature and SIDS. Humidity data were collected at the same meteorological center as temperature and were measured using mean daily percent relative humidity (continuous). We did not adjust for air pollution, a potential causal intermediate in the pathway between temperature and mortality, because this could bias estimates of the total impact of temperature on SIDS (Buckley et al. 2014).
Statistical Analysis
We computed descriptive statistics and compared the proportion of SIDS deaths that occurred after days with maximum temperature ≥ 28°C with the proportion on the same days for all other causes.
We used the case-crossover design for its strength in assessing associations between transient exposures such as temperature and acute outcomes that are rare, such as SIDS (Maclure and Mittleman 2000). Case-crossover analysis is increasingly used to estimate impacts of temperature on mortality (Basagaña et al. 2011; Basu and Ostro 2008). In this design, each SIDS case serves as its own control, and the statistical analysis consists of comparing temperature exposure at the time of the event to temperatures during a short interval around the time of the case. Because each case is its own control, case-crossover designs inherently adjust for potential individual confounders that vary little over time, such as socioeconomic status, smoking, co-morbidities, and year of birth. In addition, this design accounts for potential bias from seasonal variation in conception and birth (Basso et al. 1995). Case-crossover studies are not ecologic, but use individual rather than aggregate data as the unit of analysis (Künzli and Tager 1997).
To select control days, we used an ambidirectional time-stratified approach where the referent period was the calendar month; we matched days on which SIDS occurred to control days consisting of the same weekdays of the month of death (Levy et al. 2001). If a death, for example, occurred on a Saturday in July 2000, control days comprised all remaining Saturdays in that month. Thus, any potential confounders that were stable during the month, such as socioeconomic status, were automatically adjusted for despite lack of data on such characteristics. We selected the same weekday as controls, thereby automatically adjusting for mortality that might vary by day of week. Because SIDS is rare, bias due to the use of control days that occur during the same month, but after the day of death, will be negligible (Lumley and Levy 2000).
We used conditional logistic regression to calculate odds ratios and 95% confidence intervals (CIs) for the association between maximum temperature of event days relative to temperatures of control days (each temperature variable was modeled separately), and included spline terms with knots at the 10th, 50th, and 90th percentiles (Durrleman and Simon 1989). We verified that use of knots located at the 5th, 50th, and 95th percentiles, and a greater quantity of knots, did not affect the shape of the curves (data not shown). All models were adjusted for relative humidity.
To estimate age-specific effects of temperature, we stratified SIDS cases into two postneonatal periods (1–2 vs. 3–12 months) for separate analyses, based on research suggesting increased susceptibility of infants older than 2 months to thermal stress (Fleming et al. 1992). In addition we performed a secondary analysis of associations at 3–6 months of age, but we did not have sufficient numbers of cases to estimate associations separately for SIDS deaths before 1 month of age (n = 13) or for 7–12 months of age (n = 14).
In sensitivity analyses, we ran models that excluded humidity, and examined associations for data restricted to summer months only (defined as June–August). We used the restricted cubic spline (RCS) macro in SAS version 9.2 (SAS Institute Inc., Cary, NC, USA) for statistical analyses (Heinzl and Kaider 1997). Data were anonymized, and the institutional review board of the University of Montreal Hospital Centre waived the requirement for formal ethics review.
