Effects of ambient temperature and fall-related injuries in Ma’anshan, Anhui Province, China: a distributed lag nonlinear analysis

Despite the significant economic cost of falls and injuries to individuals and communities, little is known about the impact of meteorological factors on the incidence of fall-related injuries (FRIs). Therefore, a time-series study was conducted to explore the effects of meteorological factors on FRIs in Ma’anshan City, East China. Injury data from 2011 to 2017 were collected from the National Injury Monitoring Station in Ma’anshan City. A distributed lag nonlinear model was used in this study to evaluate the correlation between ambient temperature and fall injuries. The results showed a significant exposure-response relationship between temperature and FRIs in Ma’anshan City. The high temperatures increased the risk of FRIs (RR = 1.110; 95% CI, 1.005–1.225; lag 0). The lag effect appeared at lag 10 (RR = 1.032; 95% CI, 1.003–1.063), and then gradually remained stable after lag 25 (RR = 1.077; 95% CI, 1.045–1.110). The effect of ambient temperature varied with age and gender. The lag effect of high temperature appeared in the male group after lag 15 (RR = 1.042; 95% CI, 1.006–1.079). In contrast, the effect of the female group appeared for the first time at lag 0 (RR = 1.187; 95% CI, 1.042–1.352). And the ≥ 60 years subgroup seemed to be more sensitive in low temperature (RR = 1.017; 95% CI, 1.004–1.031; lag 0; RR = 1.003; 95% CI, 1.000–1.007; lag 25). The cumulative result is similar to the single-day effect. From the results, this study would help the establishment of fall-related injury prediction and provide evidence for the formulation and implementation of preventive strategies and measures in the future.


Introduction
The 2017 global burden of disease (GBD) study showed that disability-adjusted life-year (DALY) of fall-related injuries (FRIs) was the first among all unintentional injuries in the world (GBD 2017 DALYs andHALE Collaborators 2018). While falls and injuries cause serious health consequences, they also increase the burden on the emergency department Mingming Liang and Xiuxiu Ding contributed equally to this work.  (Murray et al. 2011;World Health Organization 2018) and the potential long-term sequelae (Davis et al. 2010;Heinrich et al. 2010). FRIs represent a major health and safety concern for people of all ages . A study by Verma et al. found that middle-aged and young adults living in the community accounted for 35.3% and 32.3% of all fall-related injuries in the population (Verma et al. 2016). There are many studies on risk factors leading to fall injuries (Nahit et al. 1998;Peeters et al. 2009), but fewer studies emphasize the effects of meteorological factors or air pollutants. Turner et al. investigated people over the age of 70 and found that lower mean temperatures were significantly associated with higher fall-related hip fracture hospitalizations (Turner et al. 2011). Hussain et al. reported that heat waves can greatly increase the risk of unintentional falls in children (Hussain et al. 2007). Gevitz et al. showed that rainfall was associated with an increase in the number of emergency department visits for fallrelated fractures, but the temperature effect results were not statistically significant (Gevitz et al. 2017). Research by Guo et al. showed that when PM 2.5 increases by 10 μg/m 3 , it corresponds to an 18% increase in fall-related injuries (Guo et al. 2018). The above studies have shown that the actual effect of meteorological factors on fall-related injuries has not been fully revealed and the results are inconsistent across populations.
China has a high health burden for FRIs (World Health Organization 2018), and FRI is one of the main causes of unintentional injuries in Anhui Province (Xing et al. 2016). Previous studies have shown that for injuries or fall-related injuries, the impact of meteorological factors is timedependent (Huang et al. 2015;Liao et al. 2018) and has a lag effect (Gasparrini et al. 2012;Zhang et al. 2020). Such lag effect might be related to the indirect effect of meteorological factors on immunity and thermoregulatory capacity (Ma et al. 2016;Min et al. 2019). Using the distributed lag nonlinear model (DLNM) framework, which has a function to explore the potential lag and nonlinear effects (Gasparrini et al. 2012), a time-series study was conducted to explore the relationship between ambient temperature and the risk of FRIs in Ma'anshan City, China. The results of the study would provide a scientific basis for intentional injury prevention and intervention.

Study Setting
Ma'anshan City is one of two Chinese national injury monitoring cities in Anhui Province. It is located in eastern China, downstream of the Yangtze River (longitude 117°53′-118°5 2′ E, latitude 31°24′-32°02′ N). The Yangtze River flows from east to west in the region. It has a northern subtropical humid monsoon climate with warm and humid seasons.

Injury data
Injury cases collected from January 1, 2011, to December 31, 2017, at the National Injury Surveillance Station in Ma'anshan City were provided by the Chinese Center for Disease Control and Prevention. The definition of fallrelated injuries is based on the National Injury Monitoring Report Card of the National Injury Surveillance Station. These injury cases are sorted into the daily data and employed to match the daily meteorological data. The data included general information on injury cases (age, sex, occupation, and education level), basic information on injury events (time, cause, location of the injury, activity at the time of injury, and whether the injury was intentional), and clinical information on the injury. This study was approved by the ethics committee from the Chinese Center for Disease Control and Prevention Institute for Environmental Health and Related Product Safety (201606).

Statistical analysis
The study explored exposure-lag-response associations through modeling. Daily FRIs are considered as small probability events, and this study used a Poisson generalized linear model with distributed lag nonlinear model to estimate the nonlinear and lag effects of ambient temperature on daily FRIs (Dai et al. 2018;Rosenberg et al. 2018).
We first investigated the correlations between FRIs and meteorological factors with Spearman's correlation analysis. Depending on the results of correlation analysis, there is a correlation between the mean temperature and the incidence of injuries (p < 0.001).
And according to the relevant analysis results, this model would incorporate meteorological factors such as atmospheric pressure (p = 0.001), snow depth (p < 0.001), and sunshine time (p = 0.002), and adjust their influence by changing the degrees of freedom (df) Ma et al. 2021).The DLNM is based on the definition of a "cross-basis," a bi-dimensional space of functions, which describes both the shape of the relationship along with the space of predictors and the lag dimension in which it occurs (Gasparrini et al. 2010). The quasi-Poisson Akaike information criteria (Q-AIC) were used to select the optimal degrees of freedom. The model used in the study to quantify the relationship between ambient temperature and FRI is as follows: þns time t; 9=year À Á þηDOW t þ νHoliday t where Y t refers to the number of FRI occurring on day t; α is the intercept of the model; TEM t , 4 represents the temperature on day t with 4 dfs; β is the matrix coefficient of the temperature and a natural cubic spline function used to estimate the lag effects of temperatures; and ns () is the smoothing function of independent variables in mgcv package in R software. PRS in the model represents the atmospheric pressure, SNOW represents the snow depth, and SUN represents the sunshine time. A natural cubic spline curve with an annual degree of freedom 9 was used to control the time trend. DOW means the day of the week, with a reference day of Friday. Public holidays during the study period are also accounted for using the categorical variable holiday due to the potential holiday effects.
According to the combination of AIC criteria and references provided by related literature, we chose 30 days as the maximum lag period in the model (Shen et al. 2021;Yi et al. 2019). In the case of a nonlinear relationship, we calculated the relative risk (RR) with a 95% confidence interval (CI) of specific temperatures on daily FRIs.
We determine the lowest risk temperature by identifying the point of the bottom of the U-shaped curve of temperature-FRI risk (Armstrong 2006), and verify our results by simulating a large number of parametric bootstrap algorithms (Tobías et al. 2017). The temperature with the lowest risk of total FRIs was the reference. RR was used to quantify the effect of temperature on the number of fall injuries, meaning the risk of FRI caused by changes in ambient temperature. Further analysis of the population of interest was performed by stratifying by age and gender groups.

Sensitivity analysis
The Spearman's correlation analysis results showed that NO 2 (p < 0.001), PM 2.5 (p < 0.001), and CO (p < 0.001) were also related to the injuries. We used this information to analyze whether they have potential confounding effects on the model. By establishing different time-series models incorporating different combinations of air pollutants (NO 2 & PM 2.5 & CO; NO 2 & PM 2.5 ; PM 2.5 & CO; NO 2 & CO; NO 2 ; PM 2.5 ; CO), it was found that the fitness of all models showed a slight decrease. Therefore, we perform sensitivity analysis based on the model with the highest goodness of fit (only PM 2.5 ). In addition, we verify that the model has obtained the best effect by changing df for time trend, atmospheric pressure, snow depth, sunshine time, and PM 2.5 .
All statistical analyses were performed using R (version 3.6) software.

Data description
As shown in Table 1, from 2011 to 2017, the Ma'anshan injury monitoring system recorded a total of 36,723 cases of fall-related injuries, with a maximum of 49 cases in a single day. Of the total cases, 21,447 were male while 15,276 were female. Of the cases, 11,533 cases were < 20 years of age; 8045 cases, 20-39 years of age; 9404 cases, 40-59 years of age; and 7741 cases, 60 years of age and older. Figure 1 shows the time-series distribution of daily injury cases in Ma'anshan from 2011 to 2017. Detailed meteorological factor information over the study period could be noted in Table 2. Table A1 lists detailed air pollutant information from 2013 to 2017. Figure 2 shows the exposure-response relationship between daily mean temperature and FRIs, and the results show a significant nonlinear correlation.

Relationship between ambient temperature and FRIs
The temperature at the lowest risk of FRIs (4.0°C) served as the reference dimension in this study to compare the singleday and cumulative lagged effects of high and low temperatures on FRIs in Ma'anshan, respectively. Detailed RRs and effect intervals are shown in Tables 3 and 4. We used the 10.0 (3.0°C) and 90.0 (27.9°C) percentile of ambient temperature in Ma'anshan City as the prescribed high and low temperatures (Min et al. 2019;Zhang et al. 2019). Figure 3 shows the dose-response relationships for FRIs at different lag days (0-30) at different mean temperatures. Table 3 shows the single-day effects of high and low temperatures on FRIs. The effect of temperature in the results showed an asymmetric U-shape. In terms of the single-day lag effect, the lag effect of high temperature appeared earlier and was more pronounced in the general population. At lag 0, high temperature significantly increased the risk of FRIs (RR = 1.110; 95% CI, 1.005-1.225). The lag effect appeared at lag 10 (RR = 1.032; 95% CI, 1.003-1.063) and increased day by day, and then gradually remained stable after lag 25 (RR = 1.077; 95% CI, 1.045-1.110). The cumulative lag effect is given in Table 4. The cumulative effect began to show a significant lag effect at lag 1 (RR = 1.175; 95% CI, 1.002-1.378), and it appeared for the second time after lag 15 (RR = 1.632; 95% CI, 1.042-2.556). In the general population, no statistically significant results of low temperature on FRI were found, whether single-day effect or cumulative effect.

Subgroup analysis
The different subgroup effects of temperatures on FRIs are shown in Fig. 4. In terms of the single-day lag effect, the lag effect of high temperature appeared in the male group after lag 15 (RR = 1.042; 95% CI, 1.006-1.079), and it increased day by day. In comparison, the female group might be more sensitive to high temperature. The lag effect of the female group appeared for the first time at lag 1 (RR = 1.105; 95% CI, 1.018-1.200) and the second time is after lag 15 (RR = 1.085; 95%, 1.041-1.132).
In the high temperature results of different age groups, the single-day lag effect of the group < 20 years old appeared at lag 15 (RR = 1.058; 95%, 1.009-1.110). The results were similar for the 20-39-and 40-59-year groups, with an immediate significant effect of high temperature on FRIs. It is worth noting that the effect of temperature on FRIs in the ≥ 60-year group was different from that of the other groups. There was a significant association and lag effect with FRIs in the ≥ 60year group at low temperature (RR = 1.017; 95% CI, 1.004-1.031; lag 0; RR = 1.003; 95% CI, 1.000-1.007; lag 25).
For the cumulative lag effect, the result is similar to the single-day lag effect. The low temperature showed a significant effect on FRIs for the group ≥ 60 years of age.

Sensitivity analysis
We conducted adjusted PM 2.5 and unadjusted PM 2.5 two timeseries analyses to compare the effects of air pollutants on the effectiveness of the model ( Figure A1). The results showed that after adjusted air pollutants, the fit of the model may decrease slightly. But the results of adjusting PM 2.5 (RR = 1.068; 95% CI, 1.008-1.132; lag 30) are basically the same as the unadjusted (RR = 1.068; 95% CI, 1.007-1.132; lag 30), except that the confidence interval of the former is slightly reduced (Table A2 and A3).
The nonlinear lag relationship between the mean temperature and FRIs after using alternative cut-off points (5th and 95th percentile) is presented in Table A4 and A5. For the high temperature (95th percentile), the results did not change significantly (RR = 1.071; 95% CI, 1.023-1.122; lag 30). However, there was a significant increase in the risk of FRIs in the total population (RR = 1.024; 95% CI, 1.010-1.038; lag 30), the female group (RR = 1.038; 95% CI, 1.019-1.056; lag  The duration of the lag effect has also become longer. This suggested that for lower temperatures, the results may be more statistically significant.

Discussion
Fall-related injury is a major public health problem (Hartholt et al. 2011). These injuries could lead to chronic pain, incapacitation, increased health care needs, and even increased risk of mortality (Tinetti et al. 1988). If the burden of related injuries will further increase, it will have a huge health, social, and economic impact on society and the country. Although previous studies have explored the risk factors for FRIs, few studies have concentrated on the effects of ambient temperature on FRIs in different populations.
In this study, we found a nonlinear effect of ambient temperature on FRIs in Ma'anshan through a time-series study, and there is also a short-term lag influence of this effect. This certainly compensates for the limitation of previous related linear model research. Compared with the reference temperature, high temperature increased the risk of FRIs, while the lag effect of low temperature on FRIs was significant for people aged 60 years or older in Ma'anshan.
The above conclusion is consistent with previous studies. Some studies have found that the incidence of fall-related injury among children increased during the high-temperature season (Morrison et al. 1999;Parslow et al. 2005). Hussain thought this is because parents of children open windows for ventilation due to the high temperature, which increased the risk of children falling from heights (Hussain et al. 2007). A study of FRIs in Guangdong, China, showed that children's falls were more common in June and September, and the weather was usually sunny when the fall occurred (78.80%) . Liu et al. (2012) considered that these times are often suitable for play or entertainment, and the caregiver's awareness of injury prevention was not enough which eventually led to the occurrence of a fall.   The FRI of the elderly is significantly different from those of other populations for temperature changes. Turner's study found an increased incidence of hip fractures in people over 75 years of age when the temperature decreased (Turner et al. 2011). Stevens's study on injuries showed that the incidence of fatal falls in older adults was 9.1% higher in colder climates (Stevens et al. 2007). In this study, the risk of FRIs in the elderly was also found to be significant at low temperatures. Especially when we use alternative cut-off points and compare with the reference temperature, the low-temperature effect would be more obvious. The reasons for the falls of the elderly are also different from those of other populations. Studies have shown that FRI in the elderly in winter is related to walking on wet and slippery roads (Xu et al. 2012). Low temperature caused lower body temperature, decreased flexibility, slower reaction speed, changes in physiological processes (Keatinge et al. 1984;Riley and Cochran 1984), and wearing heavy clothes, all of which would increase the risk of falls for the elderly (Lin et al. 2015).
This study is the first to examine the epidemiological relationship between ambient temperature and FRIs in the eastern Chinese city of Ma'anshan. In our sensitivity analysis, we also included air pollutant information from 2013 to 2017 to test the impact on the goodness of model fit. Previous studies have hypothesized that air pollutants may increase the risk of fallrelated injuries through indirect ways such as chronic systemic oxidative stress and affecting nervous system health (Guo et al. 2018). Of course, further population experiments are needed to verify these conclusions.
For the biological mechanism of ambient temperature on injury, previous studies have also provided some other hypotheses. Studies on occupational injuries have shown that excessively high temperatures could lead to individual negligence, fatigue, and decline in cognitive function and athletic ability (Nindl et al. 2013;Otte im Kampe et al. 2016). For high temperatures, causing dehydration and heat cramps in individuals over the thermoregulatory system will impair the ability to work (Cohn and Rotton 1997). Ambient temperature also affects people's activity level, and a suitable temperature would increase the amount of activity and expand the scope of activity (Morency et al. 2012;Sumukadas et al. 2009), and the risk of FRIs would increase.
The lag effects of FRIs in this study might be related to the long-and short-term effects of temperature on the physical and psychological state. Anderson et al. found that the effect of cold on mortality usually lasts for a long time (Anderson and Bell 2009). Extreme temperatures may trigger a series of reactions in people who are already weak, and this process is not completed in an instant (Basu 2009;Bhaskaran et al. 2013). Studies have found that after the heat wave, the hospitalization rate of patients with underlying diseases such as asthma and heart disease will also increase significantly (Khalaj et al. 2010). The occurrence of such underlying diseases will undoubtedly increase the risk of falling in related patients. Similar to the results of our study, Lee found that injury by falls significantly increased at high temperatures in South Korea (Lee et al. 2020). They also used the DLNM to explore the lag effects of injury, although they did not further differentiate the risks of different populations (Lee et al. 2020). These studies show that a few days after a certain temperature is reached, it still affects people's risk of falling.
Our results indicated that the cumulative effect of ambient temperature should cause greater concern. But several limitations also need to be acknowledged. First of all, this timeseries analysis directly uses the data of the injury monitoring site, which may not be able to avoid the ecological fallacy of the research in terms of methodology. It is difficult to adjust such as age, drug use, and physical condition as confounding factors. But according to the published papers, we believe that as a long-term ecological study, these confounding factors will not significantly change our conclusions.
Second, since our temperature is estimated based on fixed monitoring points, there may be exposure measurement errors. In addition, this study cannot distinguish where the injury occurred. Some fall-related injuries may occur at home or indoors, and their direct temperature exposure may be different from the ambient temperature exposure. If this exposure measurement error occurs in this study, it may lead to bias Fig. 2 Three-dimension plot for relative risk (RR) of ambient temperature on FRIs in Ma'anshan, China, 2011 against the null hypothesis, thereby underestimating the true association. These fallacies may be difficult to completely eliminate. The exact relationship between the injury and exposure may require further investigations.
Third, the DLNM framework used in this study may not be able to fit high-dimensional complex variables well, which is another limitation of the study. However, it should be pointed out that the DLNM is a generally recognized modeling tool for describing potential nonlinearities and delay dependence when solving the risks of meteorological factors. More flexible approaches based on distributed lag linear or nonlinear models provide noteworthy advantages, in particular when complex lagged associations are assumed (Gasparrini 2016). The application in various diseases or injuries proves the reliability of this model.
Although the research and related work of fall prevention have gradually increased in recent years, there are still few reports on the effects of meteorological factors on fall-related injuries. In terms of injury prevention, internationally, it is emphasized to be as specific as possible when formulating relevant injury prevention policies, strategies, and action plans (Linnan 2007). Preventing fall-related injuries requires the joint efforts of individuals, families, and society to improve public health knowledge through health education, change undesirable high-risk behaviors, reduce and eliminate environmental risk factors, and effectively prevent and reduce the incidence of falls and the severity of the injury (Murray et al. 2016). At present, the meteorological warning has received considerable attention in public health planning and preventive health care (Casanueva et al. 2019). For example, the UK has developed a monitoring system that can monitor the impact of severe winter weather on the demand for emergency departments in near real time (Hughes et al. 2014). In Canada, when freezing rain is expected to be ≥2 h or is expected to cause harm to transportation or property, Environment Canada will issue an official freezing rain warning (Yan et al. 2020). These warnings were issued during and before severe weather. Research results indicated that these weather warnings can be used to promptly initiate injury prevention and health promotion strategies to reduce the incidence of fall-related injuries (Mondor et al. 2015). Public health agencies should consider using this early warning information to initiate FRI prevention strategies and health resource allocation before severe weather. In addition, Fig. 3 Lag effects of ambient temperature at different lag days on FRIs in Ma'anshan, China, 2011 choosing better non-slip shoes, going out under appropriate weather conditions, increasing sun exposure or dietary supplements which can maintain adequate vitamin D, and moderate exercise can increase the body's strength and balance, which can prevent falls to a certain extent (Wong et al. 2020).

Conclusion
This study showed that there is a lag and nonlinear relationship between temperatures and FRIs. Different age groups have different risk outcomes. The results of the study will help Fig. 4 The cumulative effects of ambient temperature on FRIs in different gender and age groups establish an early warning system for injury and reduce the burden of FRIs, especially among high-risk groups.