# Prediction of somatic and non-somatic depressive symptoms between inpatient rehabilitation and follow-up

## Abstract

Longitudinal.

### Objective:

We identified changes in the association of somatic and non-somatic symptoms (as measured by the Patient Health Questionnaire-9, PHQ-9) between inpatient rehabilitation after spinal cord injury (SCI) and 1 year after discharge.

### Setting:

A specialty hospital in the Southeastern USA.

### Methods:

A total of 584 adults with traumatic SCI were administered the PHQ-9 during inpatient rehabilitation. Of them, 227 completed the PHQ-9 by survey at 1-year follow-up. We performed time-lagged regression between times of measurement for somatic and non-somatic factors of the PHQ-9.

### Results:

The non-somatic factor at baseline was significantly predictive of the non-somatic (r=0.67, P=0.002) and somatic factors at follow-up (r=0.53, P=0.019). The somatic factor did not significantly predict either the somatic (r=0.10, n.s.) or non-somatic factors at follow-up (r=−0.01, NS). Factor analysis also indicated changing factor structure between inpatient rehabilitation and follow-up.

### Conclusions:

Our results question the interpretation of somatic items during inpatient rehabilitation, as they are not predictive of either somatic or non-somatic symptoms at follow-up.

## Introduction

Spinal cord injury (SCI) is associated with immediate and generally permanent changes in sensory and motor functions.1 Depressive disorders are more prevalent among persons with SCI and are associated with poor outcomes in several areas including rehabilitation, health, activities of daily living and community integration, as well as lower levels of quality of life.2

Diagnostic criteria for major depression include somatic symptoms that parallel physiologic responses to SCI or prolonged hospitalization. Differential validity of psychological measures with SCI was first noted more than 40 years ago in studies of the MMPI.3 Similar concerns have been noted in conditions with HIV,4 as some depressive symptoms correspond with symptoms of HIV infection. Removing these somatic items improved the clinical utility of the Beck Depression Inventory5 and the Center for Epidemiologic Studies Depression Scale.6 They concluded somatic symptoms will inflate depression scores in people living with HIV infection.4

Several studies have addressed the factor structure of depressive symptoms after SCI using the Patient Health Questionnaire-9 (PHQ-9),7 a nine-item self-report measure intended to assist in identification of major depressive disorders (MDD). It has been used in a number of recent studies with SCI,8, 9, 10 as well as with traumatic brain injury11 and stroke.12

Because the PHQ-9 items are based on the Diagnostic and Statistical Manual of Mental Disorders-IV (DSM-IV)13 criteria for major depression, assessing the factor structure has the potential to differentiate somatic items from non-somatic ones. Evaluating patterns of item endorsement also addresses the potential differential item validity of somatic items.

Richardson and Richards8 evaluated the factor structure of the PHQ-9 using data from the SCI Model Systems in the USA during preliminary 1-year follow-up and then again at each 5-year follow-up through 25 years after injury (that is, 1, 5, 10, 15, 20 and 25 years). A two-factor solution best fit the data on each occasion. The findings differed in some respects across time, as psychomotor changes loaded with the somatic factor on some occasions but not others. Both the somatic and non-somatic factors were negatively correlated with life satisfaction.

Bombardier et al.9 investigated depressive symptoms in 849 participants 1 year after injury. Of them, 22% scored at or above the cutoff for MDD (10 or higher), and 11.4% met the criteria for probable major depression based on the presence of one of two cardinal symptoms of depression and endorsement of five of the symptoms on more than half of the days. The most frequently endorsed symptoms (occurring on more than half the days) were sleep disturbance (25.9%), poor energy (24.3%) and anhedonia (19.8%).

Krause et al.10 used confirmatory factor analysis to compare four alternative structures of the PHQ-9 during inpatient rehabilitation after SCI. A two-factor solution, with three distinctive somatic items (sleep disturbance, poor energy and appetite change) best fit the data. These three items also had the highest endorsement rates. The physiologic complications of SCI are most likely to parallel or mimic the somatic symptoms of a depressive disorder during the inpatient hospitalization, as this occurs immediately after the onset of SCI.

The presence of the somatic factor, in and of itself, is not problematic, in that somatic content is highly predictive of major depressive episodes. For instance, using an international primary care population, Barkow et al.14 found that pain and somatic complaints were associated with depressive episodes after 1-year follow-up. Hein et al.15 conducted a study among elderly participants who had never previously reported a depressive disorder and identified several preclinical symptoms of a depressive disorder that included dysphoria, change in appetite, insomnia, lack of energy, morning depth, lack of joy and interest, inferiority feeling, lack of self-confidence, poor concentration, indecisiveness, thinking about death, wish to die and joint pain. Yet, other investigators have identified an association between sleep loss and major depression.16

Taken together, these studies suggest that somatic complaints may represent preclinical symptoms of a depressive disorder in primary care and elderly participants. However, because of the profound nature of physiologic changes associated with SCI, it is important to determine whether the somatic symptoms will be predictive of later depressive symptoms and disorders among those with SCI.

### Purpose

The purpose of this study was to identify the association of somatic and non-somatic symptoms measured during inpatient rehabilitation with somatic and non-somatic symptoms measured 1 year after hospital discharge. Our working hypothesis is that, compared with somatic symptoms, non-somatic symptoms measured during inpatient rehabilitation will be more highly correlated with depressive symptoms at 1 year after discharge (both somatic and non-somatic symptoms). Our rationale is that somatic symptoms endorsed during inpatient rehabilitation are confounded with actual somatic changes due to SCI or hospitalization and will not be highly correlated with depressive symptoms that are not otherwise accounted for by the condition (that is, those not related to the symptoms of SCI or prolonged hospitalization) 1 year after discharge.

## Materials and methods

### Procedures

Participants were recruited from the inpatient census of a specialty hospital in the Southeastern USA. The inclusion criteria were (1) traumatic SCI, (2) currently hospitalized for initial rehabilitation and (3) at least 16 years of age. A total of 707 participants consented and 584 completed the research materials before discharge (82.6%). Median number of days between onset of SCI and time of interview was 44.

The peer support coordinator met with participants and obtained informed consent. As needed, a research assistant helped the participant complete the materials. Participants were given US$25 remuneration. The follow-up was initiated 12 months after discharge. Participants were again offered$25 remuneration to complete the materials by mail. Because of constraints in the timeline of the study, only 410 of the 584 participants were included in the follow-up. (Of the 410 cases, only 377 received both a follow-up mailing and phone call due to time restrictions and termination of the funding cycle.) A total of 227 participants returned completed materials approximately 1 year after discharge (55%).

### Statement of ethics

We certify that all applicable institutional and governmental regulations concerning the ethical use of human volunteers were followed during the course of this research.

### Measures

The PHQ-97 consists of nine items, each identifying symptoms that are commonly associated with a depression diagnosis based on DSM-IV criteria. The participants are asked to identify how each symptom has been a problem over the past 2 weeks using the following four categories: (1) not at all, (2) several days, (3) more than half the days or (4) nearly every day. The PHQ-9 is both reliable and valid as a brief measure of depression. Kroenke et al.7 found 0.89 internal consistency and 0.84 test–retest reliability.

### Data analysis

Primary data analyses were conducted using Mplus,17 (Los Angeles, CA, USA) specialized software for a wide range of structural equation models (SEM), which calculates parameters in the presence of incomplete data by using the missing at random (MAR) assumption.18 This means all available data, including that of those participants who only responded at baseline, were used to derive estimates used in the time-lagged regression analyses. The MAR assumption applies only to those variables not in the equation. Therefore, maximum likelihood was used to create unbiased estimates of the parameters using all available information19 from baseline responses to adjust follow-up estimates as if the data were complete.

The primary analyses included factor analysis and time-lagged regression analysis. Factor analysis is a component of SEM that is used to identify the underlying dimensions, or measurement model, of a specific set of items or an instrument. We conducted factor analysis of participant responses during inpatient rehabilitation and at 1 year (a single analysis) using maximum likelihood with a Promax rotation. The root mean square error of approximation (RMSEA) was used to evaluate the fit of the model. The RMSEA is a function of N, the χ2 and the degrees of freedom and is determined by the discrepancy per degrees of freedom and corrects for model complexity. The fit of the model improves as the RMSEA decreases, and values of less than 0.05 represent an excellent fit.20

We then conducted a time-lagged regression analysis. Latent somatic and non-somatic dimensions at follow-up were regressed on somatic and non-somatic dimensions observed during inpatient rehabilitation. Because the goal of this study was to identify the extent to which somatic and non-somatic items are predictive of similar content between inpatient rehabilitation and follow-up, we used the inpatient factor structure on both occasions. The time-lagged regression analyses used correlations between factors across time, as scores on one factor at time 1 were used to predict scores on the other factor at time 2. Beta weights were generated as coefficients, and RMSEA and the χ2 statistics were used to determine model fit.

There was no attempt to include biographic or injury variables as covariates, as Mplus calculates parameters using the MAR assumption such that additional factors only complicate the interpretation. However, SPSS Inc. (Chicago, IL, USA) was used to compare the 227 follow-up respondents to the 357 who did not participate in the 1-year follow-up. Descriptive statistics were also generated to characterize the participant sample. The mean number of items endorsed for the somatic (three items) and non-somatic factors (six items) was compared between the inpatient baseline and 1-year follow-up using paired t-tests. The McNemar statistic was used to compare changes in the percentage of endorsement of individual items between baseline and the 1-year follow-up when items were dichotomized based on participants stating the symptom occurred on at least half the days, except for self-harm where it was based on reporting the symptom any days.

## Results

### Participants

At follow-up, 73.6% of the participants were Caucasian; 74.4% were men. Average age at onset was 35.5 years. Nearly half of the participants (48%) had cervical injuries. The primary etiology of injury was motor vehicle crash (48.9%). The average number of years of education was 13.1.

### Characteristics of follow-up respondents and nonrespondents

Respondents were older, had higher levels of education and were more likely to be women (Table 1).

### Changes in item and score distribution over time

Significant changes in the proportion of participants who reported the symptom on at least half of the days between inpatient and follow-up were noted for three symptoms (Table 2). The portion of participants endorsing appetite change decreased from inpatient (29.0%) to follow-up (17.3%). In contrast, there were significant increases in the proportion of participants endorsing feeling down (inpatient=9.5%, follow-up=16.7%) and feeling bad about self (inpatient=10.7%, follow-up=18.2%).

A summative score for the three somatic items did not significantly change from baseline, although non-somatic items increased (t(208)= 2.42, P=0.016).

Whereas only 9.8% of the participants reported no depressive symptoms at baseline, 16.4% reported no symptoms at the 1-year follow-up (Table 3). Similarly, whereas only 8.5% reported scores of 15 and higher at baseline, 11% scored in this range at follow-up.

### Factor analysis

The exploratory factor analysis across the two times of measurement resulted in four factors (RMSEA=0.045), χ2 (d.f.=87)=191.14, P=0.000. The first and third factors reflect the baseline (those used in the time-lagged regression), whereas the others reflect follow-up data (Table 4). The baseline factors include six items loading with the non-somatic factor and three items loading with the somatic factor (sleep disturbance, poor energy and appetite change). The follow-up factors appear to be quite different, as four items loaded with each factor and the final item, sleep disturbance, loaded nearly identical in both factors (0.42, −0.38).

### Regression: time 1 to time 2

Figure 1 summarizes the model regressing follow-up somatic and non-somatic latent scores on baseline latent scores. The vertical lines represent the correlation between the two factors at a single point in time (either at baseline or follow-up), whereas the horizontal lines represent the correlation of a factor measured at baseline with the same factor measured at follow-up. The diagonal lines are the essential component of the time-lagged correlation, as they represent the beta weight of one factor at baseline with the other factor at follow-up (that is, the extent to which each factor at baseline predicts scores on the other factor at follow-up).

The model proved to be a good fit (RMSEA=0.054). The non-somatic factor at baseline was significantly related to both the non-somatic (+0.67, P=0.002) and somatic factors (+0.53, P=0.019) at follow-up. In contrast, the somatic factor during baseline was not significantly related to either the somatic (+0.10, n.s.) or the non-somatic (−0.01, n.s.) factors at follow-up. Both latent factors were significantly correlated at baseline (+0.24, P=0.000) and follow-up (+0.39, P=0.000).

## Discussion

Overall, it appears that somatic symptoms during inpatient rehabilitation are not predictive of future depressive symptoms, either somatic or non-somatic. In contrast, the presence of non-somatic symptoms during inpatient rehabilitation is predictive of future depressive symptomatology, including somatic symptoms.

There is ambiguity over the meaning of somatic items measured during inpatient rehabilitation. Because somatic symptoms during hospitalization were not related to somatic symptoms at 1-year after discharge, the presence of such symptoms during hospitalization is of limited value in predicting future depressive symptoms. These findings are consistent with earlier research with SCI3 and with HIV research that indicated difficulty interpreting somatic symptoms.4 However, these findings are inconsistent with research on elderly participants in primary care settings, so the findings do not generalize to all populations and settings.

### Limitations

First, although the data are longitudinal, they span only the first approximate 15 months after SCI onset and represent only two points in time of data collection. Second, as with most studies of depressive symptoms, the data are limited to self-report. Similarly, there is no psychological assessment that could be used for diagnostic purposes to help differentiate whether the somatic items are reflective of depression or another parameter. Third, we have no data on types of treatment participants had received. Finally, attrition was relatively high in this study with about 45% lost cases among those who were the focus of the follow-up. Although higher response rates are always preferable, it is important to note that SEM represents a powerful methodology for imputing missing data based on the characteristics and responses of those who do and do not complete follow-up materials. Similarly, we did not include biographic and injury characteristics in the regression analyses, as the differences in characteristics between respondents and nonrespondents were relatively small.

### Clinical implications

Because somatic items are of limited value in predicting future depressive symptoms, the presence of somatic complaints during inpatient rehabilitation should be interpreted with caution by rehabilitation psychologists and other professionals. When used in diagnosis, they should only be considered in conjunction with other key symptoms, particularly depressed mood and loss of interest.

Treatment approaches should target non-somatic symptoms. Specifically, treatment strategies may focus on improving affect and increasing participation in activities. When psychotropic medications are used, they should be evaluated based on changes in non-somatic symptoms.

Although these findings suggest that the PHQ-9 must be used with caution in both clinical evaluations and research studies, the findings do not preclude its use, even during inpatient rehabilitation. However, independent measurement of somatic and non-somatic domains is appropriate.

### Future research

At least three types of research are needed. First, more research is needed on the natural course of development of symptoms over time. Second, we need to identify the coping strategies, both favorable and unfavorable, that relate to depressive symptomatology. Lastly and most importantly, we need well-designed intervention studies to develop and test interventions to reduce or eliminate depressive symptoms.

## References

1. 1

National Spinal Cord Injury Statistical Center (NSCISC). Annual Statistical Report. University of Alabama: Birmingham, AL, 2006.

2. 2

Elliott TR, Frank RG . Depression following spinal cord injury. Arch Phys Med Rehabil 1996; 77: 816–823.

3. 3

Kendall P, Edinger J, Eberly C . Taylor's MMPI correction factor for spinal cord injury: empirical endorsement. J Consult Clin Psychol 1978; 46: 370–371.

4. 4

Kalichman SC, Rompa D, Cage M . Distinguishing between overlapping somatic symptoms of depression and HIV disease in people living with HIV-AIDS. J Nerv Ment Dis 2000; 188: 662–670.

5. 5

Beck AT, Ward CH, Mendelson M, Mock J, Erbaugh J . An inventory for measuring depression. Arch Gen Psychiatry 1961; 4: 561–571.

6. 6

Radloff LS . The CES-D Scale: a self-report depression scale for research in the general population. Appl Psychol Measur 1977; 3: 385–401.

7. 7

Kroenke K, Spitzer RL, Williams JB . The PHQ-9: validity of a brief depression severity measure. J Gen Intern Med 2001; 16: 606–613.

8. 8

Richardson EJ, Richards JS . Factor structure of the PHQ-9 screen for depression across time since injury among persons with spinal cord injury. Rehabil Psychol 2008; 53: 243–249.

9. 9

Bombardier CH, Richards JS, Krause JS, Tulsky D, Tate DG . Symptoms of major depression in people with spinal cord injury: implications for screening. Arch Phys Med Rehabil 2004; 85: 1749–1756.

10. 10

Krause JS, Bombardier CH, Carter RE . Assessment of depressive symptoms during inpatient rehabilitation for spinal cord injury: is there an underlying somatic factor when using the PHQ? Rehabil Psychol 2008; 53: 513–520.

11. 11

Fann JR, Bombardier CH, Dikmen S, Esselman P, Warms CA, Pelzer E et al. Validity of the Patient Health Questionnaire-9 in assessing depression following traumatic brain injury. J Head Trauma Rehabil 2005; 20: 501–511.

12. 12

Williams LS, Brizendine EJ, Plue L, Bakas T, Tu W, Hendrie H et al. Performance of the PHQ-9 as a screening tool for depression after stroke. Stroke 2005; 36: 635–638.

13. 13

APA. Diagnostic and Statistical Manual of Mental Disorders, 4th edn American Psychiatric Association: Washington, DC, 1994.

14. 14

Barkow K, Heun R, Bedirhan UT, Maier W . Identification of items which predict later development of depression in primary health care. Eur Arch Psychiatry Clin Neurosci 2001; 251: 21–26.

15. 15

Hein S, Bonsignore M, Barkow K, Jessen F, Ptok U, Heun R . Lifetime depressive and somatic symptoms as preclinical markers of late-onset depression. Eur Arch Psychiatry Clin Neurosci 2003; 253: 16–21.

16. 16

Roberts RE, Shema SJ, Kaplan GA, Strawbridge WJ . Sleep complaints and depression in an aging cohort: a prospective perspective. Am J Psychiatry 2000; 157: 81–88.

17. 17

Muthen L, Muthen B . Mplus, The Comprehensive Modeling Program for Applied Researchers User's Guide, 4th edn Muthen & Muthen: Los Angeles, CA, 2006.

18. 18

Rubin DB . Inference and missing data. Biometrika 1976; 63: 581–592.

19. 19

Little RJA, Rubin DB . Statistical Analysis with Missing Data. Wiley: New York, 1987.

20. 20

Browne M, Cudeck R . Alternative ways of assessing model fit. In: Bollen K, Long S (eds). Testing Structural Equation Models. Sage: Beverly Hills, CA, 1993, pp 136–162.

## Acknowledgements

This research was supported by a field-initiated grant from the National Institute for Disability and Rehabilitation Research of the Office of Special Education and Rehabilitation Services (H133G030151) and the Model Spinal Cord Injury Systems Grant (H133N060009 and H133N000005). The opinions here are those of the grantee and do not necessarily reflect those of the funding agencies. We thank the following people who made substantial contributions to the study: Richard Aust, Jennifer Coker, Philip Edles, Emily Johnson, Kristian Manley, Dr Lee L Saunders, Sean Twohig and Yusheng Zhai, and Sarah Lottes and Christina McCleery of the Shepherd Center.

## Author information

Correspondence to J S Krause.

## Rights and permissions

Reprints and Permissions

Krause, J., Reed, K. & McArdle, J. Prediction of somatic and non-somatic depressive symptoms between inpatient rehabilitation and follow-up. Spinal Cord 48, 239–244 (2010) doi:10.1038/sc.2009.113

• #### DOI

https://doi.org/10.1038/sc.2009.113

### Keywords

• spinal cord injury
• depression
• rehabilitation
• disability
• psychosocial

• ### Screening for depression in pregnant women from Côte d׳Ivoire and Ghana: Psychometric properties of the Patient Health Questionnaire-9

• Dana Barthel
• , Claus Barkmann
• , Stephan Ehrhardt
• , Stefanie Schoppen
•  & Carola Bindt

Journal of Affective Disorders (2015)

• ### Depressive symptoms are associated with allostatic load among community-dwelling older adults

• Roni W. Kobrosly
• , Edwin van Wijngaarden
• , Christopher L. Seplaki
• , Deborah A. Cory-Slechta
•  & Jan Moynihan

Physiology & Behavior (2014)

• ### Association of Internet Use and Depression Among the Spinal Cord Injury Population

• I-Hsuan Tsai
• , Daniel E. Graves
• , Ching-Huang Lai
• , Lu-Yu Hwang
•  & Lisa A. Pompeii

Archives of Physical Medicine and Rehabilitation (2014)

• ### Multisystem physiological dysfunction is associated with depressive symptoms in a population-based sample of older adults

• Roni W. Kobrosly
• , Christopher L. Seplaki
• , Deborah A. Cory-Slechta
• , Jan Moynihan
•  & Edwin van Wijngaarden

International Journal of Geriatric Psychiatry (2013)

• ### The factor structure of the PHQ-9 in palliative care

• Joseph Chilcot
• , Lauren Rayner
• , William Lee
• , Annabel Price
• , Laura Goodwin
• , Barbara Monroe
• , Nigel Sykes
• , Penny Hansford
•  & Matthew Hotopf

Journal of Psychosomatic Research (2013)