Short‐, Medium‐, and Long‐Term Temperature and Sudden Cardiac Death in a Cohort of US Women

Abstract

Background

Sudden cardiac death (SCD) is an important contributor to the global burden of cardiovascular disease. Previous studies have reported an association with extremes in temperature in the hours and days leading up to SCD, but there has been little investigation focusing on the months and year before an SCD event. Therefore, the objective of this study was to investigate the relationship between short‐, medium‐, and long‐term temperature and SCD.

Methods

Temperature predictions using 800‐m Parameter‐Elevation Regressions on Independent Slopes Model data were used to estimate monthly moving average temperature. Data on temperature were appended to each Nurses' Health Study participant's geocoded residential addresses. SCD cases were determined on the basis of medical record review and next‐of‐kin reports. Time‐varying Cox proportional hazards models were used to assess the association between exposure 1‐ and 12‐month moving averages of temperature and SCD. Individual risk factors, socioeconomic status, and cardiovascular comorbidities were considered as confounders in the final models.

Results

Over the follow‐up period, there was a total of 428 definite and 36 probable SCD cases among the 118 636 participants. One‐ and 12‐month moving average temperatures were inversely associated with SCD in the basic, risk factor, and fully adjusted models. For example, in the parsimonious individual risk factor model with definite cases only, per interquartile range (15.27 °C for the 1‐month moving average temperature, 4.72 °C 12‐month moving average temperature) increase in temperature, the hazard ratios for definite SCD were 0.58 (95% CI, 0.44–0.77) for the 1‐month and 0.76 (95% CI, 0.66–0.88) for the 12‐month moving average.

Conclusions

After controlling for risk factors and cardiovascular morbidities, lower temperatures were associated with SCD. This study demonstrates that short‐, medium‐, and long‐term cold temperature also play an important role in SCD.

Nonstandard Abbreviations and Acronyms

NHS

Nurses' Health Study

SCD

sudden cardiac death

Research Perspective.

What Is New?

• Sudden cardiac death has been associated with low and high temperatures in studies focusing on short‐term temperature trends, and this study suggests that monthly and yearly temperature trends are also associated with sudden cardiac death.

• After adjusting for multiple covariates and cardiovascular comorbidities, the study suggests that lower 12‐ and 1‐month moving average temperatures were associated with an increased risk of SCD incidence.

What Question Should Be Addressed Next?

• Additional studies in other populations should be conducted to confirm the result that long‐term exposure to lower temperatures may be an important unidentified risk factor for sudden cardiac death.

Temperature extremes have been associated with both cardiovascular disease (CVD) and death. ¹ , ² , ³ In particular, multiple studies have investigated the relationship between temperature, notably in the form of heatwaves ¹ , ³ , ⁴ , ⁵ and CVD outcomes, although some studies have reported an association with lower temperatures. ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ Likewise, cardiac arrest and sudden cardiac death (SCD) have been associated with temperature changes and extreme temperatures in the hours and days leading up to the event, ¹¹ , ¹² , ¹³ , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ but there has been little investigation into the effect of medium‐ or long‐term temperature trends on risk of SCD, especially in women.

SCD is a pervasive public health problem. In women, CVD remains the number 1 killer and is responsible for approximately one third of deaths, ²² and awareness of CVD as a cause of death in women has been on the decline. ²³ As most victims of SCD have no previous diagnosis of heart disease, ²⁴ it is critical to identify population level risk factors that contribute to the burden of SCD in women. The literature on short‐term temperature and sudden cardiac events have tended to focus on out‐of‐hospital cardiac arrest. ¹² , ²⁵ , ²⁶ , ²⁷ Moreover, of the few studies that used SCD as their outcome, most have used registry or vital statistics and not clinically confirmed SCD. ¹³ , ²⁸ , ²⁹

The NHS (Nurses' Health Study) is a prospective cohort of US self‐identified female nurses. This cohort contains detailed health and demographic data, as well as addresses that have been geocoded to facilitate studies on environmental exposures with validated outcome data. To date, the time windows during which temperature poses a risk for SCD remains unknown. Therefore, the objective of this study was to examine the association between short‐, medium‐, and long‐term temperature exposures and SCD in the NHS.

Methods

Study Population

The NHS is a prospective cohort of 121 700 participants that began enrolling in 1976. Participants were married female nurses aged 30 to 55 years at the time of enrollment. Each study participant completed an in‐depth questionnaire, and participants complete follow‐up mail‐in questionnaires every 2 years with questions on comorbidities, risk factors, disease states, and demographics. With the return of each questionnaire, addresses for the participants are updated and geocoded. To protect participant privacy and comply with relevant regulations, the data are not publicly available. Further information, including the procedures to obtain and access data from the Nurses' Health Studies, is described at https://www.nurseshealthstudy.org/researchers (contact email: nhsaccess@channing.harvard.edu).

At initial enrollment in 1976, participants resided in 11 states (California, Connecticut, Florida, Massachusetts, Maryland, Michigan, New Jersey, New York, Ohio, Pennsylvania, and Texas), but as participants moved, by the 1990s there were least 10 participants in all of the 50 states and the District of Columbia. ³⁰ Participants who continued to return biennial questionnaires and had geocoded addresses were eligible for inclusion in our study. The Institutional Review Board of Brigham and Women's Hospital (Boston, MA) approved the original study protocol for the NHS, and informed consent was implied with the biennial return of the questionnaires.

Exposure

Monthly moving average temperatures were estimated using data from the Parameter‐Elevation Regressions on Independent Slopes Model climate group at Oregon State University (www.prism.oregonstate.edu) at 800‐m resolution. Data inputs for Parameter‐Elevation Regressions on Independent Slopes Model temperature models were obtained from >10 000 surface stations throughout the United States and Canada. Temperatures were predicted daily and summarized to provide monthly predictions. Parameter‐Elevation Regressions on Independent Slopes Model predictions have high accuracy, assessed through cross‐validation, and have been used in studies in the NHS, as well as in other cohorts. ³¹ , ³² , ³³ Monthly predictions of average temperature were assigned to all residential addresses on the basis of the 800‐m grid cell where they were located through a spatial join using the sf package in R version 4.2 (R Foundation for Statistical Computing, Vienna, Austria) and were linked to each participant's corresponding person‐month starting in 1986, the first year for which exposure data are available.

Outcome

The primary outcome for the analyses was definite SCD. Details on the classification of SCD in this cohort have been published previously. ³⁴ Briefly, SCDs were confirmed by physician review of medical records and next‐of‐kin reports about the circumstances surrounding the death if not adequately documented in the medical record. Cardiac deaths were considered definite SCDs if the death or cardiac arrest occurred within 1 hour of the onset of symptoms and there was no evidence of circulatory collapse or a neurologic event before the disappearance of the pulse. ³⁵ Unwitnessed deaths or deaths that occurred during sleep where the participant was documented to be symptom free within the preceding 24 hours were considered probable SCDs if an autopsy suggested that the death could have been arrhythmic. ³⁶ Patients with structural heart disease at baseline are excluded from our analysis. Validated death data for SCD are available in the NHS cohort until 2013.

Potential Confounders

We considered multiple individual‐ and neighborhood‐level confounders on the basis of their inclusion in previous studies and biological plausibility. Directed acyclic graphs were used to examine the structure of potential confounding relationships. Data for individual‐level potential confounders were obtained from biennial questionnaires. To capture and address potential confounding from neighborhood contextual factors, we included a neighborhood socioeconomic status score composed of Z‐standardized census tract variables to reflect multiple dimensions of socioeconomic status. ³⁷ Data for missing confounders were carried forward to the next questionnaire return cycle.

Statistical Analysis

Time‐varying Cox proportional hazards models were used to assess the association between 1‐ (short‐term) and 12‐month (long‐term) temperature exposures and confirmed SCD in separate models. Time‐varying temperature at each month was used to calculate the 12‐ and 1‐month moving average temperature for each participant. Additionally, we conducted a sensitivity analysis of 3‐ and 6‐month temperature exposure windows (medium‐term) and SCD. Person‐months were calculated from time origin in June 1986 until death, loss to follow‐up, or the end of available follow‐up for the SCD end point (May 2013), whichever came first. Potential confounders considered in our basic model were year, age, region, and season. Our parsimonious risk factor model included the basic model confounders along with smoking status, pack‐years, Alternative Healthy Eating Index diet score, neighborhood socioeconomic status, educational attainment, and marital status. Finally, our cardiovascular comorbidity model additionally included factors that could plausibly be confounders or mediators of the associations, including hypertension, diabetes, body mass index, and hyperlipidemia.

The basic model, parsimonious risk factor model, and comorbidity‐adjusted model were fit separately, and all models were stratified by time period and age in months. We examined the linearity of temperature at 1 and 12 months using cubic splines. ³⁸ We tested the proportional hazards assumptions using likelihood ratio tests with interaction terms for time period and age. A sensitivity analysis including probable cases of SCD was conducted to assess the robustness of our findings to a less restrictive definition of SCD.

Hazard ratios (HRs) with corresponding 95% CIs were calculated for each interquartile range (IQR) increase in each temperature window. All tests were 2‐tailed, and results were considered statistically significant at the α=0.05 level. Analyses were performed in SAS version 9.4 (SAS Institute Inc., Cary, NC).

Results

During follow‐up, 118 636 participants were observed over 32 883 943 person‐months between 1986 and 2013. A total of 428 definite and 36 probable cases of SCD were confirmed throughout follow‐up. There was an approximate normal distribution of the temperature exposures, and baseline demographics were broadly similar when compared across quintiles of exposure (Table 1). The average age of participants over follow‐up was 64.47±9.89. Of the participants included in our analyses, ≈44% reported hypertension, 9% reported diabetes, and 50% reported hyperlipidemia. There was little variation in these comorbidities across quintiles of temperature.

Table 1.

Age‐Standardized Characteristics of the Study Population and Temperature Distribution for Nurses' Health Study Participants (n=118 636) Throughout Follow‐Up From 1986 to 2013 Overall and by Quintile of 1‐Month and 12‐Month Moving Average Temperatures

Quintile, moving average month	Overall	First, 1‐mo	12‐mo	Third, 1‐mo	12‐mo	Fifth, 1‐mo	12‐mo
Moving average temperature range, °C		(−22.55 to 3.15)	(−0.60 to 9.41)	(10.70 to 16.94)	(10.61 to 12.04)	(21.5 to 38.82)	(16.37 to 34.62)
Person‐years	(n=32 883 943)	(n=6 578 730)	(n=6 576 793)	(n=6 576 148)	(n=6 576 790)	(n=6 580 158)	(n=6 576 789)
Age, y, mean±SD	64.47±9.89	63.81±9.83	63.62±9.85	64.86±9.81	63.81±9.86	65.14±9.95	66.28±9.67
Race and ethnicity
White	94	95	96	93	94	93	92
Black	2	1	0	2	2	2	3
Other race	4	4	4	5	4	4	5
Hispanic	1	0	0	1	0	2	3
Season of SCD
Summer	25	0	26	2	25	78	26
Autumn	25	4	25	37	25	14	26
Winter	25	83	25	15	25	0	25
Spring	24	14	25	45	24	8	24
Married	42	44	47	42	42	41	40
RN degree	73	73	75	72	72	72	71
Census region
Northeast	45	62	64	38	63	30	0
Midwest	15	23	24	12	20	11	0
West	15	3	0	14	2	34	53
South	12	1	2	23	2	11	33
Body mass index, continuous, mean±SD	23.25±9.87	23.61±9.73	23.96±9.25	23.13±9.85	23.42±10.01	23.03±9.99	22.52±10.01
Hypertension	44	44	44	44	45	45	44
Diabetes	9	9	9	8	9	9	8
Hyperlipidemia	50	50	50	49	49	50	50
Smoking status
Never smoker	44	43	44	44	43	44	46
Past smoker	41	41	42	41	41	41	40
Current	12	13	12	12	13	12	11
Pack‐years, mean±SD	11.25±17.03	11.42±16.99	11.53±17.05	10.97±16.89	11.31±16.98	11.32±17.18	10.93±17.26
Alternative Healthy Eating Index Diet Score, mean±SD	54.29±9.84	53.63±9.72	53.52±9.81	54.69±9.93	53.80±9.70	54.51±9.79	55.31±9.74
Neighborhood socioeconomic status, mean±SD	0.00±3.58	−0.32±3.39	−1.28±3.24	0.38±3.81	0.37±3.37	−0.15±3.40	−0.27±3.61

Values are means±SD or medians (interquartile ranages) for continuous variables; percentages or numbers or both for categorical variables and are standardized to the age distribution of the study population. RN indicates registered nurse; and SCD, sudden cardiac death.

Temperature distributions for both moving averages are shown in Table 2. The medians and IQRs were 13.79 °C and 15.27 °C (56.82 °F and 59.49 °F) for the 1‐month moving average temperature and 11.24 °C and 4.72 °C (52.23 °F and 40.50 °F) for the 12‐month moving average temperature. There was no evidence of nonlinearity in the 1‐month and 12‐month moving average temperature dose–responses (FigureFigure); therefore, we present estimates per IQR increase in temperature from models including only a linear term for temperature.

Table 2.

Temperature Distributions in °C for 118 636 Nurses' Health Study Participants Throughout Follow‐Up: 1986–2013 (32 883 943 Person‐Months)

	1‐mo moving average	12‐mo moving average
Mean	12.71	12.62
SD	9.24	4.26
Maximum	38.82	34.56
95th percentile	25.94	22.03
75th percentile	20.41	14.45
Median	13.79	11.24
25th percentile	5.14	9.73
5th percentile	−2.94	7.89
Minimum	−22.55	−0.60

Figure 1. Cubic splines describing the relationship between 1‐ and 12‐month moving average temperature (°C) and all SCDs in a study population of 118 636 Nurses' Health Study participants throughout follow‐up from 1986 to 2013 for 32 883 943 person‐months.

SCD indicates sudden cardiac death.

We observed a negative association between 1‐month moving average temperature and SCD in each of the basic parsimonious risk factors and comorbidity‐adjusted models (Table 3). Forest plots of potential confounders considered in our 3 models are presented in Figures S1 and S2. The sensitivity analysis of 3‐ and 6‐month exposure windows is included in Tables S1 through S3 and Figure S3. In our parsimonious risk factor model, the HR of SCD was 0.58 (95% CI, 0.44–0.77) per IQR increase of exposure of 1‐month moving average temperature for confirmed SCD and 0.58 (95% CI, 0.44–0.76) for all SCDs (confirmed and probable). Likewise, the 12‐month moving average temperature was also inversely associated with SCD in all 3 model types (Table 3), but the association was less pronounced than for 1‐month temperature. The HR of SCD was 0.76 (95% CI, 0.66–0.88) per IQR of exposure of 12‐month moving average temperature for confirmed SCD and 0.79 (95% CI, 0.68–0.90) for all SCD.

Table 3.

Basic and Multivariable HRs (95% CI) for 1‐ and 12‐Month Moving Average Temperature Exposures (per IQR Temperature in °C) and Risk of SCD Among 118 636 Women in the Nurses' Health Study Cohort Followed 1986 to 2013 (32 883 943 Person‐Months) With Sensitivity Analysis for Definite and Probable Cases

Exposure window	Definite SCD (428 cases)			Definite and probable SCD (464 cases)
	Basic*	Risk factor†	Comorbidity‡	Basic*	Risk factor†	Comorbidity‡
1‐mo moving average temperature	0.57	0.58	0.58	0.57	0.58	0.59
(IQR,15.27)	(0.43–0.76)	(0.44–0.77)	(0.43–0.78)	(0.44–0.75)	(0.44–0.76)	(0.45–0.77)
12‐mo moving average temperature (IQR=4.72)	0.76 (0.65–0.88)	0.76 (0.66–0.88)	0.77 (0.66–0.89)	0.78 (0.67–0.89)	0.79 (0.68–0.90)	0.79 (0.69–0.91)

HR indicates hazard ratio; IQR, interquartile range; and SCD, sudden cardiac death.

^*Basic model: year, age, season, region.

^† Parsimonious risk factor: Basic+smoking status, pack‐years, Alternative Healthy Eating Index dietary score, neighborhood socioeconomic status, registered nurse degree (yes/no), and marital status.

^‡ Comorbidity model: ^†Multivariable risk factor model+hypertension, body mass index, diabetes, and hyperlipidemia.

Discussion

Among the women in the NHS living across the United States, 1‐month and 12‐month moving average temperatures were inversely associated with SCD. To our knowledge, this is the first study to examine an association between temperature and SCD in time periods longer than the hours or days preceding the event. Effect estimates in our study were stable after adjusting for multiple confounders at the individual and neighborhood level. The sensitivity analysis including both confirmed and probable SCD were similar to our main analyses restricted to only confirmed SCD.

The decrease in the HR for SCD associated with each IQR increase in temperature was substantial. A 1‐IQR increase in temperature (15.27 °C for 1‐month and 4.72 °C for 12‐month moving averages) was associated with a 43% decrease in the HR of SCD at 1‐month moving average and 25% decrease in the HR of SCD at the 12‐month moving average. The results of this study are consistent with previous literature on SCD and cardiovascular disease. In general, there is a greater burden of CVD at lower temperatures compared with higher temperatures in the short and long term. ¹ There is a relatively consistent association between heatwaves and cold snaps with SCD, and lower temperatures have continually demonstrated stronger effects on cardiovascular outcomes than higher temperatures. ¹ It was previously unknown how hot and cold temperatures are associated with SCD in the medium and long terms. Our analyses provide additional clarification into this association between temperature and SCD and demonstrate that the association is primarily driven by lower temperatures.

These results should be interpreted in the context of the other literature in that short‐term temperature exposures are associated with SCD at both extremes while long‐term exposures are associated with SCD primarily at lower temperatures. In a variety of geographic settings ranging from warm environments, such as Israel, to colder environments, such as Norway, Finland, and Canada, both cold ¹² , ²⁵ , ²⁸ , ²⁹ , ³⁹ , ⁴⁰ , ⁴¹ , ⁴² , ⁴³ , ⁴⁴ , ⁴⁵ , ⁴⁶ , ⁴⁷ , ⁴⁸ , ⁴⁹ , ⁵⁰ , ⁵¹ , ⁵² , ⁵³ , ⁵⁴ , ⁵⁵ , ⁵⁶ , ⁵⁷ , ⁵⁸ , ⁵⁹ and heat ²⁵ , ²⁶ , ²⁸ , ⁴³ , ⁴⁷ , ⁵⁰ , ⁶⁰ , ⁶¹ , ⁶² , ⁶³ , ⁶⁴ were associated with elevated short‐term risks of out‐of‐hospital cardiac arrest and SCD in the hours and days leading up to the event. Several of these studies simultaneously investigated the association between short‐term temperature and out‐of‐hospital cardiac arrest or SCD at both lower and higher temperatures and reported J‐, U‐, or V‐shaped distributions. ¹¹ , ¹² , ¹³ , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹

There is strong biological plausibility between temperature and SCD. Low body temperature induces hyperviscosity and hemoconcentration of the blood, which can result in hypercoagulability. ⁶⁵ Cholesterol crystals also have increased deposition in low temperatures, thereby increasing the risk of atherosclerotic plaque rupture. ⁶⁵ Exposure to cold leads to systemic vasoconstriction, a rise in blood pressure via sympathetic catecholamine activity, and resultant increased myocardial oxygen demand. ⁶⁶ Likewise, in short‐term studies, cold has been associated with a precipitation of angina pectoris ⁶⁷ and the ischemic subtype out‐of‐hospital cardiac arrest. ²⁹ , ³⁹ , ⁵⁸ In the long term, colder temperatures have been associated with an increase in blood lipids, ⁷ blood pressure, ⁹ and inflammatory mediators. ⁸ The combination of short‐ and long‐term changes in cardiovascular physiology in response to cold contribute to the biological plausibility of the observed inverse association between temperature and SCD.

Although warmer temperatures appear to have a protective effect in the short and long terms, this study does not refute the growing body of literature on the hours and days leading up to out‐of‐hospital cardiac arrest/SCD which show a J‐shaped distribution with increased risk at both cold and warm temperatures. ¹¹ , ¹² , ¹³ , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ , ²¹ Moreover, the distribution of SCD events in the long term may be more linear when compared with that of short‐term studies, which tend to follow a J‐shaped distribution. Combined with previous studies that have posited that short‐term temperature changes are associated with SCD, this study provides additional clarification on how different time intervals of exposure before the event pose different risks for SCD. In particular, colder temperatures are associated with SCD risk in the short, medium, and long term.

This study has a few limitations. Despite extensive data on confounders, there remains potential for unmeasured and residual confounding. With respect to the latter, previous validation studies in the NHS have demonstrated that participants accurately report both cardiovascular disease outcomes and risk factors in their follow‐up questionnaires. ⁶⁸ There also remains potential for misclassification in assignment of temperature exposure since some participants may spend significant amounts of time in climates that differ from their geocoded address or move seasonally during the year. This potential exposure misclassification, however, is likely to be nondifferential with respect to outcome, thus biasing results toward the null.

This study also lacks a comparator group as the SCD rate in this cohort is low. We considered the use of comparator groups in cohorts with high SCD rates, and there was not a suitable cohort with the same detail of covariates, a validated outcome, and length of follow‐up as the NHS. Moreover, we did not find another comparator group with address‐level temperature data, as most studies used zip code–level data. The incidence of SCD in women in similar contemporary studies is slightly lower in the NHS cohort. ⁶⁹ , ⁷⁰ , ⁷¹ , ⁷² This finding, however, is consistent with the cohort being composed of healthy nursing professionals. Nonetheless, the findings from this study confirm that the majority of SCD cases still occur in women without known underlying cardiovascular disease.

This study is from a cohort of women who mostly self‐identify as White individuals. The homogeneity of the cohort potentially limits the generalizability of the results to other segments of the US population. The NHS cohort is representative of the demographics of US nurses at the time of study commencement in 1976. Moreover, the NHS has a follow‐up rate of >90% with participants providing biennial questionnaires with accurate and detailed data provided by nurses who are familiar with reporting health outcomes and risk factors, thereby leading to high internal validity. ⁷³ It is possible that the homogeneity of the NHS may impact the generalizability of these results to populations with different distributions in exposure to temperature. Despite potential limits to the generalization, the homogeneity of the cohort in terms of age, sex at birth, and occupation at enrollment provide for high internal validity and reduce the potential for unmeasured confounding.

This study has multiple strengths. The NHS is a source of in‐depth data on potential confounders, modifiers, and mediators of an association between temperature and SCD in women. Moreover, we have a robust measurement of neighborhood socioeconomic status to address additional contextual factors that could potentially confound the relationship between temperature and SCD. The temperature data that are appended to this cohort provide precise estimates at the address level compared with models that use lower resolution estimates often at the zip code level. Participants from the NHS live throughout the United States and have varying levels of temperature exposure, allowing us to assess the relationship in a broad setting. The large number of rigorously confirmed SCD events, a difficult phenotype to classify population studies, is also a major strength. Although the rigorous definition likely resulted in some degree of underdetection of SCD in this cohort, the high specificity of our definition limits misclassification and provides a less biased risk estimate. Therefore, the results from this study have reliable estimates of exposure, outcome, and confounders.

Conclusions

In a population of women living throughout the entire United States, lower 1‐ and 12‐month moving average temperatures were associated with a higher HR of SCD over a time span of nearly 3 decades. The association between 1‐ and 12‐month moving average temperatures and SCD was stable after adjusting for individual and neighborhood characteristics, risk factors, and cardiovascular comorbidities. This study emphasizes the importance of investigating longer‐term environmental exposures and acute cardiovascular events. Temperature is an unavoidable environmental exposure, and a significant proportion of SCD victims involve individuals who are considered to be at low risk for CVD. Consequentially, the findings of this study underscore the relevance of ubiquitous environmental exposures and the public health burden of SCD.

Sources of Funding

This work was funded by National Institutes of Health (NIH) Grants UM1 CA186107 and R01 HL034594. W.B. and G.E.W. were supported by NIH Grant T32 ES007069. C.R. was supported by R01ES026246‐01 and W81XWH2210030. R.C.N. was supported by NIH K01ES032458. J.H. was supported by American Heart Association 13SDG14580030 and NIH P30 ES000002, R01 ES028033, R01 ES029840, U01 HL145386, and R24 ES028521. F.L. was supported by NIH P30 ES000002, R01 ES028033, R01 ES029840, U01 HL145386, and R24 ES028521 The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH or other funding sources.

Disclosures

None.

Supporting information

Data S1. STROBE Statement

Tables S1–S3

Figures S1–S3