Depression phenotypes in spinal cord injury and impact on post-injury healthcare utilization and cost: Analysis using a large claim database

Abstract

Context/Objective: Depression is the most common psychological comorbidity associated with spinal cord injury (SCI) and affects healthcare utilization and costs. This study aimed to use an International Classification of Disease (ICD) and prescription drug-based depression phenotypes to classify people with SCI, and to evaluate the prevalence of those phenotypes, associated risk factors, and healthcare utilization.

Design: Retrospective Observational Study

Setting: Marketscan Database (2000–2019)

Participants: Individuals with SCI were classified into six ICD-9/10, and prescription drugs defined phenotypes: Major Depressive Disorder (MDD), Other Depression (OthDep), Antidepressants for Other Psychiatric Conditions (PsychRx), Antidepressants for non-psychiatric condition (NoPsychRx), Other Non-depression Psychiatric conditions only (NonDepPsych), and No Depression (NoDep). Except for the latter, all the other groups were referred to as “depressed phenotypes”. Data were screened for 24 months pre- and 24 months post-injury depression.

Interventions: None

Outcome Measures: Healthcare utilization and payments

Results: There were 9,291 patients with SCI classified as follows: 16% MDD, 11% OthDep, 13% PsychRx, 13% NonPsychRx, 14% NonDepPsych, 33% NoDep. Compared with the NoDep group, the MDD group was younger (54 vs. 57 years old), predominantly female (55% vs. 42%), with Medicaid coverage (42% vs. 12%), had increased comorbidities (69% vs. 54%), had fewer traumatic injuries (51% vs. 54%) and had higher chronic 12-month pre-SCI opioid use (19% vs. 9%) (all P < 0.0001). Classification into a depressed phenotype before SCI was found to be significantly associated with depression phenotype post-SCI, as evidenced by those who experienced a negative change (37%) vs. a positive change (15%, P < 0.0001). Patients in the MDD cohort had higher healthcare utilization and associated payments at 12 and 24 months after SCI.

Conclusion: Increasing awareness of psychiatric history and MDD risk factors may improve identifying and managing higher-risk patients with SCI, ultimately optimizing their post-injury healthcare utilization and cost. This method of classifying depression phenotypes provides a simple and practical way to obtain this information by screening through pre-injury medical records.

Introduction

In addition to the myriad of physical impairments seen after spinal cord injury (SCI) – e.g. respiratory impairment, neurogenic bowel/bladder, urinary tract infections, etc. – depressive disorders are among the most common psychological comorbidities following SCI, thus compounding the effects of these physical complications (1, 2).

Depression affects almost one-quarter of the SCI population, with rates three times higher than the general population (3). Its prevalence has been estimated to be approximately 22% (1). Socio-demographic characteristics, such as sex (4), age (5), race/ethnicity (5), education (4), psychological and mental health factors, such as suicidal thoughts or history of suicide attempt, and injury characteristics such as level and severity of injury (6–8) have been identified to be risk factors for depression after SCI. In the general population, depression has been a significant contributor to increasing healthcare expenditures, accounting for around $43 billion in excess medical costs, and has been cited as a leading cause of disability in the United States (9). In veterans with SCI, depression has resulted in the highest utilization of health care and medication (10).

Depression is routinely diagnosed through symptoms screening questionnaires (11, 12), for which data are stored in electronic medical records and clinical notes. However, the lack of screening standardization (13), the disagreement of questionnaire responses with the International Classification of Disease (ICD) diagnosis codes (14), and the limited generalizability of natural language processing analytical methods to de-identified data make it difficult to identify depression profiles in medical records. This is particularly important in SCI, where the care team may only rely on the individual's pre-injury medical records to understand and factor prior health profiles in post-injury health care plans. In these situations, to check for depression in pre-injury medical data, many prefer to look at ICD diagnosis codes or check the evidence of antidepressant prescriptions. However, some researchers have shown that used alone, ICD codes have low sensitivity and precision compared to when combined with other sources of information (15), and antidepressants are not exclusive to depression, as they are used for many mental and non-mental conditions (16–19). The depression phenotypes were developed and validated by Ingram et al. (9) to overcome these limitations. They used ICD 9th revision (ICD-9) and medication orders to classify patients into one of five mutually exclusive, ordinal groups: major depressive disorder, other depression, antidepressants for other psychiatric conditions, antidepressants for physical conditions, and no depression. In their paper, Ingram and colleagues showed that using this ordinal classification algorithm which takes into account both ICD diagnosis codes and medications orders together leads to a better simple grouping than each one alone, ranking individuals in patterns of varying outcomes. It is essential to evaluate the prevalence of different depression phenotypes in the SCI population to determine if it is similar to the documented epidemiology of depression and to use them in evaluating depression-associated outcomes.

This study aimed (1) to characterize the prevalence of depression after SCI using the above-mentioned ICD-defined depression phenotypes, to evaluate the factors associated with depression post-injury, and (3) to correlate these phenotypes with post-SCI health care utilization and cost. We hypothesized that ICD-defined depression phenotypes would be similar to published statistics of depression in SCI which would validate this algorithm in this study population, that socio-demographic factors would be associated with the incidence of depression after injury, and that those with depression would have higher healthcare utilization and costs after injury.

Methods

Data source

MarketScan data, 2000–2019, from Truven Health Analytics – IBM Watson Health (20) were used for this project. These data, grouped into Commercial Claims and Encounter, Medicaid, and Medicare Supplemental, are from claims covered by large employers, health plans, and government and public organizations. Researchers have obtained it for a fee and used it in medical field research since 1990 (20). We have a neurological/neurosurgical custom database. The data is longitudinal, with a unique identifier helping to track the trajectory of each included individual as they interact with the healthcare system (hospital admissions, outpatient visits and services, outpatient prescribed and inpatient medication use, laboratory tests, and dental care procedures). Additional available data include short- and long-term disability, workers’ compensations, and self-reported data related to medical conditions. Medical conditions and procedures are recorded using diagnostic and procedure codes (International Classification of Diseases, 9th and 10th Revisions, Clinical Modification [ICD-9-CM before October 2015, ICD-10 after October 2015], Current Procedural Terminology, 4th edition [CPT-4]). Medications are coded with National Drug Code (NDC) and the medication names. Demographics are also documented in the database and enrollment information (beginning and end). Given the administrative nature of this project, the study qualified for expedited approval by the University of Louisville Institutional Review Board (IRB #10.0559).

Study cohort

From the inpatient database, cases of SCI (ICD-9-CM codes 806, 952, ICD-10-CM codes S14.0, S14.1, S24.0, S24.1, S34.0, S34.1) were extracted. To ensure an incidence cohort, 24 months of look-back data were screened to exclude anyone claiming SCI or tetraplegia, or paraplegia (ICD-9-CM: 3440, 3441, ICD-10-CM: G82). Those enrolled in their insurance for less than 24 months before injury were excluded. Outcomes were evaluated within 24 months post-index hospitalization. Therefore, those enrolled in their health insurance for less than 24 months post-injury were also excluded. Only individuals 18 years or older were retained.

Pre- and post-injury enrollment time

Pre-injury enrollment time was calculated as the difference between the health insurance enrollment start date (or first claim in the data) and the injury hospitalization admission date. Post-injury enrollment time was calculated as the difference between the injury hospitalization discharge date and the insurance enrollment end date (or last claim in the data).

Definition of depression phenotypes

We used the method proposed by Ingram et al. (9), defining five depression phenotypes and extended this with one additional phenotype. The term “depression phenotype” represents healthcare and medication utilization indicative of a depression category. Ingram et al. defined 5 phenotypes using ICD-9 codes and medication orders using Electronic Health Records (EHR) of the Geisinger Health System: Major Depressive Disorder (MDD), Other Depression (OthDep), Antidepressants for Other Psychiatric Conditions (PsychRx), Antidepressants for Physical Condition (non-psychiatric condition) (NoPsychRx), and No Depression (NoDep). We used the same ICD-9 codes, adding their ICD-10 equivalents and the same prescription medications as this paper. To the 5 categories, we added another psychiatric category: Other Non-Depression Psychiatric condition only (NonDepPsych). Therefore, the current analysis has 6 depression phenotypes: 1-MDD, 2-OthDep, 3-PsychRx, 4-NoPsychRx, 5-NonDepPsych, and 6-NoDep. We have used this method of 6 classes in our previous project (21). Groups 1–5 were termed depressed phenotypes, and “no depression” was used interchangeably with “not depressed” to refer to group 6. We looked into 24 months of claims data before injury (look-back period) to define pre-injury phenotype classification and 24 months of claims data after injury (follow-up period) to define post-injury phenotype classification. A schema, the ICD-9 and ICD-10 for depression and psychiatric conditions, and the therapeutic classification for antidepressants are provided in Figure 1.

Figure 1.

Schematic representation of ICD-9 and ICD-10 based depression phenotypes. Step 1 was to classify people into Major Depressive Disorder (MDD) vs not if they were diagnosed with the listed ICD-9 and ICD-10 in their 24-months records. Step 2 was to take those who were not in MDD and classify them into other depression (OthDep) vs not using the listed ICD-9 and ICD-10. Step 3 was to take those in the no OtherDep and classify them into antidepressants use with non-depression psychiatric condition (PsychRx), antidepressants use without non-depression psychiatric conditions (NoPsychRx), non-depression psychiatric conditions without antidepressants use (NoDepPsych), and the remaining into no depression, no non-depression psychiatric conditions, and no antidepressants (NoDep).

Individual characteristics

Demographics, injury characteristics, medical conditions, and 12-month history of opioid use were noted at the time of injury. Age, sex, and insurance type (Commercial, Medicare, Medicaid) were documented at the SCI hospitalization admission. The burden of comorbidities present at SCI hospitalization was accounted for by the Elixhauser score (22) using Quan et al.’s adaptation to ICD-9 and ICD-10 codes (23). In addition, tobacco use, osteoporosis, hypertension, congestive heart failure (CHF), chronic obstructive pulmonary disease (COPD), myocardial infarction (MI), diabetes, and obesity were screened from 3 months prior to the injury, using ICD-9 and ICD-10 (Table 1). The level of injury (cervical, thoracic, lumbar/sacral) was defined from the SCI ICD-9-CM and ICD-10 codes’ decimal digits. The type of injury (traumatic, non-traumatic) was delineated from the inclusion of ICD-9/ICD-10 codes. Medication files were screened 12 months before the injury for opioid use (Table 2). The use was categorized into none, 1–10 prescriptions, or more than ten (often classified as chronic) (24–27).

Table 1.

ICD-9 and ICD-10 for individual comorbidities.

Diagnosis	ICD-9 code	ICD-10 code
Tobacco	3051, V1582, 98984, 6490	F17, O9933, T652, Z87891
Osteoporosis	7330, V1781,7313, V8281	M810, M818, Z8262, M8970, Z13820
Hypertension	401–405	I10, I11, I12, I13, I15
CHF	39891, 40201, 40211, 40291, 40401, 40403, 40411, 40413, 40491, 40493, 4254, 4255, 4257, 4258, 4259, 428	I099, I110, I130, I132, I255, I420, I425-I429, I43, I50, P290
COPD	4168, 4169, 490-496, 500-505, 5064, 5081, 5088	I278, I279, J40-J47, J60-J67, J684, J701, J703
MI	410, 411, 412, 4297, 9971, 9980	I20-I24, I51, I252, I977, I978, T811
Diabetes	250	E08, E09, E10, E11, E13
Obesity	2780	E66

Table 2.

List of opioid prescription medications screened.

buprenorphine	butorphanol	Codeine
dihydrocodeine	fentanyl	Hydrocodone
hydromorphone	levorphanol	Meperidine
methadone	morphine	Nalbuphine
Opium	oxycodone	Oxymorphone
pentazocine	propoxyphene	Tapentadol
tramadol	buprenorphine	Naltrexone

Outcomes

The outcomes of interest were the risk of depression defined by ICD-derived phenotypes after SCI, the healthcare resources utilization, and associated payments. Emergency room (ER) visits, hospital readmissions, outpatient resources use, and medications were totaled over 12 and 24 months after discharge. Their payments, accounting for all sources (insurance, co-pays, co-insurance), were also cumulated over the same periods. They were adjusted to 2019 US dollars using the medical component of the consumer price index, accessible through the US Bureau of Labor Statistics.

Statistical analysis

Summary statistics were used to present characteristics and outcomes. Comparison across depression phenotypes was performed with the Wilcoxon Rank Sum test for continuous variables and the Chi-Square test for categorical variables. The difference in characteristics was also evaluated with Cohen’s (28) and Cohen’s h (29) effect sizes. Per Sawilowsky’s (30) extension of Cohen’s criteria (28), effect sizes are classified as very small (<0.2), small (0.2–0.49), medium (0.5–0.79), large (0.8–1.2), very large (1.2–1.99), and huge (>2.0). The risk of depression after SCI was evaluated with generalized logistic regression in a model including all characteristics as independent variables. The effect was presented with Odds Ratio (OR) with an associated 95% confidence interval (95% CI). Pre–post SCI depression category transition was evaluated with a generalized McNemar test, and the comparison between a positive change (towards less severe depression), a negative change (towards more severe depression), and a no change (same depression category) was performed by the sign test. The outcomes were compared between the six groups to evaluate the healthcare utilization and cost associated with depression. First, observed bias was eliminated by balancing all the characteristics using the inverse probability of treatment weight (IPTW) technique (31). In this method, a propensity score (PS) was calculated using a generalized logistic regression in which post-SCI depression phenotype was the outcome and all the characteristics were the factors. For each individual, the weight was calculated as the stabilized (32, 33) inverse of the propensity score of the depression group one belongs to; i.e. $W_i={\displaystyle \frac{n_i}{n_{\mathrm{total}}}\times {\displaystyle \frac{1}{\mathrm{PS}}}}$ where i = 1,2,3,4,5,6 representing the phenotype, n_i = the sample size of the corresponding phenotype group. The balance after IPTW was evaluated with p-values from the IPTW-weighted Kruskal Wallis test for continuous variables and the Chi-square test for categorical variables. The IPTW created a pseudo cohort where the characteristics are comparable between the groups. Next, healthcare utilization and payments were analyzed with IPTW-weighted generalized linear regressions: Logistic regression was used for ER visits and hospital readmissions; negative binomial regression was used for the number of ER visits, readmissions, outpatient services, medications; and quantile regression to the median was used for hospital length of stay, and payments. All tests were two-sided, with a significance level of 5%. Analyses were performed in SAS 9.4 (SAS Institute Inc, Cary, NC, USA).

Results

Demographics, medical conditions, injury characteristics, and pre-injury opioid use

The study cohort was composed of 9291 individuals with an average age of 56 years old (SD: 20) and 48% female. At injury hospitalization, 45% had commercial insurance, 30% had Medicare Supplemental, and 25% had Medicaid. A total of 70% had at least one comorbidity, with 39% having a prevalence of hypertension, 17% diabetes, and 15% COPD. Traumatic injuries made up 56% of the cases, most in the cervical region (48%). Look back data screening revealed that 47% had used opioids in the previous 12 months, with 19% being dependent users. Twenty-four months after injury, the breakdown of phenotypes was 16% MDD, 11% OthDep, 13% PsychRx, 13%NonPsychRx, 14% NonDepPsych, and 33% NoDep. Individuals in those depression phenotypes were statistically different at the time of injury. Those with no depression were older, consisting of fewer females and more individuals with Commercial insurance relative to those with MDD. Also, this group had a lower rate of comorbidities and were composed of more traumatic injuries, but a similar level of injury compared to MDD. More people in NoDep cohort had no opioid use in the previous 12 months than MDD. Characteristics are summarized in Table 3.

Table 3.

Characteristics of individuals with spinal cord injury (SCI) at the time of injury hospitalization. The cohort is stratified by international classification of disease (ICD)-derived phenotypes. MDD: Major depressive disorder, OthDep: Other depression, PsychRx: Antidepressants for Other Psychiatric Condition, NonPsychRx: Antidepressants for non-Psychiatric Condition, NonDepPsychy: Other Non-Depression Psychiatric Condition only, NoDep: No Depression. Nearly all the characteristics were statistically different between the 6 groups. The NoDep group were older had fewer females, had the highest proportion of commercial insurance, and had the highest proportion without comorbidities. Traumatic injuries made up 54% of this group where most levels of injury were cervical. In this group, 62% did not use opioids in the previous 12 months. On the other end of phenotype spectrum, the MDD group were younger, had more females, and had the highest proportion of Medicaid insurance. There was an even split of traumatic and non-traumatic injuries, and most injuries were cervical. Those with at least one opioid were 47% with 30% having more than 11 prescription drugs in the previous year. Effect sizes of the most different groups were added for each characteristic. Cohen’d was used for age and Cohen’s h was used for all categorical variables. Effect sizes are classified as very small (<0.2), small (0.2–0.49), medium (0.5–0.79), large (0.8–1.2), very large (1.2-1.99), and huge (>2.0).

			Total	MDD	OthDep	PsychRx	NonPsychRx	NonDepPsych	NoDep	P-value	Effect size
			n = 9291 (100%)	n = 1513 (16%)	n = 1013 (11%)	n = 1176 (13%)	n = 1180 (13%)	n = 1335 (14%)	n = 3074 (33%)
Demographics	Age	Mean (SD)	56 (20)	54 (19)	54 (20)	56 (21)	60 (18)	56 (21)	57 (20)	<.0001	8.36
		Median (Q1, Q3)	57 (43, 72)	55 (41, 66)	55 (40, 70)	56 (41, 73)	60 (50, 74)	56 (43, 72)	58 (45, 74)
		Range, min-max	18–99	18–96	18–97	18–99	18–95	18–98	18–97
	Sex	Female, n (%)	4471 (48%)	832 (55%)	571 (56%)	624 (53%)	620 (53%)	546 (41%)	1278 (42%)	<.0001	0.3
	Insurance	Commercial, n (%)	4182 (45%)	525 (35%)	365 (36%)	484 (41%)	600 (51%)	523 (39%)	1685 (55%)	<.0001	0.08
		Medicaid, n (%)	2324 (25%)	639 (42%)	399 (39%)	334 (28%)	117 (10%)	462 (35%)	373 (12%)		0.77
		Medicare, n (%)	2785 (30%)	349 (23%)	249 (25%)	358 (30%)	463 (39%)	350 (26%)	1016 (33%)		0.35
Medical conditions	Elixhauser	0, n (%)	2802 (30%)	303 (20%)	216 (21%)	333 (28%)	395 (33%)	364 (27%)	1191 (39%)	<.0001	0.4
		1, n(%)	2647 (28%)	408 (27%)	272 (27%)	339 (29%)	377 (32%)	359 (27%)	892 (29%)		0.11
		2, n(%)	1911 (21%)	368 (24%)	237 (23%)	263 (22%)	220 (19%)	278 (21%)	545 (18%)		0.15
		3+, n(%)	1931 (21%)	434 (29%)	288 (28%)	241 (20%)	188 (16%)	334 (25%)	446 (15%)		0.34
	Comorbidities	Tobacco use, n (%)	1095 (12%)	322 (21%)	144 (14%)	147 (13%)	53 (4%)	244 (18%)	185 (6%)	<.0001	0.55
		Osteoporosis,n (%)	578 (6%)	90 (6%)	81 (8%)	85 (7%)	75 (6%)	71 (5%)	176 (6%)	0.0533	0.12
		Hypertension, n (%)	3669 (39%)	684 (45%)	423 (42%)	461 (39%)	464 (39%)	533 (40%)	1104 (36%)	<.0001	0.18
		CHF, n (%)	677 (7%)	123 (8%)	82 (8%)	104 (9%)	92 (8%)	92 (7%)	184 (6%)	0.0093	0.11
		COPD, n (%)	1395 (15%)	304 (20%)	182 (18%)	216 (18%)	161 (14%)	204 (15%)	328 (11%)	<.0001	0.25
		MI, n (%)	359 (4%)	71 (5%)	37 (4%)	47 (4%)	42 (4%)	59 (4%)	103 (3%)	0.2556	0.1
		Diabetes, n (%)	1534 (17%)	288 (19%)	175 (17%)	190 (16%)	220 (19%)	220 (16%)	441 (14%)	0.0006	0.13
		Obesity, n (%)	435 (5%)	104 (7%)	61 (6%)	45 (4%)	35 (3%)	69 (5%)	121 (4%)	<.0001	0.19
		1 + of the above, n (%)	5629 (61%)	1044 (69%)	651 (64%)	726 (62%)	698 (59%)	844 (63%)	1666 (54%)	<.0001	0.31
Injury characteristics	SCI Type	Non-traumatic, n (%)	4104 (44%)	743 (49%)	388 (38%)	477 (41%)	479 (41%)	618 (46%)	1399 (46%)	<.0001	0.22
		Traumatic, n (%)	5187 (56%)	770 (51%)	625 (62%)	699 (59%)	701 (59%)	717 (54%)	1675 (54%)		0.22
	SCI Level	Cervical, n (%)	4484 (48%)	726 (48%)	486 (48%)	571 (49%)	527 (45%)	684 (51%)	1490 (48%)	0.0266	0.06
		Thoracic, n (%)	2052 (22%)	341 (23%)	238 (23%)	251 (21%)	307 (26%)	288 (22%)	627 (20%)		0.07
		Lumbar/Sacral, n (%)	1750 (19%)	287 (19%)	186 (18%)	219 (19%)	218 (18%)	221 (17%)	619 (20%)		0.08
		Unknown, n (%)	1005 (11%)	159 (11%)	103 (10%)	135 (11%)	128 (11%)	142 (11%)	338 (11%)		0.03
Opioid use	12-months prior to injury	0, n (%)	4918 (53%)	691 (46%)	499 (49%)	484 (41%)	488 (41%)	847 (63%)	1909 (62%)	<.0001	0.44
		1-10, n(%)	2645 (28%)	365 (24%)	265 (26%)	386 (33%)	416 (35%)	314 (24%)	899 (29%)		0.24
		11+, n(%)	1728 (19%)	457 (30%)	249 (25%)	306 (26%)	276 (23%)	174 (13%)	266 (9%)		0.55

CHF = Congestive Heart Failure, COPD = Chronic Obstructive Pulmonary Disease, MI = Myocardial Infarction.

Prevalence of depression

Within 24 months after injury, 6217 individuals (67%) had depression as defined by the 5 categories of depression phenotypes. MDD and Other depression, the phenotypes defined by ICD codes of depression, comprised 26%.

Transition of depression phenotypes pre-post injury

A total of 4534 (49%) had no depression or other psychiatric condition before injury (Figure 2), and only 55% of them (n = 2477) remained not depressed after injury. MDD was diagnosed in 924 individuals (10%), and 64% (n = 592) remained depressed. Individuals were more likely to remain within the same group (Figure 3). However, there were transitions in both directions after the injury. The distribution of these phenotypes shifted pre- to post-injury with more major depressive disorders, other depression, and antidepressants use after injury than prior (Figure 4). There was more likelihood of becoming depressed than not, as evidenced by those who experienced a negative change (37%) vs. a positive change (15%, P < 0.0001, Figure 2).

Figure 2.

Transition between depression from 24 months pre-injury to 24 months post-injury. MDD: Major depressive disorder, OthDep: Other depression, PsychRx: Antidepressants for Other Psychiatric Condition, NonPsychRx: Antidepressants for non-Psychiatric Condition, NonDepPsych: Other Non-Depression Psychiatric Condition only, NoDep: No Depression. Columns represent phenotypes sample size prior to injury. Rows represent post injury. Cells contain the number of people leaving each phenotype and where they end up. The percentage within each cell quantifies the number of individuals on that trajectory, and they are calculated based on the sample size in that column. The shade color of the cell summarizes which type of change it is (light blue for remaining in the same phenotype, light brown for transitioning to a more severe phenotype, and light green for transitioning to a less severe phenotype). On the bottom, the text statistically summarizes those transitions, and provides a statistical comparison. This figure shows that 37% transition to a more severe phenotype compared to 15% that transition to a less severe phenotype (p < .0001). About half of the individuals (52%) remain in the same phenotype.

Figure 3.

Post-SCI depression within each pre-SCI depression group. MDD: Major depressive disorder, OthDep: Other depression, PsychRx: Antidepressants for Other Psychiatric Condition, NonPsychRx: Antidepressants for non-Psychiatric Condition, NonDepPsych: Other Non-Depression Psychiatric Condition only, NoDep: No Depression. This figure portrays the different changes in depression phenotypes from the perspective of pre-injury status. Each pie represents a pre-injury phenotype; each part within each pie represents post-injury phenotypes. The figure reveals that most people remain in the same phenotype as reflected by the largest portion within each cluster. However, we also see that there are transitions to other phenotypes as well.

Figure 4.

Comparative distribution of the depression phenotypes pre- and post-spinal cord injury. MDD: Major depressive disorder, OthDep: Other depression, PsychRx: Antidepressants for Other Psychiatric Condition, NonPsychRx: Antidepressants for non-Psychiatric Condition, NonDepPsych: Other Non-Depression Psychiatric Condition only, NoDep: No Depression. This figure shows that people were more likely to be depressed after injury: the bars in the depressed phenotypes are longer post injury than prior, and the bar of no depression is shorter post injury than prior.

Risk Factors of Depression after injury

Factors associated with post-injury depression (depression phenotypes vs. no depression) were prior depression, younger age, presence of comorbidities, type of injury (traumatic versus non-traumatic), cervical injury, and pre-injury chronic opioid use (Table 4).

Table 4.

Risk of post-SCI depression in relation with pre-injury depression adjusted for demographics, medical conditions, injury characteristics, pre-injury opioid use. MDD: Major depressive disorder, OthDep: Other depression, PsychRx: Antidepressants for Other Psychiatric Condition, NonPsychRx: Antidepressants for non-Psychiatric Condition, NonDepPsych: Other Non-Depression Psychiatric Condition only, NoDep: No Depression. Each column of the table contains the odds ratio (OR) and 95% confidence intervals (95%CI) evaluating the risk factors (in the rows) predicting each depression phenotype vs no depression. The table shows that, overall, being depressed prior to injury had higher odds of being depressed post injury. Other significant factors were younger age, female sex, higher comorbidities, traumatic injury vs non-traumatic injuries, and lumbar/sacral injuries. These had significant OR per 95% CI excluding 1.

			MMD vs NoDep	OthDep vs NoDep	PsychRx vs NoDep	NonPsychRx vs NoDep	NonDepPsych vs NoDep
			OR (95% CI)	OR (95% CI)	OR (95% CI)	OR (95% CI)	OR (95% CI)
Pre-injury depression	Depression phenotypes	MDD vs NoDep	126.09 (85.361, 186.254)	28.762 (18.825, 43.943)	16.133 (10.468, 24.864)	6.149 (3.768, 10.034)	5.735 (3.637, 9.043)
		OthDep vs NoDep	19.195 (13.501, 27.291)	30.657 (21.757, 43.199)	11.287 (7.821, 16.289)	6.676 (4.512, 9.879)	4.306 (2.952, 6.281)
		PsychRx vs NoDep	20.852 (14.984, 29.017)	15.571 (10.985, 22.073)	27.653 (20.265, 37.735)	9.588 (6.858, 13.403)	3.986 (2.769, 5.737)
		NonPsychRx vs NoDep	7.795 (5.886, 10.325)	9.018 (6.816, 11.931)	9.701 (7.555, 12.456)	14.778 (11.864, 18.407)	1.187 (0.831, 1.695)
		PsychOnly vs NoDep	3.377 (2.669, 4.271)	3.104 (2.422, 3.978)	2.972 (2.356, 3.749)	1.025 (0.754, 1.392)	3.948 (3.301, 4.723)
Demo-graphics	Age	1 year older	0.988 (0.982, 0.993)	0.987 (0.981, 0.993)	0.988 (0.982, 0.994)	0.992 (0.986, 0.999)	1.004 (0.998, 1.009)
	Sex	Female vs Male	1.217 (1.041, 1.423)	1.291 (1.094, 1.524)	1.17 (1.002, 1.366)	1.082 (0.927, 1.263)	0.918 (0.795, 1.061)
	Insurance	Medicaid vs Commercial	3.038 (2.494, 3.701)	3.038 (2.472, 3.734)	2.25 (1.837, 2.758)	0.896 (0.701, 1.146)	2.702 (2.248, 3.247)
		Medicare vs Commercial	1.518 (1.19, 1.937)	1.358 (1.05, 1.756)	1.407 (1.108, 1.787)	1.168 (0.923, 1.479)	1.086 (0.872, 1.353)
Medical conditions	Elixhauser	1 vs 0	1.903 (1.543, 2.347)	1.732 (1.389, 2.159)	1.433 (1.177, 1.744)	1.433 (1.187, 1.729)	1.171 (0.974, 1.407)
		2 vs 0	2.321 (1.835, 2.935)	2.247 (1.758, 2.873)	1.656 (1.321, 2.076)	1.345 (1.07, 1.689)	1.325 (1.071, 1.639)
		3 + vs 0	3.156 (2.441, 4.081)	3.054 (2.334, 3.996)	1.804 (1.396, 2.331)	1.392 (1.071, 1.808)	1.827 (1.446, 2.308)
	Comorbidities	Tobacco use, yes vs no	1.775 (1.358, 2.321)	1.198 (0.892, 1.607)	1.298 (0.974, 1.728)	0.798 (0.559, 1.14)	1.992 (1.549, 2.56)
		Osteoporosis, yes vs no	0.77 (0.549, 1.08)	1.107 (0.794, 1.544)	1.01 (0.733, 1.393)	0.851 (0.612, 1.183)	0.849 (0.616, 1.171)
		Hypertension, yes vs no	0.887 (0.712, 1.104)	0.895 (0.709, 1.131)	0.905 (0.725, 1.128)	0.955 (0.762, 1.196)	0.913 (0.744, 1.121)
		CHF, yes vs no	0.899 (0.666, 1.214)	0.941 (0.685, 1.293)	1.162 (0.866, 1.56)	1.028 (0.759, 1.394)	0.865 (0.649, 1.154)
		COPD, yes vs no	0.956 (0.761, 1.203)	0.92 (0.721, 1.175)	1.126 (0.893, 1.42)	0.98 (0.766, 1.254)	1 (0.802, 1.248)
		MI, yes vs no	1.069 (0.725, 1.575)	1.02 (0.665, 1.563)	1.071 (0.721, 1.592)	1.013 (0.675, 1.519)	1.149 (0.807, 1.638)
		Diabetes, yes vs no	0.853 (0.682, 1.068)	0.821 (0.645, 1.044)	0.857 (0.681, 1.078)	1.058 (0.844, 1.325)	0.95 (0.769, 1.174)
		Obesity, yes vs no	0.993 (0.701, 1.406)	1.032 (0.713, 1.493)	0.766 (0.518, 1.132)	0.696 (0.457, 1.06)	1.052 (0.753, 1.468)
		1 + of the above vs none	0.956 (0.738, 1.239)	0.895 (0.68, 1.176)	0.848 (0.656, 1.097)	0.88 (0.676, 1.144)	0.906 (0.714, 1.15)
Injury chara-cteristics	SCI Type	Traumatic vs non-traumatic	1.132 (0.96, 1.335)	1.698 (1.422, 2.027)	1.567 (1.328, 1.85)	1.337 (1.133, 1.577)	1.15 (0.988, 1.338)
	SCI Level	Thoracic vs Cervical	1.214 (0.994, 1.484)	1.012 (0.821, 1.247)	0.904 (0.74, 1.105)	1.226 (1.009, 1.489)	1.103 (0.916, 1.327)
		Lumbar/Sacral vs Cervical	0.806 (0.655, 0.992)	0.696 (0.558, 0.867)	0.679 (0.552, 0.834)	0.791 (0.644, 0.971)	0.785 (0.647, 0.953)
		Unknown vs Cervical	0.908 (0.7, 1.178)	0.964 (0.729, 1.273)	1.045 (0.811, 1.345)	0.996 (0.772, 1.285)	1.02 (0.806, 1.29)
Opioid use	12-months pre injury	1–9 vs 0	0.929 (0.775, 1.114)	1.028 (0.851, 1.243)	1.46 (1.229, 1.735)	1.405 (1.187, 1.663)	0.847 (0.72, 0.997)
		10 + vs 0	1.734 (1.392, 2.16)	1.749 (1.386, 2.208)	2.174 (1.742, 2.714)	2.156 (1.725, 2.693)	1.138 (0.906, 1.43)

CHF = Congestive Heart Failure, COPD = Chronic Obstructive Pulmonary Disease, MI = Myocardial Infarction.

Association of Depression with Healthcare utilization and cost post-injury

All characteristics were statistically comparable in the pseudo-cohort balanced with IPTW (Table 5). During the index hospitalization, those who ended up in the antidepressant use groups (PsychRx and NonPsychRx) had the most extended stay (Table 6). Depression and antidepressants use groups had 1–2 days longer (P < .0001), corresponding to an additional cost of over $6000 for the OthDep and PsychRx groups (P < 0.05) compared to the no depression group, and fewer individuals were discharged home (P < .0001). In the first year following SCI, compared to those in the NoDep group, those in the MDD group had approximately double the ER visit rate (44% vs. 25%), the number of ER visits (12 vs. 6), and payments ($969 vs. 495) per person who had at least one visit; and higher hospitalization rate (50% vs. 23%) and total hospitalization payments for those who were readmitted at least once ($28,362 vs. $22,899), number of outpatient services (135 vs. 65) and their payments ($17146 vs. $7369), and number of medications (45 vs. 19) and their payments ($2825 vs. $1100). Post-discharge cumulative payments were about three times for the MDD group than the NoDep group ($34227 vs. $12474); combined index and post-discharge payments were about $28,000 more. Similarly, with a few exceptions, the other groups (OthDep, PsychRx, NonPsychRx, and NonDepPsych) had 1.2–1.8 higher utilization. Similar results were noted 24 months after injury (Table 6, Figure 5).

Table 5.

Individual characteristics in a pseudo-cohort analytically created with Inverse Probability of Treatment Weight (IPTW) balancing all the factors across the ICD-derived phenotypes. MDD: Major depressive disorder, OthDep: Other depression, PsychRx: Antidepressants for Other Psychiatric Condition, NonPsychRx: Antidepressants for non-Psychiatric Condition, NonDepPsych: Other Non-Depression Psychiatric Condition only, NoDep: No Depression. The table shows that all the characteristics were balanced (all p-values > 0.05) creating statistically similar groups that are then used for outcomes comparisons.

			MDD	OthDep	PsychRx	NonPsychRx	NonDepPsych	NoDep	P-value
			n = 1514 (16%)	n = 1008 (11%)	n = 1179 (13%)	n = 1199 (13%)	n = 1336 (14%)	n = 3066 (33%)
Demographics	Age	Mean (SD)	56 (19)	57 (20)	56 (21)	55 (20)	56 (21)	57 (20)	0.148
		Median (Q1, Q3)	57 (44, 71)	58 (44, 73)	57 (42, 73)	56 (42, 71)	57 (42, 73)	58 (43, 74)
		Range, min-max	18–96	18–97	18–99	18–95	18–98	18–97
	Sex	Female, n (%)	733 (48%)	489 (49%)	568 (48%)	547 (46%)	639 (48%)	1521 (50%)	0.320
	Insurance	Commercial, n (%)	682 (45%)	434 (43%)	517 (44%)	531 (44%)	596 (45%)	1348 (44%)	0.908
		Medicaid, n (%)	376 (25%)	257 (25%)	294 (25%)	318 (27%)	336 (25%)	747 (24%)
		Medicare, n (%)	456 (30%)	317 (31%)	368 (31%)	351 (29%)	403 (30%)	971 (32%)
Medical conditions	Elixhauser	0, n (%)	438 (29%)	299 (30%)	365 (31%)	344 (29%)	404 (30%)	918 (30%)	0.999
		1, n(%)	440 (29%)	289 (29%)	335 (28%)	340 (28%)	373 (28%)	879 (29%)
		2, n(%)	318 (21%)	211 (21%)	236 (20%)	253 (21%)	280 (21%)	631 (21%)
		3+, n(%)	318 (21%)	209 (21%)	242 (21%)	262 (22%)	279 (21%)	638 (21%)
		Tobacco use, n (%)	173 (11%)	119 (12%)	136 (12%)	161 (13%)	158 (12%)	342 (11%)	0.492
	Comorbidities	Osteoporosis,n (%)	96 (6%)	63 (6%)	70 (6%)	72 (6%)	86 (6%)	218 (7%)	0.684
		Hypertension, n (%)	608 (40%)	402 (40%)	460 (39%)	500 (42%)	529 (40%)	1215 (40%)	0.821
		CHF, n (%)	115 (8%)	72 (7%)	85 (7%)	85 (7%)	99 (7%)	237 (8%)	0.972
		COPD, n (%)	226 (15%)	156 (15%)	178 (15%)	158 (13%)	204 (15%)	441 (14%)	0.620
		MI, n (%)	58 (4%)	47 (5%)	47 (4%)	41 (3%)	49 (4%)	122 (4%)	0.783
		Diabetes, n (%)	257 (17%)	165 (16%)	193 (16%)	196 (16%)	212 (16%)	490 (16%)	0.9700
		Obesity, n (%)	71 (5%)	49 (5%)	55 (5%)	73 (6%)	68 (5%)	134 (4%)	0.331
		1 + of the above, n (%)	921 (61%)	618 (61%)	706 (60%)	729 (61%)	801 (60%)	1886 (62%)	0.901
Injury characteristics	SCI Type	Non-traumatic, n (%)	658 (43%)	456 (45%)	513 (44%)	502 (42%)	582 (44%)	1341 (44%)	0.766
		Traumatic, n (%)	856 (57%)	552 (55%)	666 (56%)	697 (58%)	754 (56%)	1725 (56%)
	SCI Level	Cervical, n (%)	738 (49%)	492 (49%)	555 (47%)	593 (49%)	650 (49%)	1489 (49%)	0.999
		Thoracic, n (%)	331 (22%)	217 (22%)	272 (23%)	257 (21%)	300 (22%)	692 (23%)
		Lumbar/Sacral, n (%)	287 (19%)	188 (19%)	221 (19%)	227 (19%)	250 (19%)	557 (18%)
		Unknown, n (%)	157 (10%)	110 (11%)	131 (11%)	123 (10%)	137 (10%)	329 (11%)
Opioid use	12-months prior to injury	0, n (%)	795 (53%)	532 (53%)	623 (53%)	650 (54%)	705 (53%)	1647 (54%)	0.997
		1-9, n(%)	439 (29%)	287 (28%)	337 (29%)	327 (27%)	380 (28%)	867 (28%)
		10+, n(%)	280 (18%)	190 (19%)	218 (19%)	223 (19%)	251 (19%)	551 (18%)

CHF = Congestive Heart Failure, COPD = Chronic Obstructive Pulmonary Disease, MI = Myocardial Infarction.

Table 6.

Healthcare utilization and payments compared between depression phenotype groups in the Inverse Probability of Treatment Weight (IPTW)-balanced pseudo-cohort.

								P-value
		MDD	OthDep	PsychRx	NonPsychRx	NonDepPsych	NoDep	overall	MDD	OthDep	PsychRx	NonPsychRx	PsychOnly
		n = 1514 (16%)	n = 1008 (11%)	n = 1179 (13%)	n = 1199 (13%)	n = 1336 (14%)	n = 3066 (33%)		vs NoDep	vs NoDep	vs NoDep	vs NoDep	vs NoDep
Index hospital
Length of stay	Median (Q1, Q3)	7 (3, 13)	6 (3, 13)	7 (4, 12)	7 (3, 13)	5 (3, 9)	5 (3, 9)	<.0001	<.0001	0.02	<.0001	<.0001	1.0000
Total payments	Median (Q1, Q3)	29142 (10780, 102440)	36844 (14496, 107828)	32525 (11586, 100505)	34551 (13556, 111524)	28922 (12172, 72745)	26899 (11540, 70342)	0.0007	0.3065	0.0003	0.035	0.0091	0.2019
Discharge home	n (%)	763 (50%)	513 (51%)	628 (53%)	704 (59%)	823 (62%)	2070 (68%)	<.0001	<.0001	<.0001	<.0001	<.0001	0.0002
1-years of healthcare utilization and cost
Rate for ER	n (%)	667 (44%)	422 (42%)	459 (39%)	357 (30%)	488 (37%)	693 (23%)	<.0001	<.0001	<.0001	<.0001	<.0001	<.0001
# of ER	Median (Q1, Q3)	13 (5, 30)	11 (4, 26)	10 (4, 23)	5 (2, 12)	9 (3, 19)	6 (3, 14)	<.0001	<.0001	<.0001	<.0001	0.1163	<.0001
ER payments	Median (Q1, Q3)	999 (247, 2484)	984 (284, 2415)	864 (220, 2267)	354 (90, 1054)	712 (169, 1744)	504 (134, 1303)	<.0001	<.0001	<.0001	0.0003	0.1492	0.0286
Rate hospitalized	n (%)	772 (51%)	486 (48%)	494 (42%)	330 (27%)	504 (38%)	687 (22%)	<.0001	<.0001	<.0001	<.0001	0.0005	<.0001
hospitalizations	Median (Q1, Q3)	1 (1, 2)	1 (1, 3)	1 (1, 2)	1 (1, 2)	1 (1, 2)	1 (1, 2)	<.0001	<.0001	<.0001	<.0001	0.065	0.0023
Hospital payments	Median (Q1, Q3)	29035 (9775, 80827)	32456 (11892, 75918)	27993 (10296, 67414)	24903 (10858, 63878)	23115 (9667, 53679)	22077 (10156, 49605)	0.0124	0.0097	0.004	0.036	0.3638	0.6782
# of outpatient	Median (Q1, Q3)	139 (79, 233)	118 (63, 205)	109 (57, 191)	89 (44, 156)	82 (42, 153)	62 (27, 118)	<.0001	<.0001	<.0001	<.0001	<.0001	<.0001
outpatient payments	Median (Q1, Q3)	17956 (7059, 39853)	18471 (7932, 39625)	15389 (5949, 35512)	11972 (4246, 25881)	10738 (3892, 25842)	7017 (2408, 17290)	<.0001	<.0001	<.0001	<.0001	<.0001	<.0001
# of medications	Median (Q1, Q3)	46 (20, 75)	36 (15, 65)	48 (28, 76)	41 (25, 64)	21 (4, 44)	19 (5, 40)	<.0001	<.0001	<.0001	<.0001	<.0001	0.0009
medications payments	Median (Q1, Q3)	2899 (551, 6822)	2285 (492, 6122)	3671 (1209, 7312)	3474 (1289, 7159)	875 (48, 3652)	1048 (71, 3651)	<.0001	<.0001	<.0001	<.0001	<.0001	0.1495
cumulative payments	Median (Q1, Q3)	36051 (13160, 83492)	35790 (14125, 79438)	28609 (11402, 71920)	22106 (8738, 47449)	20901 (7081, 48590)	11591 (4071, 30551)	<.0001	<.0001	<.0001	<.0001	0.0005	<.0001
Index hospitalizatin + post discharge	Median (Q1, Q3)	83247 (34802, 199671)	88239 (43164, 196756)	75760 (33636, 183880)	71490 (32855, 173349)	62595 (30732, 128294)	47980 (23859, 109895)	<.0001	<.0001	<.0001	<.0001	0.0005	<.0001
2-years of healthcare utilization and cost
Rate for ER	n (%)	787 (52%)	503 (50%)	584 (50%)	464 (39%)	601 (45%)	923 (30%)	<.0001	<.0001	<.0001	<.0001	<.0001	<.0001
# of ER	Median (Q1, Q3)	19 (8, 45)	17 (7, 41)	13 (5, 30)	8 (4, 20)	13 (5, 29)	8 (3, 19)	<.0001	<.0001	<.0001	<.0001	0.2762	<.0001
ER payments	Median (Q1, Q3)	1589 (451, 4065)	1436 (460, 4044)	1394 (401, 3176)	753 (159, 1872)	1081 (349, 2480)	697 (223, 1773)	<.0001	<.0001	<.0001	<.0001	0.5993	<.0001
Rate hospitalized	n (%)	983 (65%)	635 (63%)	645 (55%)	494 (41%)	703 (53%)	993 (32%)	<.0001	<.0001	<.0001	<.0001	<.0001	<.0001
hospitalizations	Median (Q1, Q3)	2 (1, 3)	2 (1, 3)	2 (1, 3)	1 (1, 2)	1 (1, 2)	1 (1, 2)	<.0001	<.0001	<.0001	<.0001	0.0127	<.0001
Hospital payments	Median (Q1, Q3)	33337 (10929, 93425)	35887 (14979, 82789)	31562 (12577, 76070)	28462 (11012, 62479)	24447 (10704, 61058)	21755 (9123, 49832)	<.0001	<.0001	<.0001	<.0001	0.0141	0.2499
# of outpatient	Median (Q1, Q3)	241 (136, 405)	199 (109, 344)	191 (103, 324)	139 (74, 254)	139 (77, 257)	99 (46, 194)	<.0001	<.0001	<.0001	<.0001	<.0001	<.0001
outpatient payments	Median (Q1, Q3)	31259 (13936, 66816)	30746 (13234, 64353)	27327 (10901, 57705)	20283 (8120, 41493)	17066 (7338, 41828)	11739 (4366, 26945)	<.0001	<.0001	<.0001	<.0001	<.0001	<.0001
# of medications	Median (Q1, Q3)	90 (41, 149)	73 (29, 126)	92 (56, 148)	74 (45, 123)	36 (9, 82)	34 (8, 75)	<.0001	<.0001	<.0001	<.0001	<.0001	0.0015
medications payments	Median (Q1, Q3)	6376 (1201, 14135)	4622 (943, 12538)	7507 (2542, 14751)	6497 (2322, 13761)	1683 (139, 7387)	1903 (151, 7196)	<.0001	<.0001	<.0001	<.0001	<.0001	0.3203
cumulative payments	Median (Q1, Q3)	63706 (27380, 142703)	62731 (27362, 127190)	51112 (23101, 117372)	37176 (16078, 82063)	38741 (13561, 80271)	21567 (7226, 52198)	<.0001	<.0001	<.0001	<.0001	<.0001	<.0001
Index hospitalizatin + post discharge	Median (Q1, Q3)	114691 (55925, 257365)	122157 (62534, 238189)	106972 (48707, 238586)	95145 (46743, 207929)	79795 (42528, 163501)	61816 (30930, 128903)	<.0001	<.0001	<.0001	<.0001	<.0001	<.0001

MDD: Major depressive disorder, OthDep: Other depression, PsychRx: Antidepressants for Other Psychiatric Condition, NonPsychRx: Antidepressants for non-Psychiatric Condition, NonDepPsych: Other Non-Depression Psychiatric Condition only, NoDep: No Depression. Depressed groups were compared to no depression group. This table shows that: (1) During the index hospitalization, most depressed groups had 1–2 days longer stay corresponding to an additional $6000 in payment and that less people were discharged home compared to no depression. (2) Within 1 year following injury, most depressed groups had 7%−19% more ER rates with 3–5 more ER visits, 5%−29% more people hospitalized, 7–57 more outpatient services, and 2–27 more prescription drugs- all corresponding to an overall additional payment of $9310–$24460. (3) Within 2 years following injury, most depressed groups had 9%−22% more ER rates with 5–11 more ER visits, 9%−33% more people hospitalized, 40–142 more outpatient services, and 2–58 more prescription drugs- all corresponding to an overall additional payment of $15609–$42139. Most of the maximum ranges were for the MDD group. Combining the index hospitalization and 2 years post-injury, the MDD payments were median $114691 compared to $61816 for NoDep.

Figure 5.

Cumulative healthcare utilization payments compared between ICD-derived depression phenotypes. 1-year and 2-years post-injury, major depressive disorder had the highest payments, and no depression had the lowest payment. This figure shows that the payments were ordered in the same rank order as the defined depression phenotypes.

Discussion

From injury to 24 months, there were 67% with depression as defined by ICD and prescription drugs defined depressed phenotypes. The analysis revealed that there was a significantly more negative change in depression phenotypes associated with injury (classified by the transition to a more severe phenotype within 24 months post-SCI) than positive change (37% vs. 15%, respectively; P < 0.0001). Additionally, depression phenotypes before SCI were the most significant predictor of depression phenotypes post-SCI. Depressed phenotypes showed significant associations with greater healthcare resource utilization and payments. The MDD cohort incurred the highest costs at 12 and 24 months post-injury, followed by OthDep, PsychRx, NonPsychRx, and NonDepPsych cohorts. The NoDep cohort had the lowest cost. Similar patterns were observed for rates of ER visits, hospital admissions, outpatient services, and medication prescriptions.

Williams et al. (1) estimated the prevalence of depression in SCI to be approximately 22% (range: 18.7%–26.3%) in a meta-analysis. In our current study, depressed phenotypes made up 67%. MDD and OthDep constituted 26%, similar to the Williams et al. research statistics. The 19 studies included in their meta-analysis were selected if they indicated that depression was measured (either by a depression diagnostic measure or by a diagnosis of major or minor depressive episodes). This means that depression was assessed by either symptom screening questionnaires or by ICD codes. Limitations of these methods alone have been expressed, notably, the fact that screening methods are not standardized (13), that classification based on questionnaire responses is not consistent with ICD codes (14), and the fact that the use of ICD alone does not capture all cases compared to when it is combined with other sources (15). In fact, using depression phenotypes, this study captures an additional 40% of individuals who would need close attention for psychological care.

A transition from NoDep to MDD, was seen in 6% of the population following SCI. Lim and colleagues observed similar results and described significant associations between traumatic SCI and new-onset anxiety and depression (34). This is also similar to rates of new depression diagnosed post-spinal surgery, in which 5-6% of patients without preexisting depression develop depressive symptoms within 24 months following surgery (21). To our knowledge, this is the first study to evaluate the transition of depressed and non-depressed classification pre- to post-injury. We found that SCI was associated with a likelihood of being depressed than not. More individuals were transitioning to more severe classes of depression phenotypes (37%) than less severe classes (15%, P < .0001).

In our study, there were significantly more females in the MDD cohort and other depressed phenotypes compared to the NoDep cohort. Several studies have also observed higher rates of depression in females compared with males post-SCI (4, 35, 36). Regarding age, we identified decreased depression rates in older individuals. This was in contrast to Krause et al., who identified a positive correlation between depressive symptoms and age at the injury (5). They noted a curvilinear relationship between symptoms and time since the injury, with the highest levels at the lower and higher ages. However, Krause et al.’s report is a single center study, which could explain the differences with our observations. Our findings are also consistent with observations in the general population, in which rates of depression are notably higher in younger and female subgroups (37). Previous studies have reported an impact of level and severity of injury on depression, with AIS D having a higher number of measured and self-reported depressive symptoms (6, 7) compared to tetra- and paraplegia AIS A-C and higher odds of depressive disorders compared to AIS A (8). The database used in this study does not capture the severity of the injury; therefore, we cannot quantify its association with depression phenotypes. It should be explored in future studies. However, the data contains the type of injury (traumatic/non-traumatic) and level of injury (cervical/thoracic/lumbar/sacral) to characterize SCI. We found that lumbar/sacral injuries were associated with lower odds of being depressed compared to cervical, consistent with prior studies (4), and that traumatic injuries had higher odds of being in depressed phenotypes than non-traumatic injuries.

Opioid use 12 months before the injury was significantly higher among individuals in the MDD cohort, while more individuals in the NoDep group had no opioid use. This is consistent with prior research, which showed that opioid use within 24 h of injury correlated with an increased risk of depression in SCI (38). Furthermore, individuals with MDD are more likely to develop chronic opioid use (9). The implications of these findings are especially critical given the risk of polypharmacy in SCI (39) and the ongoing nationwide opioid crisis (40).

A depression phenotype post-SCI leads to increased overall healthcare utilization and payments. A combined index and post-discharge payments were significantly higher in the MDD group ($28,000) than in the NoDep group. Similar results were seen regarding ER utilization, outpatient services, medication prescriptions, and hospital utilization. The association between post-SCI depression and increased healthcare utilization is consistent with evidence from prior studies, with individuals receiving more inpatient and outpatient services, particularly SCI Veterans with multiple medical comorbidities (10, 41). This is also seen in post-spinal surgery, in which patients with MDD and other depressed phenotypes incurred the highest healthcare utilization (9).

Implications for practice

The clinical implications of our findings cannot be understated. Healthcare providers need to understand the psychological challenges after SCI, including the impact on adjustment to disability and its long-term consequences (42). Addressing depression following SCI is paramount, given the increased risk for adverse outcomes, adding to suffering, increased cost, and healthcare utilization among the SCI population. Moreover, the screening for depression history following SCI, such as evaluating pre-injury EHR to classify individuals into depression phenotypes, becomes vital to guide further inpatient and post-acute care interventions. The treatment of depression is associated with improving its clinical symptoms (43). Therefore, ongoing depression screening may facilitate timely intervention, which may, in turn, lead to appropriate treatment as mental health disorders are generally undertreated (44), improving clinical outcomes and ultimately reducing patient healthcare utilization. This continuous psychiatric check is important given that about 6% of individuals without depression prior to SCI were found to be in a depressive phenotype within 24 months. Future studies should evaluate predictors of depression in this group to help clinicians with early intervention for those at high risk.

Limitations

The main limitation of this study is the reliance upon the accuracy of ICD codes within our dataset. Although the use of ICD-9 codes has been validated using the algorithm by Ingram et al. (9), we cannot exclude errors resulting from inaccurate coding. Additionally, medication use (including outpatient, antidepressant, and opioid medications) was evaluated using only prescription records, not actual patient ingestion. Thus, we cannot account for illicit opioid use or nonadherence to antidepressant medication regimens. Another significant limitation is the lack of an injury severity measure, the American Spinal Injury Association (ASIA) Impairment Scale (AIS). It is also important to note that claims data only captures individuals getting help through the healthcare system, those who are active enough about their mental/psychological health and/or those who are severely affected to engage in the healthcare system for their treatment. This misses those who might be experiencing depressive symptoms and/or those getting care for depression in avenues other than a medical system.

Conclusion

Using a classification of depression phenotypes, individuals classified in the depressed phenotypes have worse clinical outcomes and higher health care utilization and cost following SCI than individuals who remained in the non-depressed cohort following SCI. The classification into a depressed phenotype before SCI was the most significant predictor of a depressed phenotype post-SCI. These results emphasize the clinical importance of routinely screening all SCI patients by obtaining a thorough psychiatric history, including past substance abuse disorder and suicide risks, with prompt initiation of treatment when indicated to optimize outcomes and resource utilization. This method of classifying depression phenotypes provides an simple and practical way to obtain this information by screening through pre-injury medical records. By combining ICD and medications into this algorithm, the depression phenotypes allow to detect more individuals with psychiatric needs, beyond MDD and other depression diagnosis, that might be otherwise missed.

Funding Statement

This work was supported by None: [Grant Number].

Disclosure statement

No potential conflict of interest was reported by the author(s).