Incidence of fall-from-height injuries and predictive factors for severity

Methods

AI predictive model

A multilayer perceptron neural network (MPNN) was utilized to identify predictive factors and to stratify the importance of these factors associated with mortality (Figure 1). The cause-and-effect relationships were considered in selecting predictors. The dependent variable was in-hospital mortality. Predicting factors included sex, age, comorbidity, injury location, fall types (border fall vs. domestic fall), and other variables that had a significant correlation with injury severity.

Figure 1:

A multilayer perceptron neural network (MPNN) machine learning (ML) method for the prediction of factors contributing to in-hospital mortality.

A 70:30 training-testing was utilized to split the dataset partition. The MPNN analysis structure included an architecture with one hidden layer of a maximum of 50 units utilizing an automatic architecture selection function. Each hidden unit is a function of the weighted sum of the inputs. An automatic batch method was selected for the training type, with a scaled conjugate gradient utilized for the optimization algorithm. The initial learning rate was 0.4. The activation function in the hidden layer was the hyperbolic tangent. In the output layer, the dependent variable was the in-hospital mortality, and the activation function was Softmax. The loss/error function was cross-entropy. The importance of predictive variables was calculated based on a sensitivity analysis, which computed the importance of each predictor in the neural network (Figure 1) [24]. MPNN analysis was performed utilizing SPSS software (Version 28, IBM).

Results

AI predictive model and ML analysis outcomes

MPNN analysis ranked the strongest predictors for in-hospital mortality to be chest injury, then head injury, followed by GCS on admission, extremity injury, and abdomen injury, followed by other predictors that were not as strong (Figure 2).

Figure 2:

Normalized importance of independent factors related to in-hospital mortality is stratified showing the strongest predictive factors to be chest injury and head injury followed by GCS on admission.

Discussion

Falls have been consistently ranked as the second main cause of unintentional injury deaths worldwide. A fall is defined as “an event which results in a person coming to rest inadvertently on the ground or floor or other lower level” by the World Health Organization (WHO) (2021) [25]. A fall can happen from an elevated height or on flat ground, often leading to potential injury.

The injuries caused by an FFH persist to be significant adverse events and a challenge regarding injury severity assessment to identify patients at high risk on admission. Understanding the incidence and the factors that predict injury severity can help in developing effective intervention strategies to save lives. Our study demonstrated that the MPNN algorithm predicted the factors with stratified importance leading to higher severity or mortality with consideration of fall height, fall mechanisms (border fence vs. domestic falls), anatomical injury location, and demographics.

In this cohort of patients sustained from FFH injuries, the border fall group showed a younger age, fewer comorbidities, and a lower ISS on admission than the domestic group. Patients of domestic falls sustained more head and chest injuries, whereas boarder fall patients sustained more leg and ankle fractures. The injury patterns of border falls are consistent with the deliberate and possibly more falling posture-controlled nature with the leg landing first, whereas the domestic falls are consistent with unintentional or accidental nature with head or trunk landing posture [14]. Although there was not a statistical difference of in-hospital mortality rates between domestic and border fall groups, ICU stay was less in border fall than domestic fall patients.

The literature has reported that undocumented immigrants crossing the US-Mexico border frequently sustained musculoskeletal injuries, predominantly lower-limb injuries including fractures from FFHs [11], [12], [13], [14], [15], [16, 26], 27]. While domestic falls have many different mechanisms with varying outcomes [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28], [29], [30]. According to the WHO, falls are the second leading cause of unintentional injury deaths worldwide. A total of 37.3 million falls occur globally each year that are severe enough to require medical attention with reported deaths of over 660,000 adults and almost 30,000 children [25], 31]. There are no previous publications to address the fall-height, falling types (border fall vs. domestic fall), anatomic injury location, and their relation to injury severity, possibly due to the complex characteristics of injuries and multiple variables in datasets leading to sophisticate analysis required upon utilizing traditional statistical analysis. The AI and ML methods are emerging and have the potential to capture underlying trends and patterns that are otherwise impossible with traditional statistics approaches [32], thus to assist in clinical diagnosis [33].

In this study, we select the MPNN, a forward artificial neural network with stacked multiple layers of perceptron to build a predictive model to understand the importance of factors associated with injury severity. MPNN is one of the most utilized neural network architectures for building predictive AI models. MPNN has been utilized in many medical and orthopedic research [24], [34], [35], [36], [37], [38], [39]. It has been utilized in clinical quality improvement projects and emergency care to reduce complications and costs [32], 40]. It produces approximate solutions for complex problems to better understand the correlation strength between multiple variables [40], 41], whereas traditional bivariate correlation analysis only determines the correlation strength between two variables. In this study, we utilized MPNN to understand the critical mechanical factors leading to in-hospital mortality considering demographic information, comorbidities, fall height, fall location, injury location, and other clinical measurement. MPNN analysis ranked the strongest predictors for in-hospital mortality as chest injury and head injury, followed by GCS on admission, as predictors, suggesting that ICU care was required, as shown in our clinical interventions.

The predictive AI model may assist and improve the diagnosis on admission and treatment protocols for this cohort of FFH patients occurred at the border or domestic sites. In domestic falls, severe injuries to the head and chest were more common than the other anatomic locations. In border falls, the incident rate of extremity injuries was significantly higher than other anatomic locations. The tibia, fibula, and ankle injury rate was significantly higher in border falls than domestic falls (Table 4). It has been believed that border falls are intentional and more prepared with feet landing on the ground [14], whereas domestic falls are typically accidental and less prepared for controlled falling posture [21], 28], 29]. A theory has thus been proposed that the border falls from fence, wall, or bridge yield an axial load and vertical deceleration, which are forces that predominantly impact the musculoskeletal system, particularly the lower-extremity joints as evidenced by more tibia, ankle, calcaneus fracture, and joint injuries found in the border fall patients. Similar observations have been reported previously [21], 28]. Overall, we observed that border fall patients had less injury severity than domestic fall patients, as evidenced by a significantly lower ISS on admission, and fewer border fall patients were admitted to the ICU among the border fall group. Other contributing factors can be that border fall patients were younger with fewer comorbidities, leading to less injury severity.

This study demonstrated that domestic FFHs yielded a higher ISS, longer ICU stay and longer mechanical ventilation than border falls. Overall injury severity was also higher in three death cases in the hospital. These three patients who died of domestic FFHs were older than the survivors (54.7 ± 6.1 vs. 35.7 ± 13.3 years, p=0.023), and had an ISS higher than survivors (36.3 ± 2.9 vs. 9.6 ± 6.5, p=0.01), with a significantly lower GCS on admission (4.7 ± 2.9 vs., p=0.026). These three patients suffered from more multiple injuries than survivors, including brain injury (p=0.002), chest injuries (p=0.003), spine injuries (p=0.043), and fewer extremity injuries (p=0.039), suggesting that anatomic-injury locations are critical factors for mortality.

The impact of a fall depends on several factors beyond height alone, including body orientation, surface type, and the body’s biomechanics at the time of ground impact. Studies have shown that falls from as low as 10 feet (approximately 3 m) can be fatal, depending on these variables [42]. Landing head first, for instance, can be far more dangerous than landing feet first. Falling onto a hard surface (like concrete) increases the likelihood of fatal injuries, whereas softer surfaces (grass, sand) can sometimes cushion the impact. What breaks the fall or absorbs some energy can make a difference. Encountering obstacles that slow the descent can reduce injury [43], [44], [45].

The type of ground surface and surroundings of falls are critical for injury but unknown in our study, making it one of its limitations. Among the other limitations are the sample size, which is not big enough for a larger-scale big-data analysis, and the data resources, which are limited to a local hospital, leading to the results not being generalized to other states. Those situations in which patients who sustained fatal falls and did not reach the trauma center make severely injured patients unaccounted for in this study. Whether a fall was interrupted by objects, such as building materials before hitting the ground, is unknown in this cohort of patients. Our data were limited to patients’ hospital stay; no investigation has been performed after patients were discharged from the hospital.

Further research will be performed through the research of multiple medical centers. Long-term outcome follow-up studies will be performed with consideration of the outcomes from rehabilitation. At what height a fall becomes fatal under intentional vs. unintentional FFHs needs further investigation.

Conclusions

Intentional vs. unintentional FFHs yielded different injury patterns and injury severity as the injured patients from border falls had a lower ISS, and fewer head or chest injuries than domestic falls, although there were more extremity injuries in this study. MPNN analysis demonstrated that chest and head injuries with a low GCS indicated a high risk of mortality from a fall.