Using Multimodal Assessments to Reevaluate Depression Designations for Spine Surgery Candidates

Abstract

Background:

Depression is common in spine surgery candidates and may influence postoperative outcomes. Ecological momentary assessments (EMAs) can overcome limitations of existing depression screening methods (e.g., recall bias, inaccuracy of historical diagnoses) by longitudinally monitoring depression symptoms in daily life. In this study, we compared EMA-based depression assessment with retrospective self-report (a 9-item Patient Health Questionnaire [PHQ-9]) and chart-based depression diagnosis in lumbar spine surgery candidates. We further examined the associations of each depression assessment method with surgical outcomes.

Methods:

Adult patients undergoing lumbar spine surgery (n = 122) completed EMAs quantifying depressive symptoms up to 5 times daily for 3 weeks preoperatively. Correlations (rank-biserial or Spearman) among EMA means, a chart-based depression history, and 1-time preoperative depression surveys (PHQ-9 and Psychache Scale) were analyzed. Confirmatory factor analysis was used to categorize PHQ-9 questions as somatic or non-somatic; subscores were compared with a propensity score-matched general population cohort. The associations of each screening modality with 6-month surgical outcomes (pain, disability, physical function, pain interference) were analyzed with multivariable regression.

Results:

The association between EMA Depression scores and a depression history was weak (r_rb = 0.34 [95% confidence interval (CI), 0.14 to 0.52]). Moderate correlations with EMA-measured depression symptoms were observed for the PHQ-9 (r_s = 0.51 [95% CI, 0.37 to 0.63]) and the Psychache Scale (r_s = 0.68 [95% CI, 0.57 to 0.76]). Compared with the matched general population cohort, spine surgery candidates endorsed similar non-somatic symptoms but significantly greater somatic symptoms on the PHQ-9. EMA Depression scores had a stronger association with 6-month surgical outcomes than the other depression screening modalities did.

Conclusions:

A history of depression in the medical record is not a reliable indication of preoperative depression symptom severity. Cross-sectional depression assessments such as PHQ-9 have stronger associations with daily depression symptoms but may conflate somatic depression symptoms with spine-related disability. As an alternative to these methods, mobile health technology and EMAs provide an opportunity to collect real-time, longitudinal data on depression symptom severity, potentially improving prognostic accuracy.

Level of Evidence:

Diagnostic Level III. See Instructions for Authors for a complete description of levels of evidence.

Depression is a frequent comorbid condition in patients with degenerative spine disease, with a reported preoperative prevalence of 31%¹. Patients with depression tend to report worse pain and disability before and after surgical procedures^2-4, and the relative benefit of a surgical procedure for these patients has remained subject to debate^2-5. Accurately identifying depression and quantifying its severity are prerequisites to better understand the complex relationships among depression, spine disease symptoms, and surgical outcomes.

Ecological momentary assessment (EMA) consists of repeated short surveys that can be completed on a mobile device. This provides longitudinal data in a natural setting and in real time, which can improve validity and reduce recall bias^6,7. The analysis of daily symptom dynamics may yield useful information, such as the associations among and chronology of pain, activity, and depressive symptoms^8,9. An improved understanding of symptoms could allow for a better determination of whether a patient will benefit from a surgical procedure. Additionally, more precise and detailed methods of evaluating spine disability and psychosocial health may support more informative assessments of surgical outcomes¹⁰.

Reliable and valid assessment of depression is particularly relevant for patients with chronic pain given the substantial overlap in symptoms and co-occurrence of these conditions^11,12. Although depression assessments including the 9-item Patient Health Questionnaire (PHQ-9) typically conceptualize depression as a unidimensional measure^13,14, there has been debate over whether depression symptoms are better characterized as somatic and non-somatic¹⁵. Non-somatic (or cognitive-affective) symptoms relate to mental or emotional states (e.g., depressed mood), whereas somatic symptoms refer to physical symptoms (e.g., fatigue and sleep disturbances)¹⁶. Previous studies have concluded that somatic depressive symptoms are inflated in patients with chronic pain, and using a depression inventory that includes somatic symptoms may overestimate depression severity^17,18. Consequently, this issue should be considered in the spine surgery population before making clinical decisions based on traditional depression screening tools.

Obtaining longitudinal EMAs does require more time from clinicians, staff, and patients compared with a 1-time assessment. Although these smartphone-based surveys are short, they require some familiarity with technology and brief daily effort from patients. Therefore, it would be beneficial to determine the extent to which EMAs provide information that overlaps with typical methods of depression classifications in the spine literature, including electronic health record (EHR) diagnoses and self-report tools, such as the PHQ-9^2,19-21. In this study, we examined the association of EMAs with traditional depression assessments and compared the prognostic utility of these assessment modalities as correlates of 6-month surgical outcomes.

Materials and Methods

Study Population and Data Collection

This was a secondary analysis of prospectively collected data from patients undergoing lumbar fusion or decompression surgery at a single center. The inclusion criteria were that the patients spoke English, were adults between 21 and 85 years of age, owned a smartphone, had ≥1 week to complete the assessments before the surgical procedure, and reported a Numeric Rating Scale (NRS) pain score ≥3 of 10 over the previous week. Participants were excluded if they had undergone another major surgical procedure within 3 months before the index surgical procedure or were undergoing a surgical procedure for infection, malignancy, or trauma. The study was approved by the institutional review board (#202012139), and informed consent was provided by all participants. The details of the study design have been previously published²². Briefly, participants used the LifeData smartphone application to complete short EMA surveys scheduled 5 times daily for approximately 3 weeks before the surgical procedure. Participants also completed traditional questionnaires preoperatively and at 6 months postoperatively. Participants with missing questionnaires or <5 completed EMAs were excluded. Race, ethnicity, and gender identity were primarily identified by self-report, with missing data obtained from the EHR.

Depression Assessments and Outcome Measures

A primary objective was to analyze associations among depressive symptoms measured using preoperative EMA, symptoms measured using validated 1-time questionnaires (PHQ-9 and Psychache Scale), and a history of depression as reported in the EHR. The presence of a depression comorbidity was documented in the EHR during a patient’s preoperative anesthesia assessment by a nurse practitioner or anesthesiology resident physician based on a review of the medical record and patient self-report.

The PHQ-9 quantifies depression symptom frequency over the preceding 2 weeks¹⁴. Responses were analyzed as total scores, as a somatic symptom subscore, and as a non-somatic symptom subscore (based on results of confirmatory factor analysis [CFA]). The Psychache Scale is a 13-item questionnaire with total scores ranging from 13 to 65 in which participants rate the frequency and severity of psychological pain on a Likert scale²³. It includes items such as “I seem to ache inside,” and “My psychological pain seems worse than any physical pain.” The psychological pain measured by this scale is strongly correlated with depression symptoms^24,25. Eight percent of participants missed a single item on either the PHQ-9 or the Psychache Scale. Therefore, multiple imputation was used and pooled results have been reported for all analyses.

Preoperative EMAs included 3 questions adapted from the Patient-Reported Outcomes Measurement Information System (PROMIS) Emotional-Distress-Depression Short Form²⁶. On a slider user interface ranging from 0 (not at all) to 100 (worst possible), participants reported how depressed, hopeless, and worthless they currently felt. The EMA Depression score was defined as the mean of the 3 items at each assessment point. The mean EMA Depression score was calculated for each participant by averaging all completed EMAs. The reliability of the PHQ-9, Psychache Scale, and EMA Depression score is reported in the Appendix.

Additional questionnaires included the NRS for back and leg pain over the previous week, the Oswestry Disability Index (ODI), PROMIS Pain Interference (PROMIS-PI), PROMIS Physical Function (PROMIS-PF)²⁷, and Pain Catastrophizing Scale^28,29. The NRS pain score was considered to be the worse of the 2 scores for leg or back pain.

Statistical Analysis

To calculate somatic and non-somatic PHQ-9 subscores, questions were classified using CFA results. CFA is used to evaluate whether measurements of observed variables are consistent with a hypothesized latent construct³⁰. Because PHQ-9 scores were not normally distributed, CFA used maximum likelihood estimation with Satorra-Bentler scaling (“lavaan” R package). The 1-factor model and several previously proposed 2-factor models¹⁵ were compared using the Comparative Fit Index (CFI)³¹; see the Appendix for additional fit statistics.

Associations among the EMA Depression, PHQ-9, Psychache Scale, and EHR-based depression were quantified using rank-biserial or Spearman correlation and were reported with 95% confidence intervals (CIs). The Mann-Whitney U test, Kruskal-Wallis test, and Dunn test with correction for multiple comparisons were used as appropriate as nonparametric tests of significance. The association between depression assessment scores and each preoperative cross-sectional assessment was quantified using the Spearman correlation. Analyses were conducted using R version 4.2.3 (The R Foundation). Significance was set at p < 0.05.

To explore whether somatic depression symptoms may be elevated in patients undergoing lumbar spine surgery compared with similar individuals not undergoing the surgical procedure, the publicly available 2017 to 2020 National Health and Nutrition Examination Survey (NHANES) was used³². Propensity score matching (“matchit” R package) was performed using a nearest neighbor matching algorithm in a 1:4 ratio for the following demographic and clinical variables: gender, age, education, race and ethnicity, body mass index, and antidepressant use. Balance achieved after matching was assessed by verifying that standardized mean differences for each covariate were <0.10. Somatic and non-somatic subscores were normalized on the basis of the number of questions in each.

To compare the strength of association of each depression assessment modality (EMA, EHR, PHQ-9, and Psychache) with patient-reported outcome measures (PROMs), multivariable linear regressions were performed. For each PROM (NRS pain, ODI, PROMIS-PI, and PROMIS-PF), improvement from baseline to 6 months postoperatively was the dependent variable, whereas the preoperative PROM value and the depression assessment modality were the independent variables. Standardized β regression coefficients and the R² of each model were reported. The Cox test was used to determine if the model using EMA had additional explanatory value beyond that provided by each other model³³. A baseline model with only the preoperative PROM as a predictor was also reported.

Results

Participant Characteristics

Of 148 enrolled participants, 122 completed the required EMAs and surveys (Fig. 1). Included participants completed a mean (and standard deviation) of 76 ± 24 (range, 9 to 145) preoperative EMAs. The median age was 60 years, and the sample was predominantly White (93%) and non-Hispanic (98%) (Table I). A depression comorbidity was listed for 40 participants (33%). The PHQ-9 score was ≥10, suggesting at least moderate depression¹⁴, for 32% of participants.

Fig. 1.

Flow diagram of included participants.

TABLE I.

Demographic and Baseline Summary Data

Characteristic	Value (N = 122)
Age* (yr)	60 (49 to 66)
Female gender†	65 (53%)
Race†
White	114 (93%)
Black	6 (5%)
Other	2 (2%)
Ethnicity†
Hispanic	3 (2%)
Non-Hispanic	119 (98%)
Education level†
Did not graduate high school	3 (2%)
High school degree	43 (35%)
College degree	41 (34%)
Graduate or professional degree	35 (29%)
Smokes cigarettes or cigars†	4 (3%)
Opioid use†	43 (35%)
Depression comorbidity in EHR†	40 (33%)
Anxiety comorbidity in EHR†	42 (34%)
Depression and anxiety in EHR†	30 (25%)
Antidepressant use†	51 (42%)
PHQ-9 ≥10†	39 (32%)
Mean EMA Depression*	4 (0.5 to 18)
PHQ-9*	6 (3 to 11)
Psychache Scale*	19 (15 to 25)
Pain Catastrophizing Scale*	18 (7 to 28)
NRS pain* (worst of back or leg)	8 (7 to 9)
Oswestry Disability Index* (%)	64 (56 to 76)
PROMIS-PI T-Score*	68 (64 to 70)
PROMIS-PF T-Score*	33 (29 to 37)

^*The values are given as the median, with the IQR in parentheses.

^†The values are given as the number of patients, with the percentage in parentheses.

CFA of PHQ-9

The CFA of the PHQ-9 showed that the 1-factor model did not have good fit. The 2-factor models, which divided questions into somatic and non-somatic categories, were significantly better based on the CFI (Table II). Additional fit statistics are provided in the Appendix, Supplemental Table S1. Model 2c performed best and was used to calculate PHQ-9 somatic and non-somatic subscores. The somatic items were Q1: anhedonia, Q3: sleep difficulties, Q4: fatigue, Q5: appetite changes, Q7: concentration difficulties, and Q8: psychomotor agitation or retardation. The non-somatic items were Q2: depressed mood, Q6: feeling of worthlessness, and Q9: thoughts of death.

TABLE II.

CFA of Proposed PHQ-9 Models^*

	Model 1 (Single Factor)	Model 2a	Model 2b	Model 2c
Non-somatic item numbers	—	1, 2, 6, 7, 8, 9	1, 2, 6, 9	2, 6, 9
Somatic item numbers	—	3, 4, 5	3, 4, 5, 7, 8	1, 3, 4, 5, 7, 8
CFI†	0.85	0.91	0.96	0.98

^*PHQ-9 Items: (1) anhedonia, (2) depressed mood, (3) sleep difficulties, (4) fatigue, (5) appetite changes, (6) feeling of worthlessness, (7) concentration difficulties, (8) psychomotor agitation or retardation, and (9) thoughts of death.

^†This ranges from 0 to 1, with larger values indicating a better model fit.

Depression Assessment Correlations

There were significant differences (p < 0.01) between patients with and without a depression comorbidity listed in the EHR for all 3 depression indices (Fig. 2, Table III). Rank-biserial correlations were weak to moderate, with r_rb = 0.34 (95% CI, 0.14 to 0.52) for EMA Depression scores, r_rb = 0.42 (95% CI, 0.22 to 0.58) for the PHQ-9, and r_rb = 0.35 (95% CI, 0.15 to 0.53) for the Psychache Scale. Notably, there was substantial variability and overlap in quantitative depression scores between patients with and without EHR-indicated depression.

Fig. 2.

Association of depression indices with depression diagnosed based on the EHR. The rank-biserial correlation coefficient and p value for the Mann-Whitney U test are shown. The interquartile range (IQR) is indicated by a box, the median is indicated by a line within the box, and points within 1.5 times the IQR width are indicated by whiskers.

TABLE III.

Summary of Correlations

	EHR Depression* (Rank-Biserial Correlation)	EMA Depression Mean* (Spearman Correlation)
EMA Depression mean	0.34 (0.14 to 0.52)	—
PHQ-9 total	0.42 (0.22 to 0.58)	0.51 (0.37 to 0.63)
PHQ-9 somatic	0.33 (0.12 to 0.51)	0.44 (0.28 to 0.57)
PHQ-9 non-somatic	0.46 (0.27 to 0.62)	0.63 (0.51 to 0.73)
Psychache scale	0.35 (0.15 to 0.53)	0.68 (0.57 to 0.76)

^*The values are given as the correlations, with the 95% CI in parentheses.

Given the relatively weak association between EMA Depression scores and EHR-indicated depression, we next examined the association between EMA and PHQ-9/Psychache Scale scores. The Spearman correlation between EMA Depression scores and the PHQ-9 was 0.51 (95% CI, 0.37 to 0.63), whereas the correlation with the Psychache Scale was 0.68 (95% CI, 0.57 to 0.76) (Fig. 3, Table III). EMA Depression scores had a correlation of 0.63 (95% CI, 0.51 to 0.73) with PHQ-9 non-somatic items but only 0.44 (95% CI, 0.28 to 0.57) with somatic items. Although the EMA prompts only asked about non-somatic symptoms (e.g., hopelessness), the mean EMA Depression scores still had an association with previously published PHQ-9 severity categories (Fig. 4)¹⁴. The Kruskal-Wallis test showed significant differences in EMA Depression scores among PHQ-9 categories (p < 0.001). A post hoc comparison with the Dunn multiple comparison test showed no significant differences in EMA Depression scores between adjacent categories (e.g., minimal and mild), but the scores always differed significantly between non-adjacent categories (e.g., minimal and moderate).

Fig. 3.

Association of mean EMA Depression scores with the PHQ-9 (total score, somatic subscore, and non-somatic subscore) and Psychache Scale. The Spearman correlation coefficient is given. For ease of visualization, linear regressions with shaded 95% CIs are plotted.

Fig. 4.

EMA Depression mean scores by PHQ-9 category. Categories: minimal = 0 to 4, mild = 5 to 9, moderate = 10 to 14, moderately severe = 15 to 19, and severe = 20 to 27. The interquartile range (IQR) is indicated by a box, the median is indicated by a line within the box, and points within 1.5 times the IQR width are indicated by whiskers.

Correlation with Cross-Sectional Assessments

The correlations of preoperative depression assessments with results on preoperative pain and disability assessments were compared (Fig. 5). Notably, the PHQ-9 somatic subscore had a stronger association with these assessments than the PHQ-9 non-somatic subscore did. For example, the correlation with the ODI was 0.60 (95% CI, 0.48 to 0.71) for PHQ-9 somatic subscores but only 0.45 (95% CI, 0.29 to 0.58) for PHQ-9 non-somatic subscores. Both the EMA Depression and the Psychache Scale typically showed similar correlations as the PHQ-9 non-somatic subscore.

Fig. 5.

Correlations between preoperative depression assessments and preoperative pain and disability assessments.

PHQ-9 Comparison with a Matched Cohort

The comparison of the 122 spine surgery candidates with 488 matched individuals from the NHANES data set showed significantly higher (p < 0.001) PHQ-9 somatic subscores, but the difference in non-somatic subscores did not reach significance (p = 0.06) (Fig. 6). The median normalized somatic subscore was 0.83 (interquartile range [IQR], 0.5 to 1.5) in the spine surgery cohort compared with 0.33 (IQR, 0 to 0.83) in the matched cohort. The median non-somatic subscore was 0 (IQR, 0 to 0.67) in both cohorts.

Fig. 6.

PHQ-9 somatic and non-somatic subscores compared with a propensity score-matched cohort of individuals not undergoing spine surgery from the NHANES database. Subscore totals are normalized by the number of questions included in each subscore. The interquartile range (IQR) is indicated by a box, the median is indicated by a line within the box, and points within 1.5 times the IQR width are indicated by whiskers. **P < 0.001, and ns = not significant.

Association with Surgical Outcomes

At 6 months, 105 participants (86%) completed follow-up and were included in the outcome analysis. For all PROMs, depression was associated with a lower magnitude of improvement from baseline to 6 months when controlling for the baseline PROM value (Table IV). Of the 4 depression assessment methods, EMA showed the strongest association with surgical outcomes based on standardized β and model R², and it was a significant predictor variable for each. For example, EMA Depression was a stronger predictor of postoperative NRS pain (β = −0.31, R² = 0.25) than other depression indices were (values ranged from β = −0.12, R² = 0.18 to β = −0.21, R² = 0.21). Furthermore, the Cox test showed p < 0.01 for each, indicating that significant additional explanatory information was provided by the EMA Depression compared with each model using a different depression assessment.

TABLE IV.

Multivariable Regression Standardized β Weights and Fits for Improvement in 6-Month PROMs^*†

Depression Index Predictor	NRS Pain			ODI			PROMIS -PI			PROMIS-PF
	β	P Value	R2	β	P Value	R2	β	P Value	R2	β	P Value	R2
EMA	−0.31	0.001	0.25	−0.39	<0.001	0.20	−0.38	<0.001	0.24	−0.26	0.02	0.11
EHR	−0.20	0.024	0.21	−0.17	0.07	0.13	−0.18	0.05	0.16	−0.07	0.47	0.06
PHQ-9	−0.14	0.135	0.18	−0.14	0.25	0.11	−0.30	0.01	0.19	−0.10	0.38	0.07
PHQ-9 somatic	−0.14	0.158	0.18	−0.04	0.73	0.10	−0.23	0.05	0.16	−0.06	0.60	0.06
PHQ-9 non-somatic	−0.12	0.223	0.18	−0.26	0.02	0.15	−0.28	<0.01	0.19	−0.14	0.18	0.07
Psychache scale	−0.21	0.022	0.21	−0.35	<0.01	0.19	−0.31	<0.01	0.21	−0.11	0.29	0.07
None (baseline model)	NA	NA	0.17	NA	NA	0.10	NA	NA	0.13	NA	NA	0.06

^*NA = not applicable.

^†For each depression index predictor and outcome pair, a multivariable linear regression model was created. The only other predictor in the model was the preoperative value of the corresponding outcome. Standardized β weights and p values are shown for each predictor. The R² quantifies the fit of each model. The only variable in the baseline model is the preoperative PROM value, which was common across all models compared.

Discussion

Despite increased focus on the potential role of depression in spine surgery outcomes, existing screening methods such as 1-time assessments and chart review have inherent limitations. In this study of patients undergoing lumbar spine surgery, we found that a substantial number of participants reporting high levels of daily depression symptoms using EMA were not identified from the EHR. Cross-sectional surveys such as the PHQ-9 and Psychache Scale provided a better method of estimating daily symptoms but still had only moderate correlation with EMAs. Critically, EMA Depression had the strongest association with surgical outcomes compared with other depression-screening modalities. The poor performance of the PHQ-9 may be related to conflation of spine-related disability with somatic depression symptoms, which were elevated in the spine surgery cohort. Overall, these results suggest that current methods of depression screening in spine surgery candidates may be inadequate or possibly misleading.

The poor performance of chart review for predicting depression symptoms was likely exacerbated by several factors. First, diagnoses may be historical, inaccurate, or not properly documented. Second, it is possible some participants without a depression diagnosis would have been diagnosed if evaluated by a psychologist. Finally, some patients with a depression history may have minimal symptoms due to successful treatment. Consequently, clinicians and researchers should not rely solely on chart review as an indicator of current depression symptoms in spine surgery candidates.

Despite strong correlations with measures such as the Psychache Scale and the non-somatic items of the PHQ-9, there were discrepancies between the EMA Depression scores and those obtained with these measures. This suggests EMA may capture novel information missed by 1-time, cross-sectional assessments. Although not explored here, it is possible that EMA reduces recall bias, yielding more accurate assessment of depressive symptoms. Future studies can compare EMA with retrospective PROMs indexing the same time period.

EMA Depression had a stronger association with the non-somatic PHQ-9 subscore than with the somatic subscore. This finding may be expected because the EMA questions assessed patient-reported feelings of depression, hopelessness, and worthlessness, which are non-somatic symptoms. However, treating the PHQ-9 as a 2-factor model led us to question whether somatic symptoms might be disproportionately increased in spine surgery candidates. This question was outside the scope of this study, and we did not include a control group without chronic pain. Therefore, we compared our results with a propensity score-matched cohort of individuals from the NHANES database. Although non-somatic scores were similar between the cohorts, somatic scores were significantly elevated in the spine surgery candidates. This aligns with previous studies showing inflated somatic symptoms in patients with chronic pain^17,18. Therefore, depression severity may be inappropriately elevated in the spine surgery literature due to pain-related somatic symptoms. Future studies are needed to validate this assertion. Nonetheless, this finding does not detract from the real mental health challenges faced by patients with chronic pain. Instead, this should serve as a reminder to clinicians and investigators of the importance of carefully chosen screening tools and patient interviews when making a depression diagnosis.

A primary motivation for incorporating mobile health technology and EMAs into preoperative assessment is to inform more accurate, personalized prognoses¹⁰. This work specifically focused on comparing EMA with more commonly used depression assessment methods. We studied the association of each screening modality with surgical outcomes and found that the EMA Depression measurement had stronger associations with all outcomes compared with the EHR, PHQ-9, or Psychache Scale. This finding highlights the potential for EMA depression screenings to improve accuracy of prognostic tools used to predict which patients are likely to benefit from the surgical procedure. Future work should also include collaboration with clinical psychologists to design appropriate interventions based on EMA of depression, considering, in part, their influence on postoperative outcomes.

This study had several limitations. First, there was not a structured clinical interview to serve as a gold standard, which would have been useful when comparing EMAs and cross-sectional assessments. Second, participants were from a single academic medical center, tended to be highly educated, were predominantly White and non-Hispanic, and were nearly all non-smokers. Although spine surgery is most common among White, non-Hispanic individuals, it was possible that our inclusion criteria (e.g., English-speaking, owning a smartphone) further limited generalizability. Third, the number of participants with severe depressive symptoms based on the PHQ-9¹⁴ was relatively small. Finally, although we used a population control as a comparison with our spine surgery cohort, a direct comparison with a cohort without chronic pain collected using identical methods would further support our conclusions with regard to somatic depression symptoms compared with non-somatic depression symptoms.

In conclusion, although depression diagnoses and cross-sectional surveys correlate with depressive symptoms in spine surgery candidates, these measures appear imprecise and subject to recall bias. The EMA of depression symptom severity demonstrates promise for more accurate prediction of surgical outcomes.