Abstract
Background
A minority of persons who have traumatic experiences go on to develop post-traumatic stress disorder (PTSD), leading to interest in who is at risk for psychopathology after these experiences. Complicating this effort is the observation that post-traumatic psychopathology is heterogeneous. The goal of this nested case-control study was to identify pre-trauma predictors of severe post-traumatic psychiatric comorbidity, using data from Danish registries.
Methods
The source population for this study was the population of Denmark from 1994 through 2016. Cases had received three or more psychiatric diagnoses (across all ICD-10 categories) within 5 years of a traumatic experience (n = 20 361); controls were sampled from the parent cohort using risk-set sampling (n = 81 444). Analyses were repeated in samples stratified by pre-trauma psychiatric diagnoses. We used machine learning methods (classification and regression trees and random forest) to determine the important predictors of severe post-trauma psychiatric comorbidity from among hundreds of pre-trauma predictor variables spanning demographic and social variables, psychiatric and somatic diagnoses and filled medication prescriptions.
Results
In the full sample, pre-trauma psychiatric diagnoses (e.g. stress disorders, alcohol-related disorders, personality disorders) were the most important predictors of severe post-trauma psychiatric comorbidity. Among persons with no pre-trauma psychiatric diagnoses, demographic and social variables (e.g. marital status), type of trauma, medications used primarily to treat psychiatric symptomatology, anti-inflammatory medications and gastrointestinal distress were important to prediction. Results among persons with pre-trauma psychiatric diagnoses were consistent with the overall sample.
Conclusions
This study builds on the understanding of pre-trauma factors that predict psychopathology following traumatic experiences, by examining a broad range of predictors of post-trauma psychopathology and comorbidity beyond PTSD.
Keywords: Trauma, post-traumatic psychopathology, comorbidity
Key Messages.
Few studies have prospectively examined pre-trauma predictors of post-trauma psychopathology; those that did so have focused on post-traumatic stress disorder as the sole outcome.
We used Danish national registry data to examine pre-trauma predictors of severe psychiatric comorbidity in the 5 years following trauma among persons with and without pre-trauma psychiatric diagnoses.
Pre-trauma psychiatric diagnoses and associated medications, demographic and social variables, anti-inflammatory medications and gastrointestinal distress were all important to prediction of severe post-trauma psychiatric comorbidity.
This study builds on the existing literature by examining risk for psychiatric diagnoses following trauma, beyond post-traumatic stress disorder.
Introduction
Despite the high prevalence of trauma worldwide,1 only a small proportion of people develop post-traumatic psychopathology that is severe enough to warrant a clinical diagnosis, usually post-traumatic stress disorder (PTSD).2 What differentiates persons who develop post-traumatic psychopathology from persons who are resilient is a topic of current interest.3 In addition, traumatic experiences are typically unexpected, and whereas a rich body of literature has documented peri- and post-traumatic predictors of PTSD,4–7 only a small handful of studies have prospectively examined pre-trauma predictors of post-traumatic psychopathology.8 This is an important area for investigation, because an early understanding of who is most at risk for psychopathology has resource and intervention implications. To date, studies examining pre-traumatic predictors of PTSD have been conducted using cross-sectional population-based surveys. For example in the World Mental Health Surveys, demographic characteristics, trauma type and retrospectively reported pre-trauma psychiatric diagnoses were associated with PTSD.9
There is also increasing awareness that post-traumatic psychopathology is heterogeneous, and that the largely single disorder focus on PTSD to date is obscuring a full understanding of post-trauma psychopathology. Several studies have documented that post-trauma psychopathology may take many forms,10–14 including psychiatric comorbidity.15–17 In previous work using Danish registry data, we showed that the risk for a range of psychiatric diagnoses is increased in the 5 years following a traumatic event, and that some individuals experience several disorders.18,19 To date, no study has examined pre-trauma predictors of post-traumatic psychopathology beyond PTSD. Further, no study has examined risk of post-trauma psychopathology among persons with no documented pre-trauma psychiatric diagnoses. This evidence gap arises in part because of data limitations: it is difficult to find datasets that include a broad range of psychiatric diagnoses with information about persons before and after traumatic events. To address this gap, we used data from the Danish national registries to examine pre-trauma predictors (demographic characteristics, social variables, psychiatric diagnoses, physical health diagnoses and medications) of severe psychiatric comorbidity in the 5 years following trauma.
Methods
The source population was the population of Denmark between 1994 and 2016 (n = 7 420 888). We chose 1994 as the start of the study period because it coincides with implementation of ICD-10 coding in Denmark (prior to 1994, ICD-8 was used). From this population, we created a traumatic experience cohort (n = 1 406 637) using methods described elsewhere (and in Supplementary Table S1, available as Supplementary data at IJE online).18 In brief, we constructed a cohort of all persons who experienced at least one of eight traumatic events from 1994 to 2016: fire/explosion, transportation accident, exposure to a toxic substance and medical complications, traumatic brain injury, physical assault, assault with a weapon, pregnancy-related trauma and suicide death of a family member. Persons with multiple experiences on the same day were classified in a ‘multiple trauma’ group. Cohort entry occurred on the date of the first traumatic event within the study period. For many traumatic experiences we were able to use ICD-10 and other registry-based codes as an initial determination of event occurrence. We then applied additional criteria to our traumatic experience definitions (related to longer than expected hospital admissions) to increase the likelihood of capturing events that were more severe than minor stressors, and correspondence with the Diagnostic and Statistical Manual for Mental Disorders, 5th edition.20
The current case-control study was nested within this traumatic experience cohort. Cases were persons with severe psychiatric comorbidity, defined as those who received at least three psychiatric diagnoses across ICD-10 diagnostic categories within 5 years of the date of their traumatic experience (n = 20 361). Three diagnoses were chosen as the definition for severe psychopathology because epidemiological studies have shown that the majority of persons with post-traumatic psychopathology meet criteria for at least three disorders in their lifetime.21 The case date was the date of the third psychiatric disorder diagnosis. Diagnoses used to construct the case definition were obtained from the following categories of ICD-10 psychiatric disorders: organic mental disorders (ICD codes F00-F09); mental and behavioural disorders due to psychoactive substance use (F10-F19); schizophrenia, schizotypal and delusional disorders (F20-F29); manic episode or bipolar disorders (F30-F31, F34-F39); depressive disorders (F32-F33); neurotic and somatoform disorders (F40-F42, F44-F48); stress disorders (F43); behavioural syndromes associated with physiological disturbances and physical factors (F50-F59); and disorders of adult personality and behaviour (F60-F69). Controls were selected from the original trauma cohort from those at risk for the outcome on their matched cases’ date via risk-set sampling (without replacement) at a ratio of 4:1 (n = 81 444). We did not match controls to cases on any factors, so that all variables could be evaluated with respect to prediction importance.
We conducted secondary analyses stratified by the presence of a recorded psychiatric diagnosis before the traumatic experience. In these analyses, there were 8405 cases and 33 620 controls in the group that had no re-coded pre-trauma psychiatric diagnoses and 11 956 cases and 47 824 controls in the group that had recorded pre-trauma psychiatric diagnoses. Importantly, for the stratified analysis of persons with and without psychiatric disorders prior to trauma, all included cases were derived from the full case-control sample presented in this manuscript; however, to preserve the 4:1 control to case matching, additional controls needed to be selected from the parent trauma cohort for these analyses.
National registries
Medical care is provided to all residents of Denmark through a tax-funded system; receipt of health care and social variables is recorded in national medical and administrative registries.22 A unique personal identifier assigned at birth or immigration (the Civil Personal Register number) was used to merge culled data from the Danish national registries.23
Demographic data (age, sex and marital status) were obtained from the Danish Civil Registration System, which was established in 1968 and has been updated daily since 1989.23 Data on income and employment were obtained from the Income Statistics Register and the Integrated Database for Labor Market Research.24,25
The Danish Psychiatric Central Research Register has recorded psychiatric inpatient care for decades, with the inclusion of outpatient clinical care added in 1995.26 It includes admission and discharge dates with a primary and up to 20 secondary diagnoses per entry. Several studies have documented the high quality of diagnoses in this registry when compared with independent symptom re-assessment.27,28
The Danish National Patient Registry has recorded diagnoses received in any somatic inpatient hospital since 1977 and outpatient care since 1995, including treatment dates and a primary and up to 20 secondary diagnostic codes, surgery codes and examination codes.29 Studies have documented the high validity of many key diagnoses recorded in this registry.30
The Danish National Prescription Registry includes information on all prescription drugs sold in Denmark, with complete and valid data recorded since 1995.31 The registry includes dates prescription were filled and drug name and Anatomical Therapeutic Classification (ATC) code. We classified medications according to level three ATC codes.
Pre-trauma predictors
Predictors spanned demographic and social variables, psychiatric diagnoses, physical health diagnoses, surgeries and prescribed medications (‘ever use’), and were evaluated before the traumatic experience date. The type of traumatic experience itself was also included as a predictor, consistent with other similar studies.9 A complete listing of predictors can be found in Supplementary Table S2 (available as Supplementary data at IJE online). The data reduction process, aimed at preventing model overfit,32,33 focused on the elimination of rare predictors (less than 10 observations).34,35 We also eliminated diagnoses occurring in the emergency room, due to data quality concerns. Following data reduction, the final number of pre-trauma predictors included in the models was 449 for the full sample, 357 among the sample with no recorded psychiatric diagnoses prior to trauma and 455 among the sample of persons with psychiatric disorders before trauma.
Statistical analysis
We used two recursive partitioning machine learning methods that automate detection of associations between predictors and outcome and interactions among predictors and provide predictor importance metrics. We estimated classification and regression tree (CART) models as an initial evaluation of the risk structure.36 CART builds a decision tree based on predictors and their combinations, which results in the highest probability of differentiating cases and controls. CART was implemented using the R package rpart, which uses a 10-fold cross-validation procedure.37 To increase visual interpretability and decrease risk of overfit, maximum tree depth and minimum number of observations in any node were set to 10.
Second, we implemented random forest, which uses bootstrap aggregation to produce results that are less prone to model overfit than CART, using the R package randomForest.38 Each forest was built with 1000 trees, a minimum of 10 observations needed to attempt a split and 20 variables sampled as split candidates at each node (18 in the secondary sample; i.e. the square root of total number of predictors, the randomForest default). Random forest was implemented using 2-fold cross-validation—the sample was randomly split into two folds, random forest implemented in each fold, and individual-level predicted values were estimated by applying the fold 1 model to fold 2 observations and fold 2 model to fold 1 observations. We used mean decrease in accuracy to evaluate each variable in terms of main effects and interactions across all trees in both folds (i.e. replicating variable importance across subsamples).39 Overall model performance of the random forest models was evaluated by using the cross-validated predicted values to calculate area under the receiver operating characteristic curves and the proportion of total cases accounted for at varying predicted risk thresholds. Analyses were conducted in SAS version 9.4 and R version 3.5.2.40
Results
Descriptive characteristics of the cases and controls from the full sample, and the stratified samples are displayed in Table 1. Similar to the composition of the overall cohort from which the current sample was obtained,18 cases and controls were predominately female, under 40 years of age and single. Cases were more likely than controls to be in a lower income quartile. Regarding incident traumatic experience within the study period, cases more frequently experienced exposure to a toxic substance and multiple traumatic experiences on the same day, with a pattern among persons who had psychiatric diagnoses prior to trauma that differed slightly from the full sample and the sample without a psychiatric diagnosis prior to trauma.
Table 1.
Characteristics of the samples, Denmark 1994 to 2016
| Full sample |
No pre-trauma psychiatric diagnoses |
Pre-trauma psychiatric diagnoses |
||||
|---|---|---|---|---|---|---|
| Cases (N = 20 361) | Controls (N = 81 444) | Cases (N = 8405) | Controls (N = 33 620) | Cases (N = 11 956) | Controls (N = 47 824) | |
| n (%) | n (%) | n (%) | n (%) | n (%) | n (%) | |
| Sex | ||||||
| Male | 7873 (39) | 25 542 (31) | 3065 (36) | 10 360 (31) | 4808 (40) | 19 068 (40) |
| Female | 12 488 (61) | 55 902 (69) | 5340 (64) | 23 260 (69) | 7148 (60) | 28 756 (60) |
| Age at traumatic experience | ||||||
| <30 years | 8934 (44) | 43 573 (54) | 4325 (51) | 18 450 (55) | 4618 (39) | 17 207 (36) |
| 31–40 years | 3482 (17) | 18 987 (23) | 1377 (16) | 7946 (24) | 2105 (18) | 9315 (20) |
| 41–50 years | 3121 (15) | 5892 (7.2) | 1002 (12) | 2225 (6.6) | 2119 (18) | 6760 (14) |
| 51–60 years | 2133 (10) | 4928 (6.1) | 660 (7.9) | 1847 (5.5) | 1473 (12) | 5750 (12) |
| >60 years | 2682 (13) | 8064 (9.9) | 1041 (12) | 3152 (9.4) | 1641 (14) | 8792 (18) |
| Employment statusa | ||||||
| Employed | 6646 (33) | 46 669 (57) | 3240 (39) | 19 654 (58) | 3406 (29) | 17 728 (37) |
| Unemployed | 5128 (25) | 10 050 (12) | 2275 (27) | 4096 (12) | 2853 (24) | 8009 (17) |
| Early retirement | 5236 (26) | 7153 (8.8) | 1226 (15) | 2642 (7.9) | 4010 (34) | 12 612 (26) |
| State pension | 2510 (12) | 6123 (7.5) | 1069 (13) | 2422 (7.2) | 1441 (12) | 7526 (16) |
| Age <14 (not employed) | 728 (3.6) | 10 803 (13) | 535 (6.4) | 4583 (14) | 193 (1.6) | 1771 (3.7) |
| Missing | 113 (0.55) | 646 (0.79) | 60 (0.71) | 223 (0.66) | 53 (0.40) | 178 (0.40) |
| Marital statusa | ||||||
| Married/registered partnership | 4486 (22) | 29 143 (36) | 2039 (24) | 12 370 (37) | 2447 (21) | 12 285 (26) |
| Single | 11 558 (57) | 42 010 (52) | 4875 (58) | 17 338 (52) | 6683 (56) | 24 439 (51) |
| Divorced | 3228 (16) | 5040 (6.2) | 1020 (12) | 1808 (5.4) | 2208 (19) | 7672 (16) |
| Widowed | 955 (4.7) | 2722 (3.3) | 397 (4.7) | 1042 (3.1) | 558 (4.7) | 3179 (6.6) |
| Unknown | 134 (0.66) | 2529 (3.1) | 74 (0.88) | 1062 (3.2) | 60 (0.50) | 275 (0.60) |
| Income quartilea | ||||||
| 1st (lowest) | 6784 (33) | 12 878 (16) | 2865 (34) | 5136 (15) | 3919 (33) | 11 323 (24) |
| 2nd | 6499 (32) | 15 408 (19) | 2165 (26) | 6028 (18) | 4334 (36) | 17 575 (37) |
| 3rd | 3826 (19) | 21 290 (26) | 1604 (19) | 8944 (27) | 2222 (19) | 10 845 (23) |
| 4th (highest) | 1995 (9.8) | 19 532 (24) | 921 (11) | 8276 (25) | 1074 (9.0) | 5819 (12) |
| Age <14 (no income) | 728 (3.6) | 10 803 (13) | 535 (6.4) | 4583 (14) | 193 (1.6) | 1771 (3.7) |
| Missing | 529 (2.6) | 1533 (1.9) | 315 (3.7) | 653 (1.9) | 214 (1.8) | 491 (1.0) |
| Type of traumatic experience | ||||||
| Fire/explosion | 815 (4.0) | 7694 (9.4) | 382 (4.5) | 3308 (9.8) | 433 (3.6) | 2696 (5.6) |
| Transportation accident | 258 (1.3) | 670 (0.82) | 122 (1.5) | 210 (0.62) | 136 (1.1) | 732 (1.5) |
| Exposure to toxic substance | 12 525 (62) | 18 109 (22) | 4552 (54) | 7106 (21) | 7973 (67) | 20 423 (43) |
| Traumatic brain injury | 4102 (20) | 18 229 (22) | 1936 (23) | 7404 (22) | 2166 (18) | 11 643 (24) |
| Physical assault | 533 (2.6) | 824 (1.0) | 212 (2.5) | 297 (0.88) | 321 (2.7) | 1199 (2.5) |
| Assault with weapon | 18 (0.088) | 21 (0.026) | 5 (0.059) | 6 (0.018) | 13 (0.1) | 12 (0.1) |
| Pregnancy-related trauma | 1857 (9.1) | 34 934 (43) | 1062 (13) | 14 922 (44) | 795 (6.6) | 10 602 (22) |
| Suicide death of family | 94 (0.46) | 641 (0.79) | 58 (0.69) | 273 (0.81) | 36 (0.3) | 210 (0.4) |
| Multiple traumas | 159 (0.78) | 322 (0.40) | 76 (0.90) | 94 (0.28) | 83 (0.7) | 307 (0.6) |
Stratified case groups are subsamples of the case group presented in the overall case-control sample, but additional controls were selected from the parent trauma cohort to maintain the matching for the subgroup analyses.
aIn the year before the incident trauma.
Results from the full sample
The total sample CART model is displayed in Figure 1. Not surprisingly, having a psychiatric diagnosis before traumatic experiences, and medications used to treat psychiatric disorders, were important predictors of severe psychiatric comorbidity following traumatic experiences. The most salient diagnoses and medications included pre-trauma alcohol use disorder, stress disorders, antidepressant use and antipsychotic use, which were generally associated with increased risk of severe psychiatric comorbidity. As an example, persons who were prescribed antidepressants before the traumatic experience had a 63% risk of severe psychiatric comorbidity following the traumatic experience. Social variables such as low-income quartile and unemployment, were also important to prediction. For example, persons who were exposed to a toxic substance, never married and in the lowest income quartile had a 53% risk of severe post-trauma psychiatric comorbidity. Regarding traumatic experience type, pregnancy-related trauma was an important predictor of low risk of severe post-trauma psychopathology (2.9%), whereas exposure to a toxic substance was part of many profiles of increased risk. The random forest results largely corroborated these findings while also providing evidence of additional potentially important predictors (Figure 2). Age, pre-trauma depression and personality disorders were all revealed as important to prediction as indicated by the large mean decrease in prediction accuracy if these variables were removed from the model. Two categories relevant to physical health were additionally important predictors of severe psychiatric morbidity following traumatic experiences—pre-trauma use of anti-inflammatory non-steroidal products and pre-trauma symptoms involving the digestive system and abdomen.
Figure 1.
Classification and Regression Tree (CART) for the full sample
Figure 2.
Variable importance plot for the full sample
Random forest model performance
The cross-validated area under the curve (AUC) was 0.91 [95% confidence interval (CI) 0.91–0.92). Individuals in the top decile and top quintile of predicted risk, respectively, accounted for 41.9% and 70.5% of all cases of severe psychiatric morbidity.
Results among persons without pre-trauma psychiatric diagnoses
The CART model among persons who did not have a recorded ICD-10 psychiatric diagnosis before a traumatic experience is displayed in Figure 3. Pre-trauma medications typically used to treat psychiatric disorders (e.g. antidepressants) were still important to prediction even though no psychiatric diagnoses were present in the sample at this time (e.g. persons on antidepressants before a traumatic experience had a 38% risk of severe post-trauma psychiatric comorbidity). Further, a larger number of demographic and social variables such as income, employment status and marital status were important to prediction among the group without pre-trauma psychiatric diagnoses. Similar to the full sample results, pregnancy-related trauma was associated with a decreased risk of post-trauma severe psychiatric morbidity (risk = 3.9% in the group that had pregnancy-related trauma but no other characteristics) whereas exposure to a toxic substance was associated with an increased risk. Figure 4 displays the results of the associated random forest analysis that largely corroborates these findings. Similar to the overall sample results, the presence of anti-inflammatory drugs and gastrointestinal drugs was important to prediction, as indicated by a large mean decrease in model accuracy if these predictors were removed.
Figure 3.
Classification and Regression Tree (CART) for persons without pre-trauma psychiatric diagnoses
Figure 4.
Variable importance plot for persons without pre-trauma psychiatric diagnoses
Random forest model performance
Cross-validated AUC was 0.86 (95% CI 0.86–0.87). Individuals in the top decile and top quintile of predicted risk, respectively, accounted for 35.5% and 60.4% of all cases of severe psychiatric morbidity among those without a pre-trauma psychiatric diagnosis.
Results among persons with pre-trauma psychiatric diagnoses
Supplementary Figures S1 and S2 (available as Supplementary data at IJE online) display the CART and random forest results for the sample who had recorded psychiatric diagnoses prior to their traumatic experience. In general, the pattern of results was consistent with the full sample results presented above, such that psychiatric disorders and associated medications (e.g. personality disorders, antidepressants, major depressive disorder) were most important to prediction of severe psychopathology 5 years after trauma. Consistent results were also found with regard to social variables and type of trauma.
Discussion
We used longitudinal, population-based data to predict severe psychiatric comorbidity following traumatic experiences, from many pre-trauma variables. Prior literature on predicting psychopathology following trauma has largely focused on PTSD.9,41 This study contributes to the literature by describing pre-trauma demographic, social, psychiatric and physical health predictors of broadly defined severe post-traumatic psychiatric comorbidity.
Consistent with the existing literature, psychiatric diagnoses and prescriptions for associated medications before traumatic experiences were very important to the prediction of severe post-trauma psychiatric comorbidity.9 Unexpectedly, however, in our analyses among persons who had no recorded psychiatric diagnoses before the traumatic experience, medications typically used for the treatment of psychiatric disorders remained important to prediction. This result could be due to at least three factors. First, it is likely that persons with subsyndromal psychopathology before trauma (who may still be prescribed medication) are at an increased risk of severe psychiatric comorbidity following trauma. A study of US National Guard members found that most people who developed PTSD over longitudinal follow-up had subsyndromal PTSD at the start of the study.42 Given our use of medical registry-based data, we were unable to fully capture all levels of psychiatric symptomatology; this is an important direction for future research. A second possibility is that psychiatric diagnoses are misclassified in the registries, so some persons were likely classified as not having pre-event diagnoses when they may in fact have had diagnoses. This may be a particular issue for persons who had ICD-8 psychiatric diagnoses assigned before our study period. Although diagnostic misclassification is a possibility, it seems unlikely that it would have a big impact on our results because (i) a large number of persons in our sample did not have psychiatric diagnoses in the study period before traumatic experiences, and (ii) many validation studies have shown high validity of psychiatric diagnoses in the Danish registries when compared with independent re-assessment.27,28 Finally, it is possible that persons in the sample were prescribed psychiatric medications for reasons other than psychiatric disorders (e.g. antidepressants for chronic pain).
Across persons with and without recorded psychiatric diagnoses before traumatic experiences, social variables such as employment status, income and marital status were important to prediction of severe post-trauma psychopathology. It is particularly interesting that social variables were still important to prediction, even in the well-resourced social context of Denmark. There is a growing appreciation of the impact of social determinants of health across domains.43 Future research should replicate these results in settings with more variable social context (such as the USA) to determine if these findings are more pronounced in these settings.
Similarly, we found evidence that certain physical health diagnoses and related medications were associated with severe post-trauma psychopathology. Specifically, pre-trauma anti-inflammatory and anti-rheumatic medications were important to prediction of severe psychiatric comorbidity following trauma, in both groups. It has long been hypothesized that stress and infections are associated in potentially reciprocal ways through biological and/or behavioural pathways.44 This study provides further evidence that pre-trauma infections and inflammation may be important predictors of psychopathology following traumatic experiences, regardless of pre-existing psychiatric diagnoses. Further, pre-trauma symptoms of digestive distress were also important to prediction. Stress-related disorders have been shown to be associated with later gastrointestinal issues,45,46 but this study indicates this association may be bi-directional.
This study has several limitations. Whereas identifying a trauma cohort within national, longitudinal, registry-based data paves the way for many advancements, the process of defining criteria for traumatic experiences in registry-based data inevitably included investigator-based choices that impacted on cohort membership. Future studies could restrict the cohort further to determine if patterns of results are consistent within subgroups with strict indications of traumatic experiences. Further, an important next step in this line of research would be to conduct a validation study of the trauma classifications through comparisons with either medical chart notes or independent re-interview. Although we were unable to conduct such a study within the resources of the current project, this effort would provide incredibly valuable information with regard to the accuracy of created trauma variables. We were also not able to include some important traumatic experiences (e.g. sexual assault) because of the sensitive nature and limited availability of these data.
Similarly, although we were not able to include data on childhood trauma, our cohort included a small proportion of children. Future research with purposeful sampling by age group could explore possible differences in the predictors of severe psychiatric comorbidity by age of traumatic experience. We chose a 5-year window for the assessment of outcomes. The choice of any specific outcome period is dependent on the assumption that the period is short enough for observed outcomes to be the effects of the traumatic experiences while also long enough to capture disorders with long induction times. It is possible that using a different outcome time period for severe comorbid post-traumatic psychopathology would have yielded different results. In addition, the nature of medical registry-based data limits the inclusion of behavioural and other lifestyle variables that may be important to prediction of psychopathology following trauma. Future studies that are able to includes these data along with medical registry data will improve upon the presented models. With regard to statistical limitations, we relied on only two machine learning classifiers: CART and random forest. Other ‘black box’ classifiers or meta-classifiers may improve prediction performance.47 The analytical dataset was left-truncated (beginning in 1994 with the implementation of ICD10) such that cases and controls from the earliest part of our study period may have had incomplete data on predictors that occurred during the ICD-8 diagnostic period. This would apply to only a small proportion of our overall sample but should be kept in mind when interpreting our results. Similarly, diagnoses received in the emergency room were excluded due to validity concerns, but it is possible that the inclusion of these data would have improved model performance. Finally, it is unclear how the results of the current study would generalize to populations with different social settings, although the consistency with the larger trauma psychopathology literature alleviates this concern.
Despite these limitations, this study provides information on combinations of pre-trauma predictors that are most likely to result in severe psychiatric comorbidity within 5 years following traumatic experiences. While corroborating existing knowledge, this study also provides additional information about these associations, particularly among persons who are without recorded psychiatric disorders before traumatic experiences. This work can be used as the basis for research aimed at better elucidating who is at most risk for severe psychiatric comorbidity following trauma and for more detailed studies of the forms of psychiatric comorbidity that may occur following trauma.
Ethics approval
This work was deemed ‘not human subjects research’ by the Institutional Review Board at Boston University and reported to the Danish Data Protection Agency (record no. 2015–57-0002).
Data availability
Access to the data used for the current study is subject to all appropriate Danish approvals. Please contact Drs Gradus or Sørensen for information about how to initiate that process.
Supplementary data
Supplementary data are available at IJE online.
Author contributions
All authors have contributed to the drafting and revision of this manuscript, as well as the scientific conceptualization of the overall study. P.S., E.H-P. and H.T.S. were responsible for the acquisition of data; P.S., E.H-P., A.J.R. and J.L.G. were responsible for the integrity of analyses.
Funding
The work presented herein corresponds to the specific aims of National Institute of Mental Health (NIMH) grant # R01MH110453. (J.L.G.). H.T.S. was supported by a grant from the Lundbeckfonden (grant # R248-2017–521).
Supplementary Material
Acknowledgement
We thank Tammy Jiang, PhD, for her assistance in generating the variable importance plots for publication.
Conflict of interest
None declared.
Contributor Information
Jaimie L Gradus, Department of Epidemiology, Boston University School of Public Health, Boston, MA, USA; Department of Psychiatry, Boston University School of Medicine, Boston, MA, USA; Department of Clinical Epidemiology, Aarhus University Hospital, Aarhus, Denmark.
Anthony J Rosellini, Center for Anxiety and Related Disorders, Department of Psychological and Brain Sciences, Boston University, Boston, MA, USA.
Péter Szentkúti, Department of Clinical Epidemiology, Aarhus University Hospital, Aarhus, Denmark.
Erzsébet Horváth-Puhó, Department of Clinical Epidemiology, Aarhus University Hospital, Aarhus, Denmark.
Meghan L Smith, Department of Epidemiology, Boston University School of Public Health, Boston, MA, USA.
Isaac Galatzer-Levy, Department of Psychiatry, NYU Grossman School of Medicine, New York, NY, USA.
Timothy L Lash, Department of Epidemiology, Rollins School of Public Health, Emory University, Atlanta, GA, USA.
Sandro Galea, Department of Epidemiology, Boston University School of Public Health, Boston, MA, USA.
Paula P Schnurr, Executive Division, National Center for PTSD, White River Junction, VT, USA; Department of Psychiatry, Geisel School of Medicine at Dartmouth, Hanover, NH, USA.
Henrik T Sørensen, Department of Epidemiology, Boston University School of Public Health, Boston, MA, USA; Department of Clinical Epidemiology, Aarhus University Hospital, Aarhus, Denmark.
References
- 1. Kessler RC, Aguilar-Gaxiola S, Alonso J. et al. Trauma and PTSD in the WHO World Mental Health Surveys. Eur J Psychotraumatol 2017;8:1353383. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Breslau N, Kessler RC, Chilcoat HD, Schultz LR, Davis GC, Andreski P.. Trauma and post-traumatic stress disorder in the community: the 1996 Detroit Area Survey of Trauma. Arch Gen Psychiatry 1998;55:626–32. [DOI] [PubMed] [Google Scholar]
- 3. Yehuda R, Hoge CW, McFarlane AC. et al. Post-traumatic stress disorder. Nat Rev Dis Primers 2015;1:1–22. [DOI] [PubMed] [Google Scholar]
- 4. Massazza A, Joffe H, Hyland P, Brewin CR.. The structure of peritraumatic reactions and their relationship with PTSD among disaster survivors. J Abnorm Psychol 2021;130:248–59. [DOI] [PubMed] [Google Scholar]
- 5. Adams RE, Boscarino JA.. Predictors of PTSD and delayed PTSD after disaster: The impact of exposure and psychosocial resources. J Nerv Ment Dis 2006;194:485–93. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Kessler RC, Galea S, Gruber MJ, Sampson NA, Ursano RJ, Wessely S.. Trends in mental illness and suicidality after Hurricane Katrina. Mol Psychiatry 2008;13:374–84. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Heron-Delaney M, Kenardy J, Charlton E, Matsuoka Y.. A systematic review of predictors of post-traumatic stress disorder (PTSD) for adult road traffic crash survivors. Injury 2013;44:1413–22. [DOI] [PubMed] [Google Scholar]
- 8. Bomyea J, Risbrough V, Lang AJ.. A consideration of select pre-trauma factors as key vulnerabilities in PTSD. Clin Psychol Rev 2012;32:630–41. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Kessler RC, Rose S, Koenen KC. et al. How well can post-traumatic stress disorder be predicted from pre-trauma risk factors? An exploratory study in the WHO World Mental Health Surveys. World Psychiatry 2014;13:265–74. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Asselmann E, Wittchen H-U, Lieb R, Perkonigg A, Beesdo-Baum K.. Incident mental disorders in the aftermath of traumatic events: A prospective-longitudinal community study. J Affect Disord 2018;227:82–89. [DOI] [PubMed] [Google Scholar]
- 11. Kessler RC, McLaughlin KA, Green JG. et al. Childhood adversities and adult psychopathology in the WHO World Mental Health Surveys. Br J Psychiatry 2010;197:378–85. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Lewis SJ, Arseneault L, Caspi A. et al. The epidemiology of trauma and post-traumatic stress disorder in a representative cohort of young people in England and Wales. Lancet Psychiatry 2019;6:247–56. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Walsh K, McLaughlin KA, Hamilton A, Keyes KM.. Trauma exposure, incident psychiatric disorders, and disorder transitions in a longitudinal population representative sample. J Psychiatr Res 2017;92:212–18. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Zlotnick C, Johnson J, Kohn R, Vicente B, Rioseco P, Saldivia S.. Childhood trauma, trauma in adulthood, and psychiatric diagnoses: results from a community sample. Compr Psychiatry 2008;49:163–69. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Debell F, Fear NT, Head M. et al. A systematic review of the comorbidity between PTSD and alcohol misuse. Soc Psychiatry Psychiatr Epidemiol 2014;49:1401–25. [DOI] [PubMed] [Google Scholar]
- 16. Stander VA, Thomsen CJ, Highfill-McRoy RM.. Etiology of depression comorbidity in combat-related PTSD: a review of the literature. Clin Psychol Rev 2014;34:87–98. [DOI] [PubMed] [Google Scholar]
- 17. Forbes D, Nickerson A, Alkemade N. et al. Longitudinal analysis of latent classes of psychopathology and patterns of class migration in survivors of severe injury. J Clin Psychiatry 2015;76:1193–99. [DOI] [PubMed] [Google Scholar]
- 18. Gradus JL, Rosellini AJ, Szentkúti P. et al. Using Danish national registry data to understand psychopathology following potentially traumatic experiences. J Trauma Stress 2022:1–12. https://doi.org/10.1002/jts.22777. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Rosellini AJ, Szentkúti P, Horváth-Puhó E. et al. Latent classes of post-traumatic psychiatric comorbidity in the general population. J Psychiatr Res 2021;136:334–42. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders: DSM-5. 5th edn. Arlington, VA: American Psychiatric Association, 2013. [Google Scholar]
- 21. Kessler RC, Sonnega A, Bromet E, Hughes M, Nelson CB.. Post-traumatic stress disorder in the National Comorbidity Survey. Arch Gen Psychiatry 1995;52:1048–60. [DOI] [PubMed] [Google Scholar]
- 22. Schmidt M, Schmidt SAJ, Adelborg K. et al. The Danish health care system and epidemiological research: from health care contacts to database records. Clin Epidemiol 2019;11:563–91. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Schmidt M, Pedersen L, Sørensen HT.. The Danish Civil Registration System as a tool in epidemiology. Eur J Epidemiol 2014;29:541–49. [DOI] [PubMed] [Google Scholar]
- 24. Baadsgaard M, Quitzau J.. Danish registers on personal income and transfer payments. Scand J Public Health 2011;39:103–05. [DOI] [PubMed] [Google Scholar]
- 25. Petersson F, Baadsgaard M, Thygesen LC.. Danish registers on personal labour market affiliation. Scand J Public Health 2011;39:95–98. [DOI] [PubMed] [Google Scholar]
- 26. Mors O, Perto GP, Mortensen PB.. The Danish Psychiatric Central Research Register. Scand J Public Health 2011;39:54–57. [DOI] [PubMed] [Google Scholar]
- 27. Svensson E, Lash TL, Resick PA, Hansen JG, Gradus JL.. Validity of reaction to severe stress and adjustment disorder diagnoses in the Danish Psychiatric Central Research Registry. Clin Epidemiol 2015;7:235–42. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Munk-Jørgensen P, Mortensen PB.. The Danish Psychiatric Central Register. Dan Med Bull 1997;44:82–84. [PubMed] [Google Scholar]
- 29. Schmidt M, Schmidt SAJ, Sandegaard JL, Ehrenstein V, Pedersen L, Sørensen HT.. The Danish National Patient Registry: a review of content, data quality, and research potential. Clin Epidemiol 2015;7:449–90. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Vest-Hansen B, Riis AH, Christiansen CF.. Registration of acute medical hospital admissions in the Danish National Patient Registry: a validation study. Clin Epidemiol 2013;5:129–33. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Pottegård A, Schmidt SAJ, Wallach-Kildemoes H, Sørensen HT, Hallas J, Schmidt M.. Data Resource Profile: The Danish National Prescription Registry. Int J Epidemiol 2017;46:798.f. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Jiang T, Gradus JL, Rosellini AJ.. Supervised machine learning: a brief primer. Behav Ther 2020;51:675–87. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Galatzer-Levy IR, Ruggles KV, Chen Z.. Data science in the Research Domain Criteria era: relevance of machine learning to the study of stress pathology, recovery, and resilience. Chronic Stress 2018;2:247054701774753. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Rosellini AJ, Monahan J, Street AE. et al. Using administrative data to identify U.S. Army soldiers at high-risk of perpetrating minor violent crimes. J Psychiatr Res 2017;84:128–36. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Kessler RC, Warner CH, Ivany C. et al. ; Army STARRS Collaborators. Predicting suicides after psychiatric hospitalization in US army soldiers: The Army Study to Assess Risk and Resilience in Service members (Army STARRS). JAMA Psychiatry 2015;72:49–57. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Breiman L, Friedman J, Stone CJ, Olshen RA.. Classification and Regression Trees. Monterey, CA: Taylor & Francis, 1984.. [Google Scholar]
- 37. Therneau T, Atkinson B, Ripley B.. rpart: Recursive partitioning and regression trees. R Package Version 2015;4:1–9. [Google Scholar]
- 38. Liaw A, Wiener M.. Classification and regression by random forest. R News 2002;2:18–22. [Google Scholar]
- 39. Huang BFF, Boutros PC.. The parameter sensitivity of random forests. BMC Bioinformatics 2016;17:331. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40. Robin X, Turck N, Hainard A. et al. pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics 2011;12:77. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41. Schultebraucks K, Shalev AY, Michopoulos V. et al. A validated predictive algorithm of post-traumatic stress course following emergency department admission after a traumatic stressor. Nat Med 2020;26:1084–88. [DOI] [PubMed] [Google Scholar]
- 42. Fink DS, Gradus JL, Keyes KM. et al. Subthreshold PTSD and PTSD in a prospective‐longitudinal cohort of military personnel: Potential targets for preventive interventions. Depress Anxiety 2018;35:1048–55. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43. Galea S, Hernán MA.. Win-win: reconciling social epidemiology and causal inference. Am J Epidemiol 2020;189:167–70. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44. Smith ML, Gradus JL.. Psychiatric disorders and risk of infections: early lessons from COVID-19. Lancet Healthy Longev 2020;1:e51–52.e2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45. Gradus JL, Farkas DK, Svensson E, Ehrenstein V, Lash TL, Toft HS.. Post-traumatic stress disorder and gastrointestinal disorders in the Danish population. Epidemiology 2017;28:354–60. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46. Pacella ML, Hruska B, Delahanty DL.. The physical health consequences of PTSD and PTSD symptoms: a meta-analytic review. J Anxiety Disord 2013;27:33–46. [DOI] [PubMed] [Google Scholar]
- 47. Naimi AI, Balzer LB.. Stacked generalization: an introduction to super learning. Eur J Epidemiol 2018;33:459–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Access to the data used for the current study is subject to all appropriate Danish approvals. Please contact Drs Gradus or Sørensen for information about how to initiate that process.