An event sequence stratified lag method for assessing the risk of cardiovascular and cerebrovascular disease mortality under complex environmental exposure

Abstract

This invention relates to an event sequence stratified lag method for assessing the risk of cardiovascular and cerebrovascular disease (CCVD) mortality under combined environmental exposures, belonging to the field of public health and environmental health risk assessment technology. This method quantifies different individual events (such as cold waves, PM2.5, etc.) based on daily CCVD mortality data and corresponding meteorological and environmental information. 2.5 This method investigates the risks posed by combined meteorological and environmental exposures (multiple events and sequence interactions) and their lags to CCVD mortality, supporting a more scientific and rational approach to capturing the impact of various events and their complex sequence relationships on risk. The method is applicable to risk assessment under integrated meteorological and environmental exposures (multiple events and sequence interactions) and provides a scalable and reusable tool for risk comparison across different regions.

Description

Technical Field

[0001] This invention belongs to the field of public health and environmental health risk assessment technology, specifically relating to an event sequence stratified lag method for assessing the risk of death from cardiovascular and cerebrovascular diseases under combined environmental exposures. Background Technology

[0002] Cardiovascular and cerebrovascular diseases (CCVD) are among the leading causes of death worldwide, causing nearly 1.8 million deaths annually. The occurrence and progression of CCVD are influenced by multiple factors, primarily including genetic susceptibility, lifestyle behaviors (such as smoking and dietary choices), and environmental exposure. Among these factors, environmental impacts are receiving increasing attention, with air pollution and extreme temperatures increasingly considered significant contributors to CCVD morbidity and mortality. Systematically assessing the impact of combined environmental exposures on CCVD and uncovering its complex spatiotemporal lag patterns is crucial for developing public health interventions.

[0003] Existing research has shown that fine particulate matter (PM2.5) is a major pollutant in the air. 2.5 PM has a clear association with CCVD. From a cardiovascular pathology perspective, PM 2.5 It can penetrate deep into the alveoli of the lungs and enter the bloodstream, triggering systemic inflammation, oxidative stress, endothelial dysfunction, and even autonomic nervous system imbalance. [1] This damages atherosclerotic plaques, leading to myocardial ischemia, infarction, and stroke. [2] Extensive epidemiological evidence suggests that PM2.5 2.5 Exposure is associated with an increased risk of mortality from CCVD. According to the report, short-term exposure to PM... 2.5 For every 10 μg / m³ increase in PM 2.5 The mortality rate from CCVD increased by approximately 1.7%. [3] Long-term PM 2.5 For every 10 μg / m³ increase in exposure, the mortality rates from ischemic heart disease and cardiovascular disease increase by approximately 23% and 24%, respectively. [4] .

[0004] In addition, cold waves (CS), characterized by persistent abnormally low temperatures, are often associated with increased mortality from chronic cerebral vasoconstriction (CCVD): CS can lead to peripheral vasoconstriction, elevated blood pressure, and increased blood viscosity. [5] This increases the vulnerability of the myocardium and cerebrovascular system, promotes the formation of a hypercoagulable state, and thus increases the risk of thrombosis, myocardial ischemia, stroke, and arrhythmia. [6] Specifically, cold waves increase the cardiovascular disease mortality rate by an average of approximately 32%. [7] For every 1°C drop in temperature, the mortality rate from cardiovascular disease increases by approximately 1.6%. Cold waves are also associated with elevated blood pressure and coagulation problems in patients with ischemic stroke. [8]Each additional day of cold weather increases the overall risk of stroke by 3%, and the risk of ischemic stroke by approximately 5%. [9] .

[0005] Although numerous meteorological and environmental exposure factors have been revealed to be associated with the risk of mortality from CCVD, most studies and methods can only analyze and address the impact of single factors, neglecting the combined effects of multiple meteorological and environmental exposures on CCVD. In reality, most situations involve multiple exposures, such as extreme weather events and air pollution often occurring simultaneously or sequentially, which may exacerbate the risk of mortality from CCVD. While some evidence suggests that CS and PM2.5 are contributing factors... 2.5 There are certain synergistic or interactive effects among these factors, but existing methods have failed to fully capture the nonlinear lag effect of combined meteorological and environmental exposures on CCVD mortality risk. The analysis of different combined exposures and their lag time intervals on CCVD mortality risk remains limited, and the methods supporting susceptibility differences among different subgroups are still insufficient.

[0006] Distributed lag nonlinear modeling (DLNM) is a method that can capture the nonlinear relationship between exposure and response. It employs a flexible statistical modeling framework that can simultaneously characterize the nonlinear exposure-response relationship and lag effects between variables, and can effectively support the estimation of the disease risk distribution of a single meteorological environmental exposure at different time lags.

[0007] When the measurement errors of exposure data (such as temperature and pollutant concentration), outcome data (such as morbidity and mortality), and confounding factor data are within acceptable ranges, without systematic bias, and assuming that the core relationship of exposure-response-lag is relatively stable, the DLNM model transforms the three-dimensional relationship of "exposure level-lag time-health outcome" into an estimable two-dimensional linear model by simultaneously fitting the nonlinear relationship and lag effect of exposure using cross-basis functions. This allows the model to simultaneously characterize the immediate and lagged, linear and nonlinear effects of exposure.

[0008] The DLNM model divides continuous exposure variables into multiple intervals based on nodes. Each interval is fitted with a smooth curve, and the intervals are continuously differentiable. These intervals are then concatenated into a smooth nonlinear curve to fit the nonlinear relationship between exposure level and outcome. It uses lag-dimensional basis functions (Lag Basis) to fit the decay, fluctuation, and superposition patterns of lag effects. DLNM uses a cross-basis matrix, generated by tensor products of exposure and lag basis functions, to form a new set of cross variables carrying information on both exposure level and lag time (representing both the nonlinearity of exposure and the lag effect). Its main steps are: 1. Data preparation: Organize time series data, including outcome variables (daily mortality / illness), core exposure variables (temperature, pollutant concentration), confounding variables, and time variables. Determine the maximum lag days (Lmax). 2. Construct cross-basis: Specify the type, degrees of freedom, and node positions of the exposure and lag basis functions, and generate the cross-basis matrix through tensor product. 3. Construct a regression model, using the cross-basis matrix as the core independent variable, incorporating time trends and confounding factors, and selecting an appropriate distribution (Poisson, quasi-Poisson, negative binomial, Gaussian, etc.) to fit a generalized linear model. 4. Model diagnostics: Examine the residuals for autocorrelation and excessive discretization, check the reasonableness of the basis function degrees of freedom, and verify the statistical significance of nonlinearity and lag effects. 5. Effect estimation and visualization: Predict and calculate the cumulative effect and lag effect at a specific exposure level, plot lag effect curves, exposure-response curves, and 3D surface plots, and interpret the results.

[0009] Although the DLNM model is widely used in assessing disease risk from a single exposure, it still has the following limitations when assessing complex environmental exposures: (1) It does not support risk assessment of sequential exposures and lags. DLNM is mainly for single exposures that change continuously over time. It cannot directly define, identify, and analyze sequences of multiple exposure events, cannot handle the complex temporal relationships between these events, and cannot handle the time-lag effects of the sequence event relationships. (2) In the DLNM analysis framework, only the core exposure variables are considered, and the risk differences resulting from different event sequences and intervals are not characterized. The concepts of events and sequences are not native modeling objects. In application, these variables are usually implicit and unstandardized, depending on the researcher's subjective understanding and settings, resulting in a lack of comparability between different studies.

[0010] In summary, the existing technology has the following drawbacks:

[0011] (1) Existing technical methods still have limitations in assessing the risk of death from cardiovascular and cerebrovascular diseases under combined meteorological environmental exposure. Numerous epidemiological studies have shown that combined environmental exposure has an aggravating effect on health. Extreme weather and air pollution events often occur simultaneously or sequentially, which may exacerbate the risk of death from cardiovascular and cerebrovascular diseases.

[0012] (2) Although the existing distributed lag nonlinear model (DLNM) can simulate the nonlinear relationship and lag effect of a single exposure, it does not support the assessment of sequential exposure and lag. In its analytical framework, it only focuses on the core exposure variable and does not characterize the risk differences caused by different event sequences, intervals, etc.

[0013] (3) Most existing methods have vague definitions of concepts such as exposure events, sequences, hysteresis and stratification, which are difficult to reuse and extend, and cannot effectively support risk assessment of complex environmental exposure. Summary of the Invention

[0014] The purpose of this invention is to provide an event sequence hierarchical lag method for assessing the risk of cardiovascular and cerebrovascular disease mortality under combined environmental exposures. This method defines the exposure events, sequences, and hierarchical lag structures, and integrates these structures into a distributed lag nonlinear model (DLNM) for assessing the risk of cardiovascular and cerebrovascular disease mortality under combined environmental exposures. The event sequence hierarchical lag (ESSL) method is used to assess different CS and PM levels. 2.5 The study of the risk ratio (RR) of co-occurring vegetative-death (CCVD) under event combinations and their lag relationships revealed that the risk posed by combined exposures is influenced by differences in exposure sequence, gender, and age. The proposed stratified lag method for event sequences has the ability to assess the differential risk resulting from combined exposures in different environments, providing new insights for health interventions targeting vulnerable populations.

[0015] To achieve the above objectives, the technical solution of the present invention is: an event sequence stratified lag method for assessing the risk of death from cardiovascular and cerebrovascular diseases under combined environmental exposures, comprising:

[0016] S1. Acquire and organize time series data for the target region and target time period, including the daily number of deaths from cardiovascular and cerebrovascular diseases, as well as the daily measurements of meteorological and environmental exposure variables.

[0017] S2. Set event thresholds for each meteorological or environmental exposure variable to be evaluated, and identify single events that meet the threshold conditions and joint events that occur simultaneously within the same time unit based on time series data.

[0018] S3. Based on the identified events, extract from the time series data a sequence consisting of two or more events occurring in a predetermined order, with the time interval between occurrences within a preset maximum lag day.

[0019] S4. Apply the predetermined sequence filtering rules to the sequences extracted in step S3 to filter out the closest related source events for each target event in order to determine the effective sequences for modeling.

[0020] S5. Construct a distributed lag nonlinear model with a hierarchical lag structure for the fusion event sequence. The distributed lag nonlinear model uses the single event, joint event, or effective sequence determined in step S2 as the core independent variable, the daily number of deaths from cardiovascular and cerebrovascular diseases as the dependent variable, and controls for long-term trends, seasonality, and confounding variables.

[0021] S6. Using the model constructed in step S5, calculate and output the risk ratio and its statistics for cardiovascular and cerebrovascular disease mortality under different single events, combined events, or event sequences of exposure.

[0022] Furthermore, in step S2, the exposure variable X of the joint event... t Defined by the following formula:

[0023]

[0024] Among them, E i,t ∈{0,1} represents event E i Whether it occurs on day t, where n is the number of event types considered; X t The value corresponds to 2 n A possible combination of events.

[0025] Furthermore, in step S4, the sequence filtering rule is: for a type B event E occurring at a predetermined time... Bj When multiple linkable Class A events exist within the maximum lag period, only those related to E are retained. Bj A sequence of A-type events that are closest in time.

[0026] Furthermore, in step S5, the distributed lag nonlinear model is a generalized linear model based on the quasi-Poisson distribution, and its basic form is as follows:

[0027]

[0028] in, This represents the expected number of deaths on day t. The threshold is the intercept, and lag is the lag period length, representing the duration of the exposure's effect on the outcome. Describe the exposed variables Cross basis matrix of nonlinear delay effect during lag period This represents a natural spline function applied to potential confounding factors such as meteorological variables and air pollutants. This represents the term used to control for residual seasonality and long-term time trends. These are covariates, representing other variables that need to be controlled at time t. Common examples include relative humidity, other air pollutants, and air pressure. These are the degrees of freedom of the natural spline function. The degrees of freedom of the spline function representing the covariates are used to characterize the nonlinear relationship between the covariates and the outcome. The degrees of freedom of a time spline function are used to control long-term time trends and seasonal variations. This indicates the matching layer in the case-crossover design, which effectively controls for time-invariant confounding factors.

[0029] Furthermore, in step S5, when the core independent variable is an event sequence, the exposed variable is... The value of is determined by the hysteresis state of the sequence, and is defined as follows:

[0030]

[0031] Where m is a positive integer representing the specific stratum to which the lag time interval between two events in the sequence belongs.

[0032] Furthermore, in step S3, the maximum lag days Lmax is determined based on the model fit index and prior knowledge.

[0033] Furthermore, in step S5, the construction of the distributed lag nonlinear model also includes adjusting the degrees of freedom of the exposure and lag dimensions in the cross basis function, the degrees of freedom of the spline function, and the maximum lag days based on the model diagnostic results.

[0034] Furthermore, in step S5, the distributed lag nonlinear model supports stratified analysis of the risk ratio, with stratification dimensions including at least one of age and gender.

[0035] Furthermore, in step S2, meteorological and environmental exposure variables include ambient temperature and fine particulate matter (PM2.5). 2.5 Concentration of inhalable particulate matter PM 10 At least two of the following: concentration of sulfur dioxide (SO2), concentration of nitrogen dioxide (NO2), concentration of ozone (O3), concentration of carbon monoxide (CO), and relative humidity.

[0036] Furthermore, in step S6, the output risk ratio includes at least one of the following: daily lagged risk ratio, cumulative lagged risk ratio, and risk ratio stratified by different age or gender.

[0037] Compared to existing technologies, this invention offers the following advantages: The method of this invention (ESSL) integrates events, sequences, and hierarchical lag structures (including intervals, age, and sex) into a DLNM model, enabling it to more effectively assess the mortality risk of cardiovascular and cerebrovascular diseases (CCVD) under comprehensive climatic environmental exposure. This method overcomes previous limitations, clarifies unstandardized and ambiguous concepts such as events, sequences, lags, and stratification, and captures various events in the disease risk assessment process, as well as the complex sequences and interactions between these events that influence risk, in a more scientific and rational manner. This provides information for targeted public health interventions and guides future mechanistic research. Attached Figure Description

[0038] Figure 1 For the sequence and its lag.

[0039] Figure 2 A comparison of the single-day lag effects of joint events.

[0040] Figure 3 For sequence comparison under different stratified lag intervals.

[0041] Figure 4 The daily and cumulative lag effects of joint events and sequences under PC04.

[0042] Figure 5 For PC04, compare the day-lag effects of joint events and sequences by gender and age stratification. Detailed Implementation

[0043] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings.

[0044] This invention provides an event sequence stratified lag method for assessing the risk of death from cardiovascular and cerebrovascular diseases under combined environmental exposures, comprising:

[0045] S1. Acquire and organize time series data for the target region and target time period, including the daily number of deaths from cardiovascular and cerebrovascular diseases, as well as the daily measurements of meteorological and environmental exposure variables.

[0046] S2. Set event thresholds for each meteorological or environmental exposure variable to be evaluated, and identify single events that meet the threshold conditions and joint events that occur simultaneously within the same time unit based on time series data.

[0047] S3. Based on the identified events, extract from the time series data a sequence consisting of two or more events occurring in a predetermined order, with the time interval between occurrences within a preset maximum lag day.

[0048] S4. Apply the predetermined sequence filtering rules to the sequences extracted in step S3 to filter out the closest related source events for each target event in order to determine the effective sequences for modeling.

[0049] S5. Construct a distributed lag nonlinear model with a hierarchical lag structure for the fusion event sequence. The distributed lag nonlinear model uses the single event, joint event, or effective sequence determined in step S2 as the core independent variable, the daily number of deaths from cardiovascular and cerebrovascular diseases as the dependent variable, and controls for long-term trends, seasonality, and confounding variables.

[0050] S6. Using the model constructed in step S5, calculate and output the risk ratio and its statistics for cardiovascular and cerebrovascular disease mortality under different single events, combined events, or event sequences of exposure.

[0051] The following is a detailed implementation process of the present invention.

[0052] This invention designs an event sequence hierarchical lag (ESSL) method for assessing the risk of death from cardiovascular and cerebrovascular diseases under complex environmental exposures. Its core involves developing event, sequence, and hierarchical lag structures and integrating them into an existing distributed lag nonlinear model (DLNM) to capture the nonlinear relationship and lag effects of health risk responses under complex (complex) meteorological environmental exposures. Specifically:

[0053] An event (E) is a fundamental concept of core exposure. When a climate or environmental core exposure variable reaches a pre-set specific condition within a certain period (e.g., date), it is recorded as E=1; otherwise, E=0. Events are divided into two types: single events and joint events.

[0054] (1) Single event

[0055] A single event refers to an exposure event that occurs independently within a specified time (such as a date), and no other events occur within that date. For example, if an exposure occurs on a day with only average daily rainfall exceeding a preset threshold (such as ≥20 mm), it is classified as a single event.

[0056] (2) Joint events

[0057] The concept of a combined event refers to two or more events occurring simultaneously at a specified time (such as on a single day). For example, if a day experiences both a daily average rainfall exceeding a preset threshold (e.g., ≥20 mm) and a daily average wind speed exceeding a threshold (e.g., ≥8 m / s), then that day is considered to have experienced a combined event.

[0058] (3) Event segment

[0059] An event segment refers to a type of event that occurs continuously within a certain number of time units (without interruption).

[0060] (4) Sequence

[0061] A chain structure of exposure events is used to represent two or more events occurring in a specific order, i.e., after one event occurs, another event occurs within a specific time interval (lag). This lag time interval represents the time distance between the two events and can be subsequently incorporated into a discrete-time nonlinear mixed-effects model (DLNM) for hierarchical lag analysis.

[0062] Suppose a sequence consists of two events (i.e., E) A and E B It consists of two events, with an interval of ≤ 6 days between them (the preset maximum lag). It can be simplified to a pair of events (denoted as "E"). A -E B ), that is, the first event (E) occurs first. A Then another event (E) occurs within the interval (Δt). B For example, in Figure 1, E A1 -E B4 This represents a valid sequence, and E A1 -E B8 Then it is not (because E) B8 Not in E A1 (occurring within the interval).

[0063] (5) Sequence lag and screening rules

[0064] For a set of time series data observed during the process, at time point j, there may be multiple events related to type B (E). Bj Related Class A events. For stratified lag analysis, sequence selection rules need to be defined. For specific Class B events E... Bj Assuming there may be many links E Ai -E Bj (j-lag <= i <= j), as shown below:

[0065] (1)

[0066] However, not all links listed in equation (1) satisfy the sequence condition. This is because there may be some links without E. A The event. Therefore, Meaning link Not a sequence. Link to E Bj The sequence is as follows:

[0067] (2)

[0068] in, Representative and E Bj Several related event segments, and and These represent the first and last event segments, respectively. For E... Bj The selection rule for the sequence is to obtain the closest event segment from equation (2), that is, except for In addition to this, all other sequences were removed. For example, Figure 1 All sequences shown in the table are marked with a checkmark (√) after applying this filtering rule. Conversely, sequences that are rejected are marked with a cross (×), as shown in Table 1.

[0069] Table 1. Examples of sequence screening results

[0070]

[0071] (6) Lag stratified modeling

[0072] This study employs a cross-basis function approach to design and quantify the association between exposure events (or sequences) and health risk responses. This method selects events from the same year, month, and day of the week as the date of cardiovascular death, mitigating potential confounding factors such as long-term trends and seasonal variations. DLNM is introduced within the framework of generalized nonlinear modeling (GNM) to handle nonlinear relationships and lag effects.

[10] And Poisson regression is used to address the excessive dispersion problem in daily death counts:

[0073] (3)

[0074] In equation (3), Indicates the first The expected daily count of health outcomes (number of deaths) for the day. It is the intercept. Indicates the description of exposure Cross-basis matrix of nonlinear delay effect during the lag period. This represents a natural spline function applied to potential confounding factors (such as meteorological variables and air pollutants), while This indicates the item used to control for residual seasonality and long-term time trends. This represents the matching layer in the case-crossover design, effectively controlling for time-invariant confounding factors. Model parameters, including the degrees of freedom and maximum lag length of the spline terms, are typically determined based on model fitting criteria (such as the Quasi-Akaike Information Criterion (Q-AIC)) and prior knowledge, i.e., previous literature.

[11] Determined. The discrete-time nonlinear model (DLNM) employs two lag settings (daily lag and cumulative lag) to evaluate the lag effect, which can be defined as:

[0075] (4)

[0076] In equation (4), It is an integer representing the lag days. Exposure variable. Different formats can be configured based on the analysis objectives. Below are examples of events and sequences with different stratification intervals. The explicit structure.

[0077] (I) Event

[0078] For events, variables It can be defined as:

[0079] (5)

[0080] In equation (5), E i,t ∈{0,1} represents event E i Whether it occurs on day t (1 = yes, 0 = no), where n is the number of event types considered; for the case of joint events, there may be at most 2 n A combination of exposure events, X t This represents each unique state corresponding to these combinations. For example, if n = 2 (there are only two events, E...) A and E B A state of 0 indicates no event, and a state of 1 indicates that only event E occurred. A , 2 indicates that only event E occurred. B 3 represents E A and E B The events occurred simultaneously.

[0081] (II) Sequences and their hysteresis stratification

[0082] For sequences, the primary focus is on their lag stratification state. Therefore, the value of Xt depends on the lag stratification setting. Stratification of lag intervals is based on three practical principles. First, select cutoff points that reflect the biological or mechanistic effect window (e.g., acute and subacute phases). Second, each stratum should contain a sufficient number of events (e.g., 20) to stabilize the estimate. Third, employ a data-driven strategy (e.g., a quantile-based strategy) to balance the sample size of each stratum. Equation (6) represents the lag stratification state of the sequence (Xt). t ).

[0083] (6)

[0084] Based on the above method, and using daily cardiovascular and cerebrovascular death case data from a certain city from 2015 to 2021, the risk profile of vulnerable subgroups under different events, sequences, and lag time windows was analyzed. The specific steps are as follows:

[0085] ① Input the time series data of daily CCVD deaths and environmental exposure variables such as air pollution, as shown in Table 2, which describes the data descriptive statistics used in this use case.

[0086] Table 2. Time series data of daily CCVD deaths and exposure variables such as air pollution from 2015 to 2021.

[0087]

[0088] Note: PM 2.5 Fine particulate matter; PM 10 Particulate matter; SO2: Sulfur dioxide; NO2: Nitrogen dioxide; O3: Ozone; CO: Carbon monoxide; SD: Standard deviation; Min: Minimum value; Max: Maximum value; RH: Relative humidity

[0089] ② Set a time unit (e.g., date), summarize the number of CCVD deaths by time unit, and set event thresholds based on the time series data of the exposure variable to identify each event in the time series, such as... Figure 1 E in A and E B .

[0090] ③ Set the maximum lag time (e.g.) Figure 1 (lag = 6 days), and determine the event segments and sequences based on the events in each time series, such as Figure 1 Event segment E in A1A2A3 , E B4B5B6 E, sequence E A1- E B2、 E A1- E B4、 E A5- E B6 wait.

[0091] ④ Filter out the sequences that meet the criteria according to the sequence filtering rules. The filtering results are similar to those shown in Table 1.

[0092] ⑤ Perform hysteresis stratification modeling on the filtered sequences.

[0093] ⑥ Generate risk assessment results for various events and sequence exposures, and conduct countermeasure analysis.

[0094] Following the steps outlined above, an analysis of CCVD in a certain city revealed that the synergistic effect of temperature and air pollution exacerbated the risk of CCVD mortality. Specifically, the daily CCVD mortality rate, temperature, and PM2.5 were among the factors contributing to this risk. 2.5 There is a clear seasonal synergistic relationship between concentrations—in summer, as temperatures rise and PM2.5 concentrations increase... 2.5 Lower levels of CCVD reduce the risk of death; however, in winter, as temperatures decrease and PM levels rise, the risk of death decreases. 2.5 Increased concentrations significantly led to a rise in CCVD mortality. To assess the impact of a single cold wave (CS) event, CS events were categorized into 12 classes (CS01-CS12), and the resulting CCVD mortality risk ratios (RR) were analyzed. Groups were formed based on daily temperatures consistently below the lowest 2.5%, 5%, 7.5%, or 10% of the annual daily temperature (i.e., the 2.5, 5, 7.5, or 10th percentile) for at least two, three, or four days, respectively. Use case analyses, including single-event, sequence, and stratified sequence lag analyses, were conducted on this data, preliminarily validating the effectiveness of the proposed ESSL method. Details are as follows:

[0095] I. Single Event Assessment

[0096] This case study explores the impact of a single cold wave CS event, including PM2.5. 2.5 The value is included as an external covariate in equation (3) Item. To assess the impact of these CS event categories, in equation (5) Only binary values ​​(i.e., 1 and 0, indicating whether a CS event has occurred) are used. Table 3 lists the 12 types of CS events and their RR values ​​and 95% confidence intervals (CI) at a cumulative lag of 14 days (lag014).

[0097] Table 3. Event categories of CS (CS01–CS12) and their Q-AIC, RR values, and 95% CI in lag014

[0098]

[0099] Table 4 shows the one-day lag effect results for 12 types of CS events. Overall, the RR generated by CS events shows an increasing trend from CS01 to CS12. For a one-day lag of 0, CS04 has the highest RR. For most CS events, the RR peaks at a lag of 5 days, and then gradually weakens with increasing lag days. At more extreme temperature thresholds and longer durations, CS events are associated with an increased RR of CCVD mortality.

[0100] Table 4. Day-Lapse Effects of 12 Types of CS Events

[0101]

[0102] II. Joint Event Risk Assessment

[0103] This case study explores CS and PM 2.5 Joint events (i.e., "PM") 2.5 The risk of "+CS" is discussed. According to formula (5), the variable X in formula (3) has only four states: 0 indicates that there is no CS or PM. 2.5 Event; 1 indicates only CS event; 2 indicates only PM event. 2.5 Event; 3 indicates "PM" 2.5 The joint event of "+CS".

[0104] Figure 2 illustrates the effects of the joint event "PM". 2.5 +CS and CS and PM 2.5 Comparison of estimated RR (95% CI) for individual events. To ensure consistency in assessing joint events, the joint event "PM" is considered. 2.5 "+CS" is divided into 12 different categories (PC01-PC12), corresponding to the 12 categories of CS events mentioned earlier (CS01-CS12). This is in contrast to single CS events (CS01-CS12) and single PMs. 2.5 Compared to time-related CCVD mortality rates, the RR caused by the 12 combined event categories (PC01-PC12) was the largest, with all of these combined events (PC01-PC12) producing high RR within a 2-day lag. Single PM 2.5 The RR generated by the event is higher than that generated by a single CS event. Although the joint event "PM" was found... 2.5 "+CS" is more harmful than a single incident, but "PM" 2.5 The RR generated by "+CS" may still be due to PM 2.5 The order in which CS events occur differs.

[0105] III. Sequence Risk Assessment

[0106] According to equation (6), the filtered sequence lag intervals are divided into three levels. The variable X in equation (3) has only four states: 0 represents no lag interval (or sequence); 1 represents a lag interval of 0-4 days; 2 represents a lag interval of 5-9 days; and 3 represents a lag interval of 10-14 days. Following the 12-class principle for single CS events and joint events mentioned above, these two sequences are further divided into 12 different cases (labeled PC01-PC12 respectively). Figure 3 shows the two types of sequences (i.e., "PM") under different stratified lag intervals. 2.5 -CS and "CS-PM" 2.5 The risk assessment comparison of "PM" is performed. For each corresponding category, the sequence "PM" is compared. 2.5 The RR of the sequence "-CS" is higher than that of the sequence "CS-PM". 2.5 Overall, the sequence "PM" 2.5 The RR generated by the "-CS" sequence showed an increasing trend from PC01 to PC12, with a significant increase, especially during the lag interval of days 10 to 14. In contrast, the "CS-PM" sequence... 2.5 The increase in RR for “” is very small. However, in DLNM, these results are not consistent for all day lags (by default, only results for lag 2 are shown). For example, for a day lag of lag 0, the sequence “CS-PM”... 2.5 The resulting RR may be higher than the sequence PM. 2.5 RR generated by "-CS".

[0107] IV. Stratified Lag Analysis

[0108] (1) Stratified lag risk analysis of events and sequences

[0109] As shown in Table 3, among the 12 categories (CS01-CS12) of CS events, CS04 has the lowest Q-AIC value, indicating that DLNM has the best fit in this case.

[12] Therefore, the hysteresis effect of its corresponding joint event and sequence (PC04) was further explored. Figure 4 shows the joint event ("PM"). 2.5 +CS") and sequence ("PM") 2.5 -CS" and "CS-PM" 2.5 A comparison of the single-day lag and cumulative lag effects of ").

[0110] For single-day lags, the following findings were made: (1) Joint events "PM 2.5The RR value of "+CS" is greater than the RR value of a single event. Furthermore, as the lag period extends from lag0 to lag14, the difference between the RR value of the joint event and the RR value of a single event gradually decreases. (2) Sequence "PM" 2.5 The RR value of "-CS" is greater than that of the sequence "CS-PM". 2.5 "and joint events" PM 2.5 The RR value of "+CS". In addition, it exhibits an inverted U-shaped trend, indicating that it initially increases and then decreases from lag0 to lag14. (3) Sequence "CS-PM 2.5 The RR value exhibits a U-shaped trend; specifically, the RR value initially decreases after lag 8, and then increases.

[0111] Regarding cumulative lag, the following findings were made: (1) with the joint event “PM 2.5 The RR value associated with "+CS" is higher than that of a single event, and the RR values ​​of both single events and joint events show a pattern of first increasing and then stabilizing. (2) The RR value associated with the sequence "PM" is higher than that of a single event. 2.5 The RR values ​​associated with "-CS" showed a continuous and significant upward trend from lag 00 to lag 014. (3) The RR values ​​associated with the sequence "CS-PM" showed a continuous and significant upward trend. 2.5 "The relevant RR values ​​show an S-shaped trend, that is, they initially increase slowly (lag00-lag04), then decrease (lag04-lag012), and finally increase slightly (lag012-lag014).

[0112] (2) Stratified analysis by age and gender

[0113] Further stratified analysis was conducted by age and gender to explore the association with joint events (“PM”). 2.5 +CS) and sequence ("PM") 2.5 -CS and "CS-PM" 2.5 The RR values ​​(PC04) associated with these events correspond to the best fit (CS04) of a single CS event DLNM.

[0114] Considering the decrease in the number of deaths per CCVD group after stratification by age and sex, DLNM results with lag 0 (lag0) rather than lag 2 (lag2) were used to assess the RR values ​​of the subgroups. This is because the DLNM output with lag 0 provides a more robust assessment of the RR values, while the output with lag 2 may inadvertently allocate some "explanation" to other lags, resulting in poorer robustness of the assessment. Figure 5The results of DLNM with lag 0 are presented, comparing the RR values ​​associated with joint events and sequences after stratification by age and sex.

[0115] For gender stratification analysis, in conjunction with the joint event "PM" 2.5 The relative response (RR) values ​​for men and women related to "+CS" were found to be similar. However, as the lag days increased, the RR value for women decreased faster than that for men. In "CS-PM"... 2.5 "In the analysis of the sequences, the RR values ​​were found to be higher in men than in women. In contrast, the CS-PM..." 2.5 "Sequence analysis showed that the RR values ​​for women were significantly higher than those for men, peaking at lag 0. Conversely, the RR values ​​for men peaked similarly at lag 8."

[0116] For age-stratified populations, the 65-year-old and older age group participated in the joint event "PM". 2.5 The relative risk (RR) for the "+CS" sequence is higher than that for the age group under 65. For individuals aged 65 and older, the risk is higher than that for the "CS-PM" sequence. 2.5 The associated RR is greater than that of PM. 2.5 -CS" sequence-related RR. In sequence "PM 2.5 In the analysis of "-CS", the RR of the population aged 65 and above peaked at a lag of 5.

[0117] V. Sensitivity Testing

[0118] To assess the sensitivity of the RR estimated by DLNM, two experiments were conducted: (I) adjusting and testing "df1" and "df2" in equation (3); (II) changing the values ​​in equation (3). Covariate testing. Specifically, for experiment (I), the range of "df1" was changed from 3 to 5, and the range of "df2" was changed from 4 to 6. For experiment (II), PM... 2.5 Replace with PM 10 NO2 and CO were added as covariates. The Spearman correlation coefficients among these covariates all met the criterion of less than 0.7. In the above settings, the RRs estimated by DLNM fluctuated within a small range (±0.016) (compared to those involving only a single CS event and the joint event "PM"). 2.5 (Comparing +CS with DLNM) shows that the output results are robust.

[0119] VI. Conclusions and Countermeasures

[0120] (1) PM2.5 levels in a certain city during winter2.5 The frequent occurrence of "+CS" combined events may be a significant factor contributing to the increased CCVD mortality. This is in contrast to individual PM events. 2.5 Compared to the CS incident, the joint incident "PM" 2.5 The estimated relative risk (RR) for the combined event "+CS" is significantly increased, and the overall RR of the combined event may even exceed the sum of the RRs of the individual events. This is consistent with conclusions drawn from existing literature, such as the possibility that hemodynamic stress induced by CS events may be associated with increased PM. 2.5 RR of event-related inflammation. PM 2.5 The event may increase susceptibility to CS-triggered arrhythmias.

[13]

[14] .

[0121] (2) PM 2.5 Post-incident CCVD may be exacerbated by subsequent CS events, resulting in elevated RR values ​​associated with inflammation or vascular damage. (Sequence "PM") 2.5 The RR value associated with "-CS" is higher than that associated with "CS-PM". 2.5 The sequence, as shown in Figure 4, exhibits an inverted U-shaped RR (Responsive Rate) variation. The RR value initially increases and then decreases with increasing lag time, peaking at 6 days. A potential explanation is that early air pollution events induce chronic inflammatory responses and endothelial dysfunction. This pathological change subsequently reduces the body's ability to adapt to subsequent chronic inflammatory events (CS) that occur several days later.

[15] Furthermore, with the CS incident and PM 2.5 The lag between events has increased, previously caused by PM 2.5 The symptoms triggered by the event may improve (related RR values ​​gradually decrease). Despite the sequence "CS-PM" 2.5 "It may also lead to an increase in RRs, but the overall effect is relatively small. This is related to early physiological adaptations or behavioral adjustments observed in previous studies."

[16]

[17]

[18] .

[0122] (3) Gender stratification analysis shows that: (I) For the joint event "PM 2.5 +CS”, the RR risk is similar for men and women. (II) For sequence “PM” 2.5 The gender difference in "-CS" is mainly reflected in the time window of risk manifestation. Women have the highest risk at a lag of 0, while men reach their peak risk at a lag of 8, indicating that women are more sensitive to short-term PM2.5. 2.5 The physiological response to the event is usually more rapid. Although the timing of peak RR differs, the peak values ​​are not significantly different.

[19]

[20] (III) For the sequence "CS-PM 2.5Because men have a poorer ability to induce cold vasodilation, their risk of death from cholestatic vascular disease (CCVD) may be significantly higher than that of women (who have almost no risk of CCVD). Existing research also indicates that individuals with poor cold-induced vasodilation experience a more significant increase in blood pressure and peripheral vasoconstriction after cold exposure, thereby increasing the risk of myocardial ischemia and thrombosis.

[21]

[22] .

[0123] (4) Regarding age structure, older adults aged 65 and above showed higher levels of CS and PM. 2.5 The synergistic effects of the event exhibit greater sensitivity and vulnerability—primarily due to factors such as immune aging, impaired endothelial repair capacity, and cardiovascular homeostasis dysregulation.

[23]

[24] Furthermore, other epidemiological studies have shown that older adults exposed to multiple environmental stressors have a significantly increased risk of cardiovascular events.

[25]

[26] This finding underscores the need to develop targeted protection strategies for this high-risk group.

[0124] References:

[0125] [1] Ding, R., Ren, X., Sun, Q., Yang, M., Wang, Y., Sun, Z. and Duan, J., 2025. Air pollution and stroke: an emerging challenge from cardio‐cerebrovascular multimorbidity. Journal of the American Heart Association, 14(13), p.e041848.

[0126] [2] Jin, Jie-Qi, Guo-Zhen Lin, Shuang-Ying Wu, Mu-Rui Zheng, Hui Liu,Xiang-Yi Liu, Min-Qian Yan, Zhao-Yue Chen, and Chun-Quan Ou. "Short-termeffects of individual exposure to PM2. 5 on hospital admissions formyocardial infarction and stroke: a population-based case-crossover study inGuangzhou, China." Environmental Science and Pollution Research 30, no. 32(2023): 78802-78810.

[0127] [3] Zhao, Y., Kang, X., Tian, X., Liu, L., Zhao, Z., Luo, L., Tao,L., Liu, X., Wang, X., Guo, X. and Xia, J., 2023. Long-term exposure to airpollution and cerebrovascular disease: findings from Beijing HealthManagement Cohort study. Frontiers of Environmental Science & Engineering, 17(7), p.84.

[0128] [4] Alexeeff, S.E., Liao, N.S., Liu, X., Van Den Eeden, S.K. andSidney, S., 2021. Long‐term PM2. 5 exposure and risks of ischemic heartdisease and stroke events: review and meta‐analysis. Journal of the AmericanHeart Association, 10(1), p.e016890.

[0129] [5] Ni, W., Areal, A.T., Lechner, K., Breitner, S., Zhang, S.,Woeckel, M., Slesinski, S.C., Nikolaou, N., Dallavalle, M., Schikowski, T.and Schneider, A., 2025. Low and high air temperature and cardiovascularrisk. Atherosclerosis, 406, p.119238.

[0130] [6] Chen, Z., Liu, P., Xia, X., Cao, C., Ding, Z. and Li, X., 2024.Low ambient temperature exposure increases the risk of ischemic stroke bypromoting platelet activation. Science of The Total Environment, 912,p.169235.

[0131] [7] Fan, J.F., Xiao, Y.C., Feng, Y.F., Niu, L.Y., Tan, X., Sun, J.C.,Leng, Y.Q., Li, W.Y., Wang, W.Z. and Wang, Y.K., 2023. A systematic reviewand meta-analysis of cold exposure and cardiovascular disease outcomes.Frontiers in Cardiovascular Medicine, 10, p.1084611.

[0132] [8] Qi, X., Wang, Z., Xia, X., Xue, J., Gu, Y., Han, S., Yao, Q.,Cai, Z., Wang, X., Wang, L. and Leng, S.X., 2021. The effects of heatwavesand cold spells on patients admitted with acute ischemic stroke. Annals ofTranslational Medicine, 9(4), p.309.

[0133] [9] Vaičiulis, V., Jaakkola, J.J., Radišauskas, R., Tamošiūnas, A.,Lukšienė, D. and Ryti, N.R., 2023. Risk of ischemic and hemorrhagic strokein relation to cold spells in four seasons. BMC Public Health, 23(1), p.554.

[0134]

[10] Gasparrini, A., Armstrong, B. and Kenward, M.G., 2010.Distributed lag non‐linear models. Statistics in medicine, 29(21), pp.2224-2234.

[0135]

[11] Dai, M., Chen, S., Huang, S., Hu, J., Jingesi, M., Chen, Z., Su,Y., Yan, W., Ji, J., Fang, D. and Yin, P., 2023. Increased emergency casesfor out-of-hospital cardiac arrest due to cold spells in Shenzhen, China.Environmental Science and Pollution Research, 30(1), pp.1774-1784.

[0136]

[12] Chen, J., Yang, J., Zhou, M., Yin, P., Wang, B., Liu, J., Chen,Z., Song, X., Ou, C.Q. and Liu, Q., 2019. Cold spell and mortality in 31Chinese capital cities: definitions, vulnerability and implications.Environment international, 128, pp.271-278.

[0137]

[13] Wang, F., Liang, Q., Sun, M., Ma, Y., Lin, L., Li, T., Duan, J.and Sun, Z., 2020. The relationship between exposure to PM2. 5 and heart ratevariability in older adults: A systematic review and meta-analysis.Chemosphere, 261, p.127635.

[0138]

[14] Xing, X., Liu, F., Yang, X., Liu, Q., Wang, X., Lin, Z., Huang,K., Cao, J., Li, J., Fan, M. and Chen, X., 2021. Declines in heart ratevariability associated with short-term PM2. 5 exposure were modified by bloodpressure control and treatment: A multi-city panel study in China.Environmental Pollution, 287, p.117572.

[0139]

[15] Wang, T., Chiang, E.T., Moreno-Vinasco, L., Lang, G.D.,Pendyala, S., Samet, J.M., Geyh, A.S., Breysse, P.N., Chillrud, S.N.,Natarajan, V. and Garcia, J.G., 2010. Particulate matter disrupts human lungendothelial barrier integrity via ROS-and p38 MAPK–dependent pathways.American journal of respiratory cell and molecular biology, 42(4), pp.442-449.

[0140]

[16] Chen, S., Dong, H., Li, M., Huang, L., Lin, G., Liu, Q., Wang,B. and Yang, J., 2022. Interactive effects between temperature and PM2. 5 onmortality: a study of varying coefficient distributed lag model—Guangzhou,Guangdong Province, China, 2013–2020. China CDC weekly, 4(26), p.570.

[0141]

[17] Mokoena, K.K., Ethan, C.J., Yu, Y. and Quachie, A.T., 2020.Interaction effects of air pollution and climatic factors on circulatory andrespiratory mortality in Xi’an, China between 2014 and 2016. Internationaljournal of environmental research and public health, 17(23), p.9027.

[0142]

[18] Ning, Z., He, S., Liao, X., Ma, C. and Wu, J., 2024. Cold wavesand fine particulate matter in high-altitude Chinese cities: assessing theirinteractive impact on outpatient visits for respiratory disease. BMC PublicHealth, 24(1), p.1377.

[0143]

[19] Kraus, U., Horstmann, S., Dandolo, L., Bolte, G., Peters, A.,Schneider, A. and INGER study group, 2025. Sex / gender in the associationbetween ambient air pollution and cardiovascular mortality: systematic reviewand meta-analysis. Ecotoxicology and Environmental Safety, 300, p.118443.

[0144]

[20] Liu, G., Sun, B., Yu, L., Chen, J., Han, B., Li, Y. and Chen,J., 2020. The gender-based differences in vulnerability to ambient airpollution and cerebrovascular disease mortality: evidences based on 26781deaths. Global heart, 15(1), p.46.

[0145]

[21] Tsoutsoubi, L., Ioannou, L.G., Mantzios, K., Ziaka, S., Nybo, L.and Flouris, A.D., 2022. Cardiovascular stress and characteristics of cold-induced vasodilation in women and men during cold-water immersion: arandomized control study. Biology, 11(7), p.1054.

[0146]

[22] Jingesi, M., Yin, Z., Huang, S., Liu, N., Ji, J., Lv, Z., Wang,P., Peng, J., Cheng, J. and Yin, P., 2024. Cardiovascular morbidity riskattributable to thermal stress: analysis of emergency ambulance dispatch datafrom Shenzhen, China. BMC Public Health, 24(1), p.2861.

[0147]

[23] Chen, S., Dong, H., Li, M., Huang, L., Lin, G., Liu, Q., Wang,B. and Yang, J., 2022. Interactive effects between temperature and PM2. 5 onmortality: a study of varying coefficient distributed lag model—Guangzhou,Guangdong Province, China, 2013–2020. China CDC weekly, 4(26), p.570.

[0148]

[24] Chanda, F., Lin, KX, Chaurembo, AI, Huang, JY, Zhang,HJ, Deng, WH, Xu, YJ, Li, Y., Fu, LD, Cui, HD and Shu, C., 2024.PM2. 5-mediated cardiovascular disease in aging: Cardiometabolic risks, molecular mechanisms and potential interventions. Science of the TotalEnvironment, 954, p.176255.

[0149]

[25] Wu, C., He, G., Wu, W., Meng, R., Zhou, C., Bai, G., Yu, M., Gong, W., Huang, B., Xiao, Y. and Hu, J., 2024. Ambient PM2. 5 andcardiopulmonary mortality in the oldest-old people in China: A national time-stratified case-crossover study. Med, 5(1), pp.62-72.

[0150]

[26] Zhang, Z., An, R., Guo, H. and Yang, X., 2025. Effects of PM2. 5exposure and air temperature on risk of cardiovascular disease: evidence from a prospective cohort study. Frontiers in Public Health, 12, p.1487034.

[0151] The above are preferred embodiments of the present invention. Any changes made to the technical solution of the present invention that do not exceed the scope of the technical solution of the present invention shall fall within the protection scope of the present invention.

Claims

1. A stratified lag method for event sequences used to assess the risk of death from cardiovascular and cerebrovascular diseases under combined environmental exposures, characterized in that, include: S1. Acquire and organize time series data for the target region and target time period, including the daily number of deaths from cardiovascular and cerebrovascular diseases, as well as the daily measurements of meteorological and environmental exposure variables. S2. Set event thresholds for each meteorological or environmental exposure variable to be evaluated, and identify single events that meet the threshold conditions and joint events that occur simultaneously within the same time unit based on time series data. S3. Based on the identified events, extract from the time series data a sequence consisting of two or more events occurring in a predetermined order, with the time interval between occurrences within a preset maximum lag day. S4. Apply the predetermined sequence filtering rules to the sequences extracted in step S3 to filter out the closest related source events for each target event in order to determine the effective sequences for modeling. S5. Construct a distributed lag nonlinear model with a hierarchical lag structure for the fusion event sequence. The distributed lag nonlinear model uses the single event, joint event, or effective sequence determined in step S2 as the core independent variable, the daily number of deaths from cardiovascular and cerebrovascular diseases as the dependent variable, and controls for long-term trends, seasonality, and confounding variables. S6. Using the model constructed in step S5, calculate and output the risk ratio and its statistics for cardiovascular and cerebrovascular disease mortality under different single events, combined events, or event sequences of exposure.

2. The event sequence stratified lag method for assessing the risk of death from cardiovascular and cerebrovascular diseases under combined environmental exposures, as described in claim 1, is characterized in that... In step S2, the exposure variable X of the joint event t Defined by the following formula: Among them, E i,t ∈{0,1} represents event E i Whether it occurs on day t, where n is the number of event types considered; X t The value corresponds to 2 n A possible combination of events.

3. The event sequence stratified lag method for assessing the risk of death from cardiovascular and cerebrovascular diseases under combined environmental exposures, as described in claim 1, is characterized in that... In step S4, the sequence filtering rule is: for a type B event E that occurs at a predetermined time... Bj When multiple linkable Class A events exist within the maximum lag period, only those related to E are retained. Bj A sequence of A-type events that are closest in time.

4. The event sequence stratified lag method for assessing the risk of death from cardiovascular and cerebrovascular diseases under combined environmental exposures according to claim 1, characterized in that, In step S5, the distributed lag nonlinear model is a generalized linear model based on the quasi-Poisson distribution, and its basic form is as follows: in, This represents the expected number of deaths on day t. The threshold is the intercept, and lag is the lag period length, representing the duration of the exposure's effect on the outcome. describe Cross basis matrix of nonlinear delay effect during lag period This represents a natural spline function applied to potential confounding factors. This represents the term used to control for residual seasonality and long-term time trends. Covariates are other variables that need to be controlled at time t. These are the degrees of freedom of the natural spline function. The degrees of freedom of the spline function representing the covariates are used to characterize the nonlinear relationship between the covariates and the outcome. The degrees of freedom of a time spline function are used to control for long-term time trends and seasonal variations. This represents the matching layer in a case-crossover design.

5. The event sequence stratified lag method for assessing the risk of cardiovascular and cerebrovascular disease mortality under combined environmental exposures according to claim 4, characterized in that, In step S5, when the core independent variable is an event sequence, the variable is exposed. The value of is determined by the hysteresis state of the sequence, and is defined as follows: Where m is a positive integer representing the specific stratum to which the lag time interval between two events in the sequence belongs.

6. The event sequence stratified lag method for assessing the risk of cardiovascular and cerebrovascular disease mortality under combined environmental exposures according to claim 1, characterized in that, In step S3, the maximum lag days Lmax is determined based on the model fit index and prior knowledge.

7. The event sequence stratified lag method for assessing the risk of death from cardiovascular and cerebrovascular diseases under combined environmental exposures according to claim 1, characterized in that, In step S5, the construction of the distributed lag nonlinear model also includes adjusting the degrees of freedom of the exposure and lag dimensions in the cross basis function, the degrees of freedom of the spline function, and the maximum lag days based on the model diagnostic results.

8. The event sequence stratified lag method for assessing the risk of cardiovascular and cerebrovascular disease mortality under combined environmental exposures according to claim 1, characterized in that, In step S5, the distributed lag nonlinear model supports stratified analysis of the risk ratio, with stratification dimensions including at least one of age and gender.

9. The event sequence stratified lag method for assessing the risk of death from cardiovascular and cerebrovascular diseases under combined environmental exposures according to claim 1, characterized in that, In step S2, meteorological and environmental exposure variables include ambient temperature and fine particulate matter (PM2.5). 2.5 Concentration of inhalable particulate matter PM 10 At least two of the following: concentration of sulfur dioxide (SO2), concentration of nitrogen dioxide (NO2), concentration of ozone (O3), concentration of carbon monoxide (CO), and relative humidity.

10. The event sequence stratified lag method for assessing the risk of cardiovascular and cerebrovascular disease mortality under combined environmental exposures according to claim 1, characterized in that, In step S6, the output risk ratio includes at least one of the following: daily lagged risk ratio, cumulative lagged risk ratio, and risk ratio stratified by different age or gender.

Images