Short-term association between ambient temperature and acute myocardial infarction hospitalizations for diabetes mellitus patients: A time series study

Background Acute myocardial infarction (AMI) is the leading cause of death among people with diabetes mellitus (DM) and has been found to occur more frequently with extreme temperatures. With the increasing prevalence of DM and the rising global mean temperature, the number of heat-related AMI cases among DM patients may increase. This study compares excess risk of AMI during periods of extreme temperatures between patients with DM and without DM. Methods Distributed lag nonlinear models (DLNMs) were used to estimate the short-term association between daily mean temperature and AMI admissions (International Classification of Diseases 9th revision [ICD-9] code: 410.00–410.99), stratified by DM status (ICD-9: 250.00–250.99), to all public hospitals in Hong Kong from 2002 to 2011, adjusting for other meteorological variables and air pollutants. Analyses were also stratified by season, age group, gender, and admission type (first admissions and readmissions). The admissions data and meteorological data were obtained from the Hong Kong Hospital Authority (HA) and the Hong Kong Observatory (HKO). Findings A total of 53,769 AMI admissions were included in the study. AMI admissions among DM patients were linearly and negatively associated with temperature in the cold season (cumulative relative risk [cumRR] [95% confidence interval] in lag 0–22 days (12 °C versus 24 °C) = 2.10 [1.62–2.72]), while those among patients without DM only started increasing when temperatures dropped below 22 °C with a weaker association (cumRR = 1.43 [1.21–1.69]). In the hot season, AMI hospitalizations among DM patients started increasing when the temperature dropped below or rose above 28.8 °C (cumRR in lag 0–4 days [30.4 versus 28.8 °C] = 1.14 [1.00–1.31]), while those among patients without DM showed no association with temperature. The differences in sensitivity to temperature between patients with DM and without DM were most apparent in the group <75 years old and among first-admission cases in the cold season. The main limitation of this study was the unavailability of data on individual exposure to ambient temperature. Conclusions DM patients had a higher increased risk of AMI admissions than non-DM patients during extreme temperatures. AMI admissions risks among DM patients rise sharply in both high and low temperatures, with a stronger effect in low temperatures, while AMI risk among non-DM patients only increased mildly in low temperatures. Targeted health protection guidelines should be provided to warn DM patients and physicians about the dangers of extreme temperatures. Further studies to project the impacts of AMI risks on DM patients by climate change are warranted.

previous studies in regions around the world with higher rates of both being observed during periods of extreme heat or extreme cold [3][4][5][6][7][8]. However the strength of these associations and the threshold temperatures above (below) which heat (cold) effects become apparent varies geographically. Studies have shown that circulation to the skin [9] and sweating response [10] is impaired in diabetics. Hospital admissions among diabetics have been found to be more sensitive to heat wave effects [11]. A study on the relationship between inflammatory and coagulation responses to cold temperatures in patients with coronary disease [12] found that the association of several blood parameters with temperature was stronger in CHD patients who had diabetes vs. those who didn't and the authors concluded that diabetics were a susceptible subgroup which reacted more strongly to cold temperatures [12]. Our previous study on hot weather effects on mortality [3] estimated that deaths due to diabetes were 9.3% higher on days with a mean temperature of 31.2C vs a day with a mean temperature of 28.2C during the hot season in Hong Kong. This was not statistically significant however due to the small number of deaths with diabetes listed as the principal cause of death. Our study on cold weather and mortality [4] found a cumulative relative risk for diabetes deaths = 1.84 (95% CI = 1.42, 2.38, p < .001) for a 10C lower temperature over a 20 day lag period during the cool season in Hong Kong. This was the strongest association observed for all of the common causes of deaths in Hong Kong [4].
Our study looking at associations between weather and public hospital admissions in Hong Kong [13] found that hospitalizations due to respiratory and infectious diseases rose during very hot weather, while admissions for circulatory, respiratory and infectious disease rose during cold weather. This analysis did not examine the influence of co-morbidities on these associations. Recent increases in both DM prevalence and the frequency of extreme high temperatures pose serious public health threats particularly if those with DM are more sensitive to variations in meteorological conditions. A study specifically looking at the relationship between hospital admissions among those diagnosed with diabetes and meteorological variables is needed in order to determine whether the thermoregulation problems noted in diabetics leads to greater sensitivity of this group to extreme temperatures in a sub-tropical climate, and to aid in the patient management.

c)
Aims and Hypotheses to be Tested:  Our aim is to develop a statistical model relating daily hospitalizations among subjects with a Section 13: Proposed Research Project 2 prior diagnosis of DM to recent meteorological and environmental conditions, including temperature, relative humidity, wind speed, solar radiation and pollutant levels.
 We hypothesize that hospitalizations among diabetics will be more sensitive to high temperatures and high humidity in the hot season than those among non-diabetics.

d) Plan of Investigation:
(i) Subjects All admissions to Hong Kong Hospital Authority hospitals not due to external causes for the years from 2002-2011 from among subjects with a prior diagnosis of DM between 1998 and 2011 will be included in the models for diabetics. Admissions among those without a previous diagnosis of diabetes will also be included as a comparison group.
(ii) Methods Distributed lag non-linear models with daily hospitalizations as the outcomes will be used to estimate the association between recent meteorological conditions and morbidity.
(iii) Study design This will be a retrospective cohort time series (daily) study.

Statistical Analysis
Distributed lag non-linear models (DLNM) [Armstrong], an extension of Generalized Additive Models, will be used to assess the associations between meteorological variables and the daily hospitalization outcomes while taking into account non-linear and lagged effects, Daily mean temperature, relative humidity, solar radiation and wind speed will be modelled using splines with Section 13: Proposed Research Project 3 5 degrees of freedom each for the non-linear effect of each variable and the lag structure. The maximum lag considered will be 28 days. Daily levels of pollutants, including NO2, RSP, SO2, and O3 will be modelled using splines with 5 degrees of freedom for non-linear effect, 3 degrees of freedom for the lag effects and maximum lags of 10 days. Long term trends will be controlled using a smooth term with maximum degrees of freedom of 10 (1 per year). Seasonality will be controlled using a smooth term for day of the year (1, 2, …365 or 366 for leap years) with maximum of 5 degrees of freedom. Day of the week and holiday effects will be controlled using indicator variables. The R statistical software package version 3.0.2 will be used for all analyses with the packages mgcv() [Wood] and dlnm () [Armstrong] being used for the modelling. The mgcv() package uses cross-validation to select the appropriate df for each non-linear term given a maximum df supplied by the user. The fit of the model will be examined by checking the partial autocorrelation of the residuals, normality of residuals, and scatterplots of residuals vs. predicted values. Adjusted cumulative relative risks and 95% confidence intervals for hospitalization risk corresponding to changes in meteorological and pollutant variables will be estimated from the models with the cumulative effect being estimated out to the maximum lag at which the observed effect is significant. For each outcome we will also perform subgroup analysis by age group (<65, A research associate biostatistician will be needed to perform the complex data management, data cleaning, and the analyses for the project. The RA will also assist with the literature review and the process of writing the grant report and the papers for journal submission. g) Purpose and Potential: The results of this study will help to determine whether hospitalizations among people with diabetes are particularly sensitive to adverse environmental conditions compared to hospitalizations among Section 13: Proposed Research Project 4 those who do not have a previous diagnosis of diabetes. This information will be useful for researchers studying the physiological effects of diabetes and those studying the relationship between meteorological conditions and health outcomes. This information will also help clinicians in providing patient care and advice to diabetics. The models developed for daily hospitalizations among diabetics can be used to identify combinations of meteorological and environmental conditions which are particularly threatening to DM patients.
h) Key References: