Introduction

Suicide is a public health problem, causing over 800,000 deaths worldwide every year (^{World Health Organization [WHO], 2019}). In Spain, a total of 3,539 suicide deaths were registered in 2018, equivalent to a suicide rate of 7.9 per 100,000 inhabitants (^{Instituto Nacional de Estadística [INE], 2019}). However, these figures only consider actual suicide deaths and do not include suicide attempts. It is estimated that for every suicide death, there are 20 previous attempts (^{WHO, 2014}), which are the most predictive risk factor for suicidal behavior (^{Parra-Uribe et al., 2017}).

Since most people who attempt suicide go to out-of-hospital Emergency Services (OES), it is essential to study the classification and correct registration of suicide attempts in the latter. However, scientific literature, although scant, has focused on studying the validity of triage systems that determine the level of priority of Hospital Emergency Service (HES) requests (^{Feldman et al., 2006}; ^{Grosgurin et al., 2019}; ^{Parenti, Bacchi Reggiani, Iannone, Percudani, & Dowding, 2014}; ^{Williamson, Gochman, Bullaro, Kaufman, & Krief, 2018}). Although their use can be extrapolated to OES (^{Gräff et al., 2014}), there has been less research in this respect, especially as regards mental problems (^{Shah et al., 2009}) and particularly suicide.

In Andalusia, the largest autonomous community in Spain, various research projects have focused on the analysis of requests related to suicidal behavior (^{Blanco-Sánchez et al., 2018}; ^{Mejías-Martín et al., 2018}; ²⁰¹⁹) and observed that the highest rate occurs in the province of Málaga. Studies conducted in this province have found differential characteristics in requests related to suicidal behavior with respect to other OES requests (^{Guzmán-Parra et al., 2016}; ^{Jiménez-Hernández et al., 2017}; ^{Moreno-Küstner et al., 2019}). However, results obtained in previous studies reveal differences in the classification of suicide attempts (^{Blanco-Sánchez et al., 2018}; ^{Guzmán-Parra et al., 2016}; ^{Jiménez-Hernández et al., 2017}; ^{Mejías-Martín et al., 2018}; ²⁰¹⁹; ^{Moreno-Küstner et al., 2019}). Some studies classify a request as involving suicidal behavior when the operator answering the call and the doctor attending the patient consider it as such (^{Mejías-Martín et al., 2018}; ²⁰¹⁹; ^{Moreno-Küstner et al., 2019}). However, others have regarded requests related to suicidal behavior by using merely the classification made by the operator (^{Guzmán-Parra et al., 2016}; ^{Jiménez-Hernández et al., 2017}). Furthermore, ^{Blanco-Sánchez et al. (2018)} observed enormous difficulty on the part of health professionals in classifying suicide attempts due to the implications this entails, since it is one of the behaviors with the lowest registration rate in health reports (^{Miret et al., 2010}).

These circumstances justify the need for a valid classification system of OES requests. For this reason, and due to the importance of having a useful classification to improve care provision, the general objective of this study is to validate the classification of the Emergency Coordinating Center (ECC) of Málaga (Spain), based on its sensitivity, specificity, and positive (PPV) and negative (NPV) predictive values for the detection of requests related to suicidal behavior.

Method

Sample

The database analyzed comprised a total of 589,013 requests. To improve data quality, 113,180 requests were excluded since they were not health care requests, together with 69,575 claims that were not classified in the main categories. Furthermore, 16,452 duplicate requests were identified and eliminated. A total of 389,806 valid health claims were obtained, of which 277,207 were excluded because information on the clinical assessment by the health professional was not recorded in the database. Lastly, the sample consisted of 112,599 valid health requests that had been attended by a health professional at the scene of the event, for which the clinical assessment had been recorded (Figure 1).    

Figure 1 Flow chart of the selection process of the study sample 2013-2014. 

Calculation of the sample size to obtain the necessary statistical power to generalize our results was based on the infinite population formula (^{Morales-Vallejo, 2012}). Accordingly, for the sample to be representative in our case, 1,067 requests would be required.

Results

Of the 112,599 valid health service requests, the mean age was 66.27 (SD = 21.18) while 52.4% (n = 58,976) of the requests were made by women. The total number of requests were divided into two groups: that of the ECC, when the patient is attended by telephone and that of the healthcare professional who attends in situ. The first group classified 1,390 requests (1.2%) as suicidal behavior; while the second classified 1,456 (1.3% as this type of behavior).

Table 1 shows the distribution of requests related to suicidal behavior according to the clinical assessment of the health professional who attends the patient at the scene of the event. The codes referring to mental disorders and poisoning are the most common diagnoses (40.6% and 24.2%, respectively), as opposed to the actual suicide codes, which account for 12% of the distribution.    

Table 2 shows the results of the calculation of sensitivity, specificity, PPV, and NPV. The sensitivity of the classification system of the coordinating center to correctly identify requests related to suicidal behaviors, according to the reference pattern, was 44.78%, while the specificity of the classification system to detect the absence of suicidal behavior, according to the clinical assessment, was 99.34%. As for the PPV of requests related to suicidal behavior, a value of 46.91% was obtained and a value of 99.28% was obtained for the NPV (Table 3).    

As can be seen from Table 2, 738 requests (.7% of the total) were regarded as false positives. In other words, the ECC classified them as suicide, but the healthcare professional who attended the user reported a different diagnosis. Eight hundred and four cases were considered false negatives (.7% of the total). These are requests that the coordinating center did not classify as suicide, whereas the health professional assigned them codes which meant they could be considered as suicidal behavior.

Discussion and conclusion

The purpose of this study is to validate the classification system of a coordinating center for urgent requests in comparison with the categorization made by doctors attending these requests in situ. The results obtained indicate that the ECC classification has a lower capacity to identify requests involving suicidal behavior than it has to detect the absence of suicidal behavior in the requests received. The strengths of this study include having a large set of data obtained from daily clinical practice and being able to compare the classification of requests at two different moments in the OES care process.

The results of the analysis of the sensitivity and specificity of the classification system used by the ECC have shown that it experiences difficulty in detecting demands for suicidal behavior. In this regard, ^{Mejías-Martín et al. (2018)} in their study on OES requests involving suicidal behavior have highlighted the lack of agreement between the demands classified by the ECC operator as suicide and the clinical assessment made by the healthcare team attending the patient. Their result showed that, out of the total number of requests identified as suicide by the operator, only 7.7% are confirmed as such by the doctor treating the patient. These findings can be explained by factors such as the lack of a specific protocol in the case of suicidal behavior to guide action (^{Mejías-Martín et al., 2018}), the absence of a uniform classification system to categorize suicidal behavior (^{Silverman, Pirkis, Pearson, & Sherrill, 2014}), and the difficulty in determining, for the professional who answers the phone, the intentionality of a suicide attempt when the circumstances are unclear or the patient cannot or does not wish to clarify matters (^{Blanco-Sánchez et al., 2018}; ^{Mejías-Martín, García-Caro, Schmidt, Quero, & Gorlat, 2011}; ^{Mejías-Martín et al., 2018}). These factors, together with the great social stigma attached to suicide, suggest that they may be interfering in the accurate classification of requests as involving suicidal behavior.

As for the lack of agreement in the classification of requests between the operator and the doctor attending the patient in situ, we found, on the one hand, that requests regarded as false positives were mainly classified as a mental illness, according to our standard of reference. The presence of mental disorders is a risk factor, but it is hardly sufficient factor (^{Franklin et al., 2017}) for suicidal behavior, and the causality of suicides due to a mental disorder has been widely disseminated (^{García-Haro et al., 2020}), it seems logical to think that an ECC operator would classify a request as involving suicidal behavior when the patient verbalizes a related ideation or behavior. However, it is the onsite physician who can explore and undertake a more thorough evaluation of the patient’s condition and definitively considering the demand as a mental health problem.

Furthermore, requests considered false negatives were mostly classified as poisonings/allergies by the ECC. It is to be expected, given the information provided by the person requesting emergency health services that the operator will not have sufficient evidence to be able to consider the request as one involving suicidal behavior. Moreover, among the false negatives, requests classified by the ECC as trauma were found, which include falls and stab/fire wounds. This group of patients can be considered the most worrying as regards their treatment. Because ECCs classify them by severity, an inaccurate classification could lead to a delay in care-delivery.

In their qualitative study, ^{Blanco-Sánchez et al. (2018)} found that professionals who attend requests for emergency health services believe that there are no specific protocols for cases of suicide attempts. Requests concerning drug intake, for example, would be dealt with from a medical point of view regardless of whether it involves suicidal behavior. In this same study, since no objective elements were found to identify these behaviors as cries for attention or actual suicide attempts, the criteria of family members were often used to classify them as one or the other. In addition, ECC professionals consider that the health professionals who attend the patient should be responsible for classifying the request rather than them. Likewise, healthcare professionals consider that suicide attempts should be assessed in hospital services rather than the ECC. In this regard, the study by ^{Moreno-Küstner and González Sánchez (2020)} points to the underreporting of cases of suicide deaths in the Death Certificates with Judicial Intervention that provide information for the INE suicide statistics. Their results indicate that 73% of the cases of death were confirmed by family members as the suicide of the deceased person but were not identified as such in the Statistical Bulletin of Deaths with Judicial Intervention. This shows that the classification of suicidal behavior is an arduous, complicated task, and that proper registration by the various official bodies is required to obtain accurate suicide figures.

However, our study is not without limitations and its results should therefore be interpreted with caution. First, some requests were unable to be analyzed due to incomplete records. This underlines the fact that when working with the information extracted from daily clinical practice, intrinsic errors may appear during the data collection stage. Second, it was not possible to take patient variables (such as personal history and previous attempts) into account to be able to accurately classify an episode or call as a suicide attempt. Likewise, aspects such as gender and age could be potential sources of bias since the sample analyzed has a higher representation of older people and women, the groups that most request OES. Finally, in the future, we could encounter changes in the classifications used to date to record suicidal behavior. However, we consider that this limitation is always present since the methods and instruments used in research undergo changes and modifications to enhance their use.

This study has significant practical implications, such as, for example, serving to orient the training of OES professionals in the importance of conducting a more exhaustive classification of requests in general and those involving suicidal behavior in particular. In addition, better classification will allow for more effective, efficient interventions in suicide cases while at the same time facilitating the identification of cases regarded as false negatives, thereby reducing the unnecessary use of health resources.

In conclusion, we consider that it is necessary to have a valid classification system to identify requests involving suicidal behavior for pre-hospital emergency services, and to conduct periodic studies on the validity of this classification to improve care delivery. The use of categories or questions that are valid for identifying requests involving suicidal behavior will allow for better detection of this problem and help develop more suitable suicide prevention plans for this at-risk population.