Attempted Suicide Is Independently Associated with Increased In-Hospital Mortality and Hospital Length of Stay among Injured Patients at Community Tertiary Hospital in Japan: A Retrospective Study with Propensity Score Matching Analysis

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper presents the results of a cohort study examining differences in outcomes following self-inflicted injury compared to injuries that were not self-inflicted. The paper addresses a topic of clinical and public health importance, with findings that can highlight the need for specific focus on providing care for people presenting with self-inflicted injuries. The comprehensive approach to addressing confounding is a major strength of this study. The manuscript is structured logically and is well written.

 

The following are aspects of the paper which would benefit from correction or clarification:

 

Introduction:

In the abstract you note that increased attention should be paid to patients with self-inflicted injury, it would be helpful to give some context in the introduction or methods section for what kind of care is currently provided for people who present to hospital in Japan with self-harm.

 

Methods:

The definition of suicide attempt and intentional injury is a little ambiguous. As it varies, it would be helpful to know what the definition of a suicide attempt is for the purpose of this paper. The authors mention that it was determined if it was a suicide attempt based partially on self-report. What was reported by the patients in order to classify it as an intentional self-injury, a expression of suicidal intent, or the patient stating that the injury was intentionally self-inflicted?

Explain what is meant by pre-hospital length of stay compared to hospital length  of stay. I see that it is in figure 2, but it would be good for this to be included in the methods also.

Consider noting how the variables included in the PS matching analyses selected. Was it due associations already established in the literature or were some crude associations conducted to select the most appropriate variables for the matching regression?

Suggest specifying the type of matching approach (e.g. greedy matching).  

Discussion:

Are there processes in place for carrying out mental health assessments with patients in the ED in Japan? If so, could that be a potential reason for hospital length of stay being longer for self-inflicted injuries.

The statement that those presenting with self-inflicted injuries would have high levels of psychological distress and uncooperative behavior requires a reference. I would also recommend rephrasing the sentence make it more neutral. As it is currently phrased, it comes across to the reader that all those who present with self-inflicted injuries have high levels of uncooperative behaviour, which is of course not the case.

There does not seem to be a mention of whether alcohol or drugs had been taken at the time of the injury. It is possible that this would have an impact on mortality risk and length of stay and may differ between the groups. If you do not have this variable to include, I recommend commenting on it.

 

Minor

Abstract

Line 35,  remove 2019 between ‘disease’ and ‘pandemic’

Include reference for injury severity rating.

In the discussion, the authors state: “Thus, the current findings emphasize the importance of measurement to prevent self-inflicted injury”. The phrasing is not very clear here, suggest changing the word measurement to something else.  

 

 

Author Response

Reviewer#1

 

Comments and Suggestions for Authors

This paper presents the results of a cohort study examining differences in outcomes following self-inflicted injury compared to injuries that were not self-inflicted. The paper addresses a topic of clinical and public health importance, with findings that can highlight the need for specific focus on providing care for people presenting with self-inflicted injuries. The comprehensive approach to addressing confounding is a major strength of this study. The manuscript is structured logically and is well written.

 

Response: Thank you for accurately summarizing the main points of our manuscript. We appreciate your positive evaluation of our manuscript.

 

 

The following are aspects of the paper which would benefit from correction or clarification:

 

Introduction: In the abstract you note that increased attention should be paid to patients with self-inflicted injury, it would be helpful to give some context in the introduction or methods section for what kind of care is currently provided for people who present to hospital in Japan with self-harm.

 

Response: Thank you for raising this point. On the basis of this comment, we have added brief explanations to improve the clarity of the text. At most EDs in Japan, including our own, patients with self-inflicted injury who are brought to the emergency department (ED) are initially evaluated by an emergency medical team consisting of attending emergency physicians, emergency medicine residents, post-graduate year 1 or 2 junior residents, and nurses. After the patients are physically stabilized, a psychiatric consultation is provided by in-house psychiatrists. If patients who attempted suicide need long-term psychiatric care, they are likely to be transferred to psychiatric hospitals. To clarify these points, we have made corresponding revisions in the R1 version of the manuscript (page 2, lines 73–79).

 

 

Methods: The definition of suicide attempt and intentional injury is a little ambiguous. As it varies, it would be helpful to know what the definition of a suicide attempt is for the purpose of this paper. The authors mention that it was determined if it was a suicide attempt based partially on self-report. What was reported by the patients in order to classify it as an intentional self-injury, a expression of suicidal intent, or the patient stating that the injury was intentionally self-inflicted?

 

Response: We agree that the definition of intentional injury in the previous version of our manuscript was unclear. Patients who were classified into the “self-inflicted group” were “patients who had trauma caused by suicide attempt.” Suicide attempts were determined via self-report by the patient, police report, or circumstantial evidence, such as the presence of a suicide note (R1 version, page 3, lines 107 to 110). To make the context clearer and convey our intention more precisely, we used the term of “suicide attempt” rather than “self-inflicted injury” throughout the manuscript. This point was also raised by another reviewer. Please also see our response to reviewer #2.

 

 

Explain what is meant by prehospital length of stay compared to hospital length of stay. I see that it is in figure 2, but it would be good for this to be included in the methods also.

 

Response: According to the reviewer’s suggestion, we have added the following sentences in the R1 version of the manuscript: (page 3, lines 110–116)

 

This study adopted prehospital LOS as a one of the trauma care parameters because prolonged prehospital LOS was known to be associated with poor outcomes of injured patients [14-17]. Many previous studies similarly considered prehospital LOS to be an important parameter of trauma care [14-17]. Hospital LOS was also deemed as a relevant care parameter that reflecting increased healthcare resources and costs [7-10, 18-20].

 

 

Consider noting how the variables included in the PS matching analyses selected. Was it due associations already established in the literature or were some crude associations conducted to select the most appropriate variables for the matching regression?

 

Response: We agree that our explanation of the method for selecting the variables included in the propensity score (PS) matching analyses was inadequate. In accord with the reviewer’s advice, we have revised the corresponding text as follows (page 3, lines 132–149):

 

A set of these variables was chosen a priori based on previous information and biological plausibility: The Injury Severity Score (ISS) and Abbreviated Injury Scale (AIS) are widely used anatomical scoring system that correlates with trauma mortality and morbidity [11-13]. The high Charlson Comorbidity Index was also known to be associated with increased mortality in patients with trauma [23, 24]. Therefore, these variables were incorporated into the logistic regression model to find the PS. To maximize model fitting, patients’ physiological parameters including GCS score, SBP, and respiratory rate were categorized (GCS score: 3, 4–5, 6–8, 9–12, and > 12; SBP: 1–49, 50–75, 76–89, and > 89 mmHg; and respiratory rate: > 29, 10–29, 6–9, 1–5, and 0 breaths/min) according to the scoring system of the Revised Trauma Score (RTS) [12]. The Charlson Comorbidity Index was also divided into four groups (0, 1, 2, and ≥ 3). Because our study period was relatively long, the data were divided into four phases (2002–2006, 2007–2011, 2012–2016, and 2017–2021) and each phase was considered as a possible confounder. A previous report employed a similar adjustment strategy [25, 26]. Mental illness, season, nighttime, and weekend admission were also incorporated into our model, as described above, because these variables are known to be associated with both suicide attempts and trauma outcomes [27-30].

 

 

Suggest specifying the type of matching approach (e.g. greedy matching). 

 

Response: We apologize for not specifying the type of matching approach in the previous version of our manuscript. In this study, matching analysis was conducted using greedy nearest neighbor one-to-one PS matching without replacement. Each patient in the suicide attempt group was matched with a patient in the no suicide attempt group, with the nearest estimated propensity on the logit scale within a specified range (0.2 of the pooled standard deviation of estimated logits) to reduce characteristic differences between the two groups. If two or more patients in the no suicide attempt group met this criterion, one patient was randomly selected for matching (R1 manuscript, page 3, lines 122 –125, and page 4, lines 154–157).

 

 

Discussion: Are there processes in place for carrying out mental health assessments with patients in the ED in Japan? If so, could that be a potential reason for hospital length of stay being longer for self-inflicted injuries.

 

Response: As described in response to your comment above, at most tertiary hospitals in Japan, patients who have trauma caused by suicide attempt and are brought to the ED are initially evaluated by an emergency medical team consisting of attending emergency physicians, emergency medicine residents, post-graduate year 1 or 2 junior residents, and nurses. After initial resuscitation, a psychiatric consultation is provided by in-house psychiatrists (page 2, lines 73–79).

 

We speculate that the prolonged hospital LOS observed in the suicide attempt group was largely caused by the circumstances described in the R1 manuscript (page 10, lines 299–316) rather than the mental health assessments carried out in the ED.

 

 

The statement that those presenting with self-inflicted injuries would have high levels of psychological distress and uncooperative behavior requires a reference. I would also recommend rephrasing the sentence make it more neutral. As it is currently phrased, it comes across to the reader that all those who present with self-inflicted injuries have high levels of uncooperative behavior, which is of course not the case.

 

Response: Thank you for noting this point. We agree with the reviewer’s advice. Accordingly, we have revised the corresponding text as follows (page 10, lines 308–310):

 

It is also plausible that suicide attempt group was more likely to have psychological symptoms compared with the no suicide attempt group, which are known to be associated with prolonged hospital stay [38, 39].

 

 

There does not seem to be a mention of whether alcohol or drugs had been taken at the time of the injury. It is possible that this would have an impact on mortality risk and length of stay and may differ between the groups. If you do not have this variable to include, I recommend commenting on it.

 

Response: As the reviewer indicated, alcohol or psychoactive drug use at the time of injury can affect the trauma outcome [44, 45]. Unfortunately, however, our database did not record these variables. Further analyses including these variables will be needed to further clarify the association between injuries caused by suicidal attempts and measured outcomes. This point is discussed in the revised limitations section (R1 manuscript, page 11, lines 334–336).

 

 

Minor

Abstract: Line 35, remove 2019 between ‘disease’ and ‘pandemic’

 

Response: According to the reviewer’s advice, we have removed “2019” from the R1 version of the manuscript (R1 manuscript, page 1, line 35).

 

 

Include reference for injury severity rating.

 

Response: According to the reviewer’s suggestion, we have added a brief explanation of injury severity rating scales, such as ISS, AIS, and RTS, as follows (R1 manuscript, page 3, lines 135–136 and page 3, lines 139–142):

 

The ISS and AIS are widely used anatomical scoring system that correlates with trauma mortality and morbidity. The patients’ physiological parameters including GCS score, SBP, and respiratory rate were categorized (GCS score: 3, 4–5, 6–8, 9–12, and > 12; SBP: 1–49, 50–75, 76–89, and > 89 mmHg; and respiratory rate: > 29, 10–29, 6–9, 1–5, and 0 breaths/min) according to the scoring system of the RTS [12].

 

 

In the discussion, the authors state: “Thus, the current findings emphasize the importance of measurement to prevent self-inflicted injury”. The phrasing is not very clear here, suggest changing the word measurement to something else.

 

Response: We agree with the reviewer’s advice, and have amended the corresponding text as follows (R1 manuscript, page 9, lines 257–258):

 

Thus, the current findings emphasize the importance of prevention of injury due to suicidal attempt.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This is an interesting and well written paper looking at the outcomes of people with severe injuries, comparing those with intentional injury (suicide attempt) and accidental injury. This is a solid paper with simple but appropriate analysis for the data. There are a couple of suggestions for improvements/clarifications below, but overall this work helps to clarify that self-inflicted injury may be more costly from a healthcare perspective, and therefore more investing in public health and preventative interventions would he helpful to reduce suicide attempts.

Title:

'self-inflicted injury' does not really capture what is being compared here as it is a broad label that could encompass many things and many levels of medical severity. 'Attempted suicide' would be more accurate as this is what is described later in the paper.

Methods:

A similar point to the above - there isn't much of a definition given about what would be included/excluded as a self-inflicted injury/suicide attempt 'case'. Does this include e.g., wounds from stabbing/cutting, attempted hanging, what about attempted drowning or traffic/height related trauma? And does self-poisoning appear anywhere or were these excluded?

Some discharge destinations, such as psychiatric hospital transfer, are going to be more common in people who attempted suicide - so I'm not sure what additional value this adds.

Analysis: seems appropriate and well conducted.

Discussion:

comment about possible impact of socioeconomic status on outcomes are really important. Was there no indicator that could be used for this in the analysis? e.g., small area economic statistics matched to patients home address, or claiming social security benefits?

The relatively small number of cases was acknowledged in the limitations section. In western populations not all self-inflicted injuries are recorded as such, for a variety of reasons. What is the situation in Japan? Is it likely that some people who are actually suicide attempts might be mistakenly classed as accidental injuries and included in the 'control' group instead? The small numbers over such a long study period suggest that some cases may have been missed.

Author Response

Reviewer#2

Comments and Suggestions for Authors

This is an interesting and well written paper looking at the outcomes of people with severe injuries, comparing those with intentional injury (suicide attempt) and accidental injury. This is a solid paper with simple but appropriate analysis for the data. There are a couple of suggestions for improvements/clarifications below, but overall, this work helps to clarify that self-inflicted injury may be more costly from a healthcare perspective, and therefore more investing in public health and preventative interventions would be helpful to reduce suicide attempts.

 

Response: Thank you very much for accurately summarizing the main points of our manuscript, and for your positive evaluation of our manuscript.

 

 

Title:

'self-inflicted injury' does not really capture what is being compared here as it is a broad label that could encompass many things and many levels of medical severity. 'Attempted suicide' would be more accurate as this is what is described later in the paper.

 

Response: We agree with the reviewer’s comment, and have used the term “suicide attempt” rather than “self-inflicted injury” throughout the revised manuscript. Reviewer #1 made a similar comment. Please also see our response to reviewer #1.

 

 

Methods: A similar point to the above - there isn't much of a definition given about what would be included/excluded as a self-inflicted injury/suicide attempt 'case'. Does this include e.g., wounds from stabbing/cutting, attempted hanging, what about attempted drowning or traffic/height related trauma? And does self-poisoning appear anywhere or were these excluded?

 

Response: Thank you for this helpful advice. Although our database included both blunt (e.g., traffic accident, fall, hanging) and penetrating (e.g., stabbing, cutting) injuries, burn injuries and suicide attempts without trauma (such as self-poisoning) were not included. Trauma etiology was dichotomized into blunt and penetrating trauma and was considered as one of the explanatory variables for PS matching.

Corresponding descriptions have been added to page 2, lines 88–91 and page 3, lines 122–133.

 

 

Some discharge destinations, such as psychiatric hospital transfer, are going to be more common in people who attempted suicide - so I'm not sure what additional value this adds.

 

Response: The current results revealed that injured patients who attempted suicide were more likely to be transferred to psychiatric hospitals (S2 Table in the supplementary material). As the reviewer indicated, this is not surprising. However, this may be one of the causes of prolonged hospital LOS in injured patients who attempted suicide. For example, physical stabilization is generally a prerequisite for psychiatric hospital transfer in Japan, which takes a relatively long time. To support the discussion on page 10, lines 298–316 of the R1 version of the manuscript, we would prefer to retain the corresponding results.

 

 

Analysis: seems appropriate and well conducted.

 

Response: Thank you very much for your positive evaluation of our manuscript.

 

 

Discussion: comment about possible impact of socioeconomic status on outcomes are really important. Was there no indicator that could be used for this in the analysis? e.g., small area economic statistics matched to patients’ home address, or claiming social security benefits?

 

Response: Thank you for this comment. Unfortunately, we did not measure variables that directly reflect the impact of socioeconomic status (e.g., income and insurance status of patients, hospital cost, functional outcome, etc.). Nevertheless, the current study revealed that hospital LOS, which is known to be related to increased care resources and cost, was significantly longer in the suicide group compared with the no suicide group. These results suggest that the suicide group required more healthcare resources and higher related costs.

 

 

The relatively small number of cases was acknowledged in the limitations section. In western populations not all self-inflicted injuries are recorded as such, for a variety of reasons. What is the situation in Japan? Is it likely that some people who are actually suicide attempts might be mistakenly classed as accidental injuries and included in the 'control' group instead? The small numbers over such a long study period suggest that some cases may have been missed.

 

Response: We appreciate the reviewer’s insight. As the reviewer indicated, it is possible that there were missed or misclassified suicide attempts. We speculate that suicide attempts may have been underestimated because some injured patients may have concealed their suicide attempts. This may have biased our results toward the null hypothesis. A corresponding discussion of this point has been added to the limitations section of the R1 version of the manuscript (page 11, lines 346–349).

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

I thank the authors the opportunity to revise the entiteled manuscript: Self-inflicted injury is independently associated with increased in-hospital mortality and hospital length of stay at a community tertiary hospital in Japan: A retrospective study with a propensity score matching analysis (2750980), submitted to MDPI IJERPH.

I've found an interesting research with soundness for public health, as well the authors present a relevant frame for decision making in emergencies attention, and a well performed methodology for delivering their conclusions.

I have some minor comments for the author's consideration, expecting this lead to possibility of clarifyin some small aspects from this valuable work.

1. Line 149, section 2.6. It is not clear what the authors intend with this section. Means the retrospective nature prevents calculating a sample? That may not always be the case. However, the second sentence refers that a power was computed, but authors do not refer for which one of the  statistical models used along the study.

2. Matching groups process: Could the authors please describe the matching process? for example diagnosis, or trauma type were matched?

3. Line 172, if matched pairs were matched, why are there these kind of differences between the groups? the decision to keep groups with these differences could lead to a bias in results interpretation. Could authors please explain some more detailed groups matching decisions? Table 1 is showing a statistic with a >.05 p value between groups in the matched cohort, however, if groups are supposed to be matched, the SD are very wide, for example in age.

4. Table 1. Was there any data related to previous medication by the in patients? Does p value in Table one correspond to Man-Whitney's U? or a different statistic to demonstrate invariance between the groups?

5. Line 321, related to section 2.6, Could the authors please explain for what particular statistical model was this power calculated?

 

Thank you once again.

 

 

 

 

It is not clear what the authors intend with this section. Means the retrospective nature prevents calculating a sample? That may not always be the case. However, the second sentence refers that a power was computed, but authors do not refer for which one of the  statistical models used along the study.

Comments for author File: Comments.pdf

Author Response

Reviewer#3

Comments and Suggestions for Authors

I thank the authors the opportunity to revise the entitled manuscript: Self-inflicted injury is independently associated with increased in-hospital mortality and hospital length of stay at a community tertiary hospital in Japan: A retrospective study with a propensity score matching analysis (2750980), submitted to MDPI IJERPH.

 

I've found interesting research with soundness for public health, as well the authors present a relevant frame for decision making in emergencies attention, and a well performed methodology for delivering their conclusions.

 

I have some minor comments for the author's consideration, expecting this lead to possibility of clarifying some small aspects from this valuable work.

 

Response: Thank you very much for your positive evaluation of our manuscript. We appreciate your positive comments.

 

 

1. Line 149, section 2.6. It is not clear what the authors intend with this section. Means the retrospective nature prevents calculating a sample? That may not always be the case. However, the second sentence refers that a power was computed, but authors do not refer for which one of the statistical models used along the study.

 

Response: We agree that the corresponding text in the previous version of the manuscript was not sufficiently clear. We intended to convey that the retrospective nature of the study predetermined the sample size, and the observed power was computed post hoc using G*Power 3 for Windows (Heinrich Heine University, Dusseldorf, Germany) for all primary outcomes examined. The primary outcome measure in this study was in-hospital mortality, which is the proportional differences between the two groups. Therefore, post-hoc power was computed for the chi-squared test.

 

 

2. Matching groups process: Could the authors please describe the matching process? for example diagnosis, or trauma type were matched?

 

Response: In accord with the reviewer’s advice, we revised the corresponding description in the methods section, as follows (page 3, lines 122–149):

 

Matching analysis was conducted using greedy nearest neighbor one-to-one PS matching without replacement. Multivariable logistic regression was used to find PS to predict the probability of being assigned to the self-inflicted group. In addition to age and sex, Charlson Comorbidity Index [21, 22], diagnosed mental illness, presentation time and period (8:00–16:59, 17:00–23:59, and 24:00–7:59, weekend or weekday and 2002–2006, 2007–2011, 2012–2016, and 2017–2021, respectively), season (spring, March–May; summer, June–August; autumn, September–November; and winter, December–February), initial recorded vital signs (Glasgow Coma Scale [GCS] score, systolic blood pressure [SBP], and respiratory rate), trauma etiology (blunt or penetrating), injury distribution with AIS ≥ 3, and ISS were selected as explanatory variables for the logistic regression.

 A set of these variables was chosen a priori based on previous information and biological plausibility: The ISS and AIS are widely used anatomical scoring system that correlates with trauma mortality and morbidity [11-13]. The high Charlson Comorbidity Index was also known to be associated with increased mortality in patients with trauma [23, 24]. Therefore, these variables were incorporated into the logistic regression model to find the PS. To maximize model fitting, patients’ physiological parameters including GCS score, SBP, and respiratory rate were categorized (GCS score: 3, 4–5, 6–8, 9–12, and > 12; SBP: 1–49, 50–75, 76–89, and > 89 mmHg; and respiratory rate: > 29, 10–29, 6–9, 1–5, and 0 breaths/min) according to the scoring system of the RTS [12]. The Charlson Comorbidity Index was also divided into four groups (0, 1, 2, and ≥ 3). Because our study period was relatively long, the data were divided into four phases (2002–2006, 2007–2011, 2012–2016, and 2017–2021) and each phase was considered as a possible confounder. A previous report employed a similar adjustment strategy [25, 26]. Mental illness, season, nighttime, and weekend admission were also incorporated into our model, as described above, because these variables are known to be associated with both suicide attempts and trauma outcomes [27-30]. All categorical variables mentioned above were dummy coded and incorporated into the PS model.

 

Reviewer #1 made a similar comment. Please also see our response to reviewer #1.

 

 

3. Line 172, if matched pairs were matched, why are there these kind of differences between the groups? the decision to keep groups with these differences could lead to a bias in results interpretation. Could authors please explain some more detailed groups matching decisions?

 

Table 1 is showing a statistic with a >.05 p value between groups in the matched cohort, however, if groups are supposed to be matched, the SD are very wide, for example in age.

 

Response: Thank you for this helpful comment. We acknowledge that in the previous version of the manuscript, our description of the method for determining matching decisions was insufficient. In the revised manuscript, we have described the process for determining group matching decisions, as follows (R1 manuscript, page 4, lines 154–158):

 

Each patient in suicide attempt group was matched with a patient in no suicide attempt group, with the nearest estimated propensity on the logit scale within a specified range (0.2 of the pooled standard deviation of estimated logits) to reduce characteristic differences between the two groups. If two or more patients in the no suicide attempt group met this criterion, one patient was randomly selected for matching.

 

In this study, we used standardized difference (SD) to evaluate the covariate balance; an absolute SD of > 10% represents meaningful imbalance [31] (R1 manuscript, page 4, lines 151–153). Please note, after PS-matching, patient distributions were closely balanced, with all SD < 10% between the two groups (Table 1).

 

 

4. Table 1. Was there any data related to previous medication by the in patients? Does p value in Table one corresponds to Man-Whitney's U? or a different statistic to demonstrate invariance between the groups?

 

Response: As the reviewer notes, psychoactive drug or alcohol use can affect the outcomes of trauma patients [44, 45]. Unfortunately, however, our database did not record this variable. This point is acknowledged in the revised limitations section (R1 manuscript, page 11, lines 335–337). Reviewer #1 raised a similar point. Please also see our response to reviewer #1.

 

The p-values listed in Table 1 were derived from the Mann–Whitney U test or chi-squared tests. We have added a corresponding explanation in the legend of Table 1 in the R1 manuscript (page 7, lines 209–210).

 

 

5. Line 321, related to section 2.6, Could the authors please explain for what particular statistical model was this power calculated?

Thank you once again.

 

Response: As we noted in response to your comment #1, the observed power was computed post hoc using G*Power 3 for Windows (Heinrich Heine University, Dusseldorf, Germany) for in-hospital mortality, which is the proportional differences. Therefore, post-hoc power was computed for the chi-squared test.

 

 

Author Response File: Author Response.pdf