Secondary Logo

Journal Logo

Featured Articles: Original Clinical Research Report

Multidimensional Perioperative Recovery Trajectories in a Mixed Surgical Cohort: A Longitudinal Cluster Analysis Utilizing National Institutes of Health Patient-Reported Outcome Measurement Information System Measures

Kent, Michael L. MD*; Giordano, Nicholas A. PhD, RN; Rojas, Winifred BSN, RN, CCRP‡,§; Lindl, Mary Jo MS, RN‡,§,∥; Lujan, Eugenio MD; Buckenmaier, Chester C. III MD; Kroma, Raymond BS‡,§; Highland, Krista B. PhD‡,§

Author Information
doi: 10.1213/ANE.0000000000005758


See Article, p 276


  • Question: In addition to pain intensity trajectories, are distinct trajectories of biopsychosocial measures also present in postsurgical models?
  • Findings: When utilizing a longitudinal clustering algorithm, distinct postoperative stratified functional clusters were observed as well as unique trajectories of pain impact, psychological health, and physical health measures.
  • Meaning: As a complement to pain intensity trajectories, the longitudinal measurement of additional biopsychosocial variables may impact the ability to detect and impact at-risk patient populations to improve overall perioperative recovery trajectories.

Recently, clinical investigations on perioperative pain have shifted from static assessments to dynamic evaluations (ie, pain intensity trajectories) as predictors for chronic postsurgical pain (CPSP).1 However, a large body of evidence delineates the bidirectional influence between pain intensity and other modulating factors such as sleep, anxiety, pain catastrophizing, and depression.2,3 This bidirectional relationship is well defined across many chronic pain conditions in which depressive or anxious traits, for example, are cross-sectionally associated with longitudinal poor outcomes.4 However, changes in psychosocial measures such as self-efficacy have been shown to dynamically mediate pain responses in models such as chronic low back pain.5 These modulators build upon the limits of pain intensity to address the comprehensive concepts of multidimensional recovery and pain impact.6–8 Characterizing multidimensional trajectories throughout the entire perioperative course may complement our understanding of how such longitudinal outcomes evolve.

Numerous studies have utilized postoperative pain intensity trajectories, or clusters, to investigate the association of preoperative patient-level risk factors, trajectory membership, and subsequent negative long-term outcomes (ie, CPSP).9,10 However, it remains unclear how to estimate patients’ risks for experiencing and modifying such pain intensity trajectories to avoid negative long-term outcomes. Further, identifying pain intensity trajectories alone fails to capture the biopsychosocial presentations of patients throughout the perioperative period, a time when patients’ physical, mental, and social health can exacerbate their pain responses. Despite their dynamic presentations, many of these investigations only assess emotional and cognitive factors preoperatively in a cross-sectional approach.11,12 There is a need to examine perioperative pain intensity trajectories in the context of a variety of biopsychosocial factors known to modulate the pain experience, such as anxiety, depression, and sleep, which may also dynamically change during the postoperative period.13 In both cardiac surgery and spine surgery, persistent anxiety and depression throughout the perioperative period was found to be associated with decreased physical function and increased pain.14,15 However, these investigations have often been limited to only evaluating short-term changes in biopsychosocial presentations. Recently, in a mixed surgical cohort, multidimensional measures such as pain catastrophizing have also been described to have unique perioperative trajectories that influence acute and subacute pain intensity.16 Thus, longitudinal monitoring of the biopsychosocial presentations of pain can inform the design of future-focused interventions capable of impacting membership within high- or low-risk pain intensity trajectories.

In this prospective observational mixed surgical cohort study, a battery of biopsychosocial measures, capable of characterizing the perioperative pain experience, was administered at standardized time points over a 6-month period. Building upon pain intensity–based trajectories, the primary goal of this study was to identify and describe postoperative pain impact trajectories based on the measures of pain interference and physical function. Our second goal was to describe concurrent trajectories in biopsychosocial measures known to modulate the postoperative pain experience. Such multidimensional trajectories build upon cross-sectional research and observational studies of shorter duration and further inform the degree of surgical impact on patient recovery by focusing on the dynamic nature of biopsychosocial pain presentations.


Participants and Procedures

This was an observational, multisite, longitudinal cohort study of patients undergoing a variety of surgeries. Recruitment occurred at 2 large tertiary care military treatment facilities in the United States, Walter Reed National Military Medical Center (WRNMMC) and Naval Medical Center San Diego (NMCSD), from May 1, 2016 to August 1, 2018. The study was approved by the institutional review board (IRB) at WRNMMC and NMCSD (IRB# 500094), and written informed consent was obtained from all subjects or a legal surrogate, or the parents or legal guardians of minor subjects. Active duty members, family members, and retirees were recruited during their presurgical clinic visit or on admission to the presurgical unit for major abdominal surgery, thoracic surgery, mastectomy, total knee arthroplasty (TKA), total hip arthroplasty (THA), or spinal fusion. These surgeries are associated with both a significant acute and chronic pain burden and have been studied in a variety of settings with regard to pain and postoperative opioid use.11,12 After expressing interest in participating, study staff screened patients for inclusion and exclusion criteria and obtained informed consent along with Health Insurance Portability and Accountability Authorization. Patients were considered for inclusion if they were undergoing one of the surgeries listed above and enrolled within the Military Health System or Veterans Administration. Patients were excluded if they were younger than 18 years, older than 80 years, could not read or communicate in English, or were unable to consent for surgery or understand protocol instructions. After enrollment, participants were asked to complete online surveys via the Research Electronic Data Capture (REDCap) system preoperatively (baseline) and again at 1 week, 2 weeks, 1 month, 3 months, and 6 months postoperatively.17 Electronic surveys could be completed at home or using a tablet provided by study staff when patients were in the clinic. Reporting is conducted via utilization of the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) criteria for cohort studies. The projected sample size listed on for this observational study was 1500 participants, much higher than the number of enrolled participants; this discrepancy was due to regulatory delays.

Variables of Interest

Three specific descriptive recovery domains were prespecified: pain impact, physical health, and psychological health. Surveys addressing these domains included several National Institutes of Health (NIH) Patient Reported Outcome Measurement Information System (PROMIS) computer adaptive testing (CAT)–enabled scales: pain impact (PROMIS Pain Interference & Physical Functioning), physical health (PROMIS Fatigue, Sleep Disturbances, and Satisfaction with Social Roles and Activities), and psychological health (PROMIS Depression, Anxiety, Anger, and Social Isolation). PROMIS measures were selected due to their favorable psychometric properties, including reduced response burden.18 The 5-item static PROMIS Neuropathic Pain Quality scale was also included. All PROMIS scales produce T scores that are normed on the general population, with a mean of 50 and standard deviation (SD) of 10. Scores within 1 SD of 50 were considered “within normal range,” though evidence indicates scores 0.5 SD (5 points) worse from 50 may be indicative of mild, yet clinically relevant symptom levels, and 1 SD (10 points) worse than 50 may be indicative of moderate symptom severity.19 Scales are interpreted per their name. For example, higher pain interference scores indicate greater dysfunction, while higher physical functioning scores indicate lower dysfunction. In addition to static severity characterizations, item bank–specific minimally clinical important differences (MCID) have also been reported for PROMIS measures in a variety of settings, and the literature was reviewed to descriptively provide MCIDs for measures utilized in this study.20–28

Additional patient-reported outcomes assessed preoperative pain intensity, pain catastrophizing levels, and opioid use. The Defense and Veterans Pain Rating Scale (DVPRS) was used to assess average pain intensity over the past 24 hours within the context of functioning.29 DVPRS responses range from 0 (no pain) to 10 (as bad as it could be, nothing else matters). The pain catastrophizing scale is a validated and reliable measure of pain catastrophizing.30 Responses for each of the 13 scale items range from 0 (not at all) to 4 (all the time). Items are summed, with total scores ranging from 0 to 52. Participants were also asked whether they were currently taking opioids (yes/no).

Data were extracted via self-report or from the electronic medical record. This included demographics (sex, race, ethnicity, and age), health (body mass index, smoking status, cancer, and American Society of Anesthesiologists [ASA] physical status), and surgery (surgery type, peripheral nerve block receipt, and days of hospital stay).

Statistical Methods

Dropout Comparisons

First univariate tests assessed data normality. Shapiro-Wilk tests indicated the continuous variables were not normally distributed, and therefore, subsequent comparisons utilized nonparametric tests. Bivariate analyses (eg, χ2 and Kruskal-Wallis tests) examined whether participants who completed 3 or more time points varied in demographic and clinical characteristics relative to those who completed less than 3 time points. For multicategory characteristic variables (eg, race or surgery type), pairwise χ2 tests with adjusted P values were interpreted.

Pain Impact Trajectories

The prespecified domain of pain impact served as the primary exploratory objective given the dynamic and documented role of pain and physical function following surgery. Granular descriptions of the measures with physical health and psychological health are scarce within the literature with regard to changes within the postoperative period. Clustered pain impact joint trajectories of PROMIS. Pain Interference and PROMIS Physical Function were modeled simultaneously with a k-means for joint longitudinal data with a nonparametric approach, using the R package kml3d.31 This 2-step approach allowed for both longitudinal correlated outcomes to be transformed into a single categorical cluster identification variable. In the first step, the cluster trajectory membership was estimated for a 3-, 4-, and 5-cluster solution. Here, the k-means procedure was redrawn 100 times from different initial configurations of cluster assignments. For inclusion in the analyses, participants could have a maximum number of 3 timepoints missing, and remaining data that were missing at random were imputed using linear interpolation. In the second step, each participant, in each cluster solution, is assigned to the nearest cluster. The longitudinal k-means clustering approach does not require a minimum suggested sample size. Determining the optimal number of clusters was aided by using comparative fit indices (eg, Calinski-Harabasz [CH] criterion and Akaike information criterion [AIC]), global average of postprobabilities, and cluster meaningfulness and interpretability.32

Cluster Comparisons

After the number of pain impact clusters was determined, the remaining domain and measure pathways or trajectories, psychosocial health (anger, anxiety, depression, and social isolation) and physical health (fatigue, sleep disturbance, and social role satisfaction), were visually inspected and described per symptom severity level (normal, mild, or moderate). Then, participant characteristics and baseline scores between clusters were compared using χ2, Fisher exact (when cell size was <5), and Kruskal-Wallis tests. This included examining differences in patient-reported outcomes, as well as demographic, health, and surgery-related characteristics. Variables significantly associated with cluster membership were further analyzed using post hoc pairwise comparisons with Bonferroni P value corrections for multiple comparisons. For categorical variables, column and row proportions were compared to the overall proportions with χ2 tests for given probabilities. Statistical significance was set to P < .05. All analyses were conducted in R.


Sample Characteristics

There were 504 participants at the presurgical timepoint, with survey noncompletion occurring across 1-week (n = 417), 2-week (n = 407), 1-month (n = 388), 3-month (n = 356), and 6-month (n = 324) follow-ups. In the k-means longitudinal clustering, 403 participants had data on at least 3 time points and were included in subsequent models (“included”), and 101 completed less than 3 time points and were excluded from analyses (“excluded”). Kruskal-Wallis and Pearson χ2 tests indicated the included subsample varied from the excluded subsample in race and ethnicity distribution (χ2[4] = 13.44, P = .001) and receipt of a peripheral nerve block distribution (χ2[1] = 17.05, P < .001). The included subsample, compared to the excluded sample, had a lower proportion of Black participants and a higher proportion of participants who did not receive a peripheral nerve block. The differences between included and excluded samples are shown in Supplemental Digital Content 1, Table 1, Otherwise, there was a lack of evidence, indicating the included versus excluded subsamples varied in other demographic characteristics, medical factors, or any of the patient-reported outcomes.

Pain Impact Trajectories

The 2-cluster (CH = 203.70, AIC = −9538.90, mean probability = 97% [95% confidence interval {CI}, 96–98]), 3-cluster (CH = 162.32, AIC = −8930.13, mean probability = 94% [95% CI, 93–96]), and 4-cluster (CH = 133.15, AIC = −8778.59, mean probability = 92% [95% CI, 91–93]) solutions were compared across fit indices, posterior probabilities, and clinical interpretability. In closer review, 1 cluster from the 2-cluster model was divided between 2 clusters of the 3-cluster model, suggesting the existence of 3 unique clusters. Moreover, the 3-cluster solution indicating low, average, and high performance was more consistent with the range of clinical presentations seen in practice and per previous literature. Therefore, the 3-cluster solution was selected for further evaluation.

Cluster Comparisons

Table 1. - Participant Characteristics and Outcomes for Full Sample and Across Clusters
Characteristic Full sample
N = 402
Low performance (n = 141, 35%) Average performance (n = 188, 47%) High performance (n = 74, 18%) P value
Posterior probabilities, mean (SD) 0.95 (0.11) 0.94 (0.12) 0.96 (0.09) 0.93 (0.14) .13
Sex, n (%) <.01
 Malea 186 (51.2) 71 (56.3) 98 (56.3) 17 (27.0)b
 Femalea 177 (48.8) 55 (43.7) 76 (43.7) 46 (73.0)c
Body mass index, median [IQR] 29.1 [26.2–33.5] 30.5 [27.1–33.9]d 29.1 [26.2–34.2]e 27.3 [24.2–31.0]d,e <.01
Age, median [IQR] 55.0 [44.0–61.0] 55.0 [46.0–61.0]d 55.5 [46.0–62.0]e 46.0 [28.2–58.0]d,e <.01
Race and ethnicity, n (%) .14
 Hispanic or Latino 29 (7.38) 8 (5.80) 13 (7.14) 8 (11.0)
 Multiple or other 20 (5.09) 8 (5.80) 5 (2.75) 7 (9.59)
 Non-Hispanic Asian 17 (4.33) 6 (4.35) 7 (3.85) 4 (5.48)
 Non-Hispanic Black 64 (16.3) 25 (18.1) 24 (13.2) 15 (20.5)
 Non-Hispanic White 263 (66.9) 91 (65.9) 133 (73.1) 39 (53.4)
Beneficiary type, n (%) .05
 Active duty 123 (30.5) 39 (27.7) 51 (27.1) 33 (44.6)
 Family member 122 (30.3) 41 (29.1) 61 (32.4) 20 (27.0)
 Retiree or other 158 (39.2) 61 (43.3) 76 (40.4) 21 (28.4)
ASA physical status, n (%) <.01
 Ia 39 (9.68) 3 (2.13)b 17 (9.04) 19 (25.7)c
 II 273 (67.7) 102 (72.3) 129 (68.6) 42 (56.8)
 III 91 (22.6) 36 (25.5) 42 (22.3) 13 (17.6)
Smoker, n (%) .23
 No 375 (93.1) 127 (90.1) 178 (94.7) 70 (94.6)
 Yes 28 (6.95) 14 (9.93) 10 (5.32) 4 (5.41)
Cancer, n (%) <.01
 No 315 (78.2) 123 (87.2) 147 (78.2) 45 (60.8)
 Yesa 88 (21.8) 18 (12.8)b 41 (21.8) 29 (39.2)c
Chemotherapy, n (%) .19
 No 57 (64.8) 9 (50.0) 26 (63.4) 22 (75.9)
 Yes 31 (35.2) 9 (50.0) 15 (36.6) 7 (24.1)
Surgery, n (%) <.01
 Major abdominala 30 (7.46) 4 (2.84)b 10 (5.35) 16 (21.6)c
 Mastectomya 79 (19.7) 8 (5.67)b 32 (17.1) 39 (52.7)c
 Spinal fusiona 86 (21.4) 53 (37.6)c 29 (15.5)b 4 (5.41)b
 Thoracica 17 (4.23) 2 (1.42) 8 (4.28) 7 (9.46)c
 Total hip arthroplastya 95 (23.6) 26 (18.4) 63 (33.7)c 6 (8.11) b
 Total knee arthroplastya 95 (23.6) 48 (34.0)c 45 (24.1) 2 (2.70)b
Presurgical opioid use, n (%) <.01
 No 311 (87.4) 91 (74.0) 159 (92.4) 61 (100)
 Yesa 45 (12.6) 32 (26.0)c 13 (7.56) 0 (0.00) b
Peripheral nerve block, n (%) .18
 No 261 (64.8) 98 (69.5) 113 (60.1) 50 (67.6)
 Yes 142 (35.2) 43 (30.5) 75 (39.9) 24 (32.4)
Hospital duration (d), median [IQR] 3.00 [2.00–4.00] 3.00 [3.00–4.00]e 3.00 [2.00–4.00]e 3.00 [2.00–4.00] .04
Preoperative patient-reported outcomes,  median [IQR]
 Average pain, median 5.00 [2.00–6.00] 6.00 [5.00–7.00]e,f 5.00 [2.00–6.00]d,e 0.00 [0.00–2.00]d,f <.01
 PROMIS Global Health 3.00 [3.00–4.00] 3.00 [2.00–4.00]d,e 4.00 [3.00–4.00]e 4.00 [3.00–4.00]d <.01
 Pain catastrophizing scale 10.0 [4.00–22.0] 21.0 [10.0–32.0]e,f 9.00 [3.00–16.0]d,e 3.00 [0.00–9.00]d,f <.01
Preoperative PROMIS measures, median [IQR]
 Pain interference 61.5 [54.4–66.9] 66.9 [63.0–70.1]e,f 61.2 [56.0–66.1]d,e 46.6 [38.7–52.6]d,f <.01
 Physical function 39.1 [34.6–46.3] 34.4 [31.4–38.0]e,f 39.5 [36.5–45.0]d,e 56.3 [50.2–62.6]d,f <.01
 Fatigue 55.4 [48.5–62.3] 60.9 [56.7–66.7]e,f 53.4 [47.4–59.3]d,e 47.3 [39.1–50.7]d,f <.01
 Sleep disturbance 56.1 [48.9–63.2] 62.3 [56.1–68.5]e,f 54.3 [49.1–62.0]d,e 45.6 [39.8–52.5]d,f <.01
 Depression 48.2 [42.7–55.0] 54.0 [48.2–59.4]d,e 46.1 [40.5–51.2]e 44.8 [38.6–51.2]d <.01
 Anxiety 52.9 [46.0–58.9] 55.5 [51.2–61.6]d,e 51.2 [45.1–56.5]e 49.4 [39.9–56.5]d <.01
 Anger 48.8 [42.2–54.7] 52.5 [47.4–59.6]e,f 48.5 [42.1–53.5]d,e 44.2 [33.9–48.8]d,f <.01
 Social role satisfaction 44.7 [37.9–51.6] 38.7 [32.4–44.6]e,f 46.2 [39.9–51.6]d,e 51.6 [48.2–61.1]d,f <.01
 Social isolation 42.3 [36.5–49.0] 47.0 [40.6–54.8]d,e 42.2 [34.2–47.2]e 41.2 [31.8–47.0]d <.01
Continuous variable comparisons analyzed with Kruskal-Wallis tests and are displayed as medians [IQRs]. Categorical variable comparisons analyzed with χ2 tests and are displayed as frequency (%).
Abbreviations: ASA, American Society of Anesthesiologists; IQR, interquartile range; PROMIS, Patient Reported Outcome Measurement Information System; SD, standard deviation.
aThe distribution of participants in a row across the 3 clusters was significantly different than the expected distribution based on the overall proportion of participant allocation to the low (35%), average (47%), and high (18%) performance clusters.
bProportion in the cell was lower than expected relative to the overall proportion.
cProportion in the cell was higher than expected relative to the overall proportion.
d–fSignificant pairwise comparison of continuous variable with false discovery rate adjustment for multiple comparisons. Similar superscripts indicate the comparison groups that are different.

Table 2.:
Median [25th Percentile, 75th Percentile] Changes in Postoperative PROMIS Scores From Preoperative Values, by Cluster
Figure 1.:
Pain impact outcomes.
Figure 2.:
Physical outcomes.
Figure 3.:
Psychological outcomes.

Three distinct clusters were identified: low, average, and high performance (Figures 1–3). Differing cluster trajectories were observed for pain impact (PROMIS Physical Function and Pain Interference), physical health (PROMIS Fatigue, Sleep Disturbances, and Social Role Satisfaction), and psychological health (PROMIS Depression, Anxiety, Anger, and Social Isolation; Figures 1–3). As shown in Table 1, clusters were characterized by sex, body mass index, age, ASA physical status, cancer history, surgery type, presurgical opioid use, hospital duration, and all patient-reported outcomes. Changes in PROMIS measures indexed to published MCIDs are descriptively presented in Table 2.

Low Performance

At baseline, the low-performance cluster was characterized by a higher-than-expected proportion of participants undergoing TKA, spinal surgery, and reporting presurgical opioid use (Table 1). Preoperative pain catastrophizing scale scores were significantly higher than other clusters, with a median score of 21 (10.0; 32.0). Similarly, the distribution of hospitalization length of stay was significantly higher in this cluster compared to the high-performance cluster. From baseline to 6 months, pain impact measures remained in the moderate-impairment range for both physical function and pain interference, but demonstrated clinically important improvements in both scales by 3-month follow-up (Figure 1; Table 2). Physical health (fatigue, sleep disturbance, and satisfaction with social role and activities), on the other hand, began in the moderate impairment range and improved to mild impairment by 6 months, with fatigue and sleep disturbances meeting MCID thresholds (Figure 2; Table 2). With regard to psychological health measures, anger, depression, and social isolation remained within normal population ranges through 6 months (Table 2; Figure 3). However, while categorized as mild impairment at baseline, anxiety improved to population norms and surpassed the MCID threshold by 3 months. Self-reported opioid utilization was significantly higher compared to the average and high-performance clusters at all timepoints (Supplemental Digital Content 2, Table 2, Opioid use at baseline was reported by 26% of patients within this cluster and demonstrated an acute increase through the first postoperative month, with a return to baseline levels by 3 months postoperatively.

Average Performance

The average performance cluster was characterized by a higher proportion of patients undergoing THA compared to other clusters. Pain catastrophizing and all PROMIS measures were significantly better than the low-performance cluster but worse than the high-performance cluster (Table 1). While demonstrating mild impairment for physical function and moderate impairment for pain interference at baseline, patients within this cluster improved past baseline values for pain impact into the normal range by 6 months (Table 2; Figure 1). For psychological health variables, patients within this cluster were within average population ranges at baseline. However, for anxiety, anger, and depression, patients surpassed MCID thresholds, demonstrating lower values for these measures at 6 months (Table 2; Figure 3). For physical health, while within population norms at baseline, the average performance cluster reported fatigue and sleep disturbance improvements that met MCID by 1-month follow-up and surpassed the MCID range by 6-month follow-up (Table 2; Figure 2). By 3-month follow-up, the average performance cluster had sleep disturbance and social role satisfaction scores that were similar to the high-performance cluster. Baseline self-reported opioid utilization (7.56%) was not statistically different compared to the high-performance cluster (0%). However, during the first postoperative month, the average-performance cluster demonstrated significantly higher (P < .01) self-reported opioid use higher than the high-performance but lower than the low-performance clusters (Supplemental Digital Content 2, Table 2,

High Performance

The high-performance cluster had a higher-than-expected proportion of female participants, ASA physical status I, cancer, major abdominal surgery, mastectomy, and thoracic surgery; as well as a lower body mass index, younger age, and lower-than-expected proportion of participants reporting presurgical opioid use (Table 1). For pain impact trajectories, the high-performance cluster demonstrated scores within the average range for pain interference and physical function, but reported scores in the moderate-impairment range at 1-week follow-up (Table 2; Figure 1). By 1-month (pain interference) and 6-month follow-up (physical function), the high-performance cluster returned to baseline (Table 2; Figure 2). For physical health outcomes (sleep disturbance, fatigue, and satisfaction with social roles), this cluster demonstrated a transient worsening of scores at 1 week but returned to baseline normal values by 1 month postoperatively. While within population averages at baseline, no significant changes from baseline for psychological health were observed across anger or social isolation measures (Figure 3). However, this cluster demonstrated improvements in depression and anxiety that met and surpassed MCID, respectively (Table 2; Figure 3). No patients reported opioid utilization preoperatively, and while a small increase in the number of patients reporting opioid using was observed during the first postoperative month, no self-reported opioid utilization was observed by 6 months (Supplemental Digital Content 2, Table 2,


This prospective observational cohort study of 403 surgical patients observed 3 distinct perioperative trajectory clusters (low, average, and high performance) across numerous biopsychosocial variables. Compared to before surgery, the high-performance cluster demonstrated an acute worsening of pain impact and physical health variables, followed by a return to preoperative levels or better by 6 months. Conversely, the low-performance cluster remained within the moderate impairment range for pain impact, improved from moderate to mild impairment across physical health outcomes, and did not change from baseline for psychological health outcomes. Finally, the average trajectory was characterized as the most dynamic cluster, demonstrating improvement across all measures by 6 months postoperatively. Distinct differences in self-reported opioid utilization were observed, with the low-performance cluster reporting higher opioid utilization across all time points compared to other clusters. No gross differences in dynamic opioid utilization patterns were observed.

Similar to previous studies, our results demonstrate distinct trajectories of pain within the acute postoperative phase up to 6 months.10 Our longitudinal time horizon also mirrors previous studies in which trajectories clearly depict worsening pain, while others depict stable or improving pain.33,34 However, our methodology and results build upon previous literature in numerous ways. Unlike previous studies where cluster or trajectory membership was determined by the unidimensional measure of pain intensity, we include a broader functional strategy by conducting a cluster analysis to include both pain interference and physical function under the conceptualization of pain impact. Further, and similar to previous studies, we found elevated pain catastrophizing scale scores and preoperative opioid use to be associated with worse pain outcomes.35,36 However, unlike earlier studies, we also uniquely illustrate that these variables are associated with the longitudinal reporting of lower function across pain, physical health, and psychological health, collectively. This expands the term “pain trajectory” to include the novel concept of “recovery trajectories,” which, like pain intensity, may be modulated by the numerous biopsychosocial presentations of pain.

Our results demonstrated that numerous known modulators of the acute pain experience (ie, anxiety, depression, and sleep disturbance, etc) are also characterized by distinct trajectories as well as differing magnitudes and timing of such changes. To our knowledge, few studies have addressed the presence of distinct psychosocial trajectories following surgery. As an example, Pagé et al’s15 investigation utilizing a growth mixture model in 1071 patients undergoing cardiac surgery demonstrated distinct longitudinal depression and anxiety trajectories, which influenced the rate of persistent postsurgical pain. In our results, we observed distinct psychosocial trajectories but also significant changes and slopes within clusters. For example, unlike the high- or low-performance clusters, only participants in the average-performance cluster demonstrated continued improvement in the trajectories of psychosocial variables (anxiety, depression, and anger) through 6 months, compared to before surgery (Figure 3). This may have implications in identifying modifiable biopsychosocial variables that may either assist in patient stratification or modulate persistent pain following surgery.

Previous investigations of pain trajectories have demonstrated patient-level characteristics such as anxiety or pain catastrophizing as being associated with trajectory membership; however, mixed results have been reported in relation to specific surgical subtypes. Whereas acute pain trajectory investigations have included mixed surgical cohorts, longitudinal studies often focus on single surgical subtypes.1,11 In our longitudinal study, we included both patient-level characteristics and a set of mixed surgical subtypes, and observed longitudinal trajectory membership was associated with both. For example, the low-performance cluster predominantly consisted of patients undergoing TKA and spinal surgeries and was significantly worse across all baseline biopsychosocial variables compared to the average- or high-performance clusters. These findings are in contrast with trajectory and epidemiologic studies focused on the acute postoperative period during which surgical subtype does not consistently demonstrate a significant impact on acute pain trajectories in relation to patient-level characteristics.11

While PROMIS measures were utilized to establish these descriptive clusters, both the magnitude and directional change of these measures adds to the emerging knowledge base of clinically meaningful changes in surgical patient-reported outcome measures.37 Depending on a specific PROMIS measure, symptoms or states are categorized as mild, moderate, or severe; however, MCIDs across time are also relevant to clinical care optimization. While PROMIS MCIDs have been addressed in some surgical samples such as cervical spine surgery, foot/ankle surgery, and upper extremity fractures, it remains unclear how MCIDs translates to clinical decision making.27,38,39 Of note, in our sample, numerous 6-month PROMIS measures within clusters significantly surpassed MCIDs quoted in the literature when compared to baseline even though clusters remaining distinctly different from one another (Table 2). However, future studies are needed to further specify the clinical utility of such MCIDs within the postoperative period on a surgery- and timeframe-specific basis.

The longitudinal observational nature of this study has inherent limitations. These limitations include loss to follow-up and unmeasured confounding. However, the statistical approaches used to model trajectories in this study, k-means, are specifically designed to model cluster joint trajectories and interpolate group level trends, even when data are missing at random within subjects. In the present analysis, missing data were accounted for via linear interpolation. It is unclear whether the clusters would maintain if other missing data methods (eg, imputation) were selected, different longitudinal cluster analyses (eg, latent class growth analysis) were used, or a larger sample size was analyzed. Additionally, we found no systematic differences in characteristics between dropouts and those who remained in the study. While the diversity of surgical procedures included in this sample helps illustrate the presence of trajectories across surgical populations, it limits the generalizability of findings given the varying procedural volume. With regard to the study population, the inclusion of active duty service members may have also skewed results given the likely decreased comorbidity burden compared to civilian communities. Additionally, as diagnoses of active duty service members such as posttraumatic stress disorder or more prevalent compared to civilian cohorts, this may have biased our results with regard to pain and psychosocial measures. While self-reported opioid utilization was different between clusters, this measure is limited by the inherent bias associated with patient-reported use of opioids. Nonetheless, opioid utilization mirrored cluster stratification and followed similar patterns over time. Despite these limitations, the heterogeneous and large sample size, use of longitudinal validated repeated measures, and prolonged follow-up period of this study advances current understanding into the development of distinct biopsychosocial symptom trajectories that emerge and persist throughout postoperative recovery.


These findings illustrate the thematic variability in pain impact trajectories that patients experience after undergoing surgical procedures. Notably, this large cohort study identified patient-level clinical characteristics associated with poor outcomes after surgery, using a robust comprehensive battery of biopsychosocial patient-reported outcomes. Cross-sectional preoperative assessments may be inadequate for planning future patient care needs, given the complex multidimensional symptom presentations observed in this large sample throughout the 6-month postoperative recovery period. Further, the formation of trajectory memberships approximately 1 month after surgery underscores the need for repeated assessments across standardized time periods. As such, clinicians will need to consider leveraging prospective longitudinal patient-reported outcome assessments to identify patients on worsening postoperative trajectories and intervene to mitigate the occurrence of poor health outcomes.


The authors thank all the participants for their time and contribution to the study. The investigators also thank the research staff from the Defense and Veteran’s Center for Integrative Pain Management for their work in recruitment, data collection, and data management, including Mary McDuffie, Maria DiMarzio, and Marisa Kinnally.


Name: Michael L. Kent, MD.

Contribution: This author helped with study design, implementation, data interpretation, and manuscript preparation.

Conflicts of Interest: Dr M. L. Kent received an investigator-initiated grant from Pacira Pharmaceuticals Inc, during the conduct of the study. The views expressed in this article are those of the authors and do not reflect the official policy of the Department of Veterans Affairs, Uniformed Services University, the Department of the Army/Navy/Air Force, Department of Defense, the United States Government, or the Henry M. Jackson Foundation for the Advancement of Military Medicine, Inc.

Name: Nicholas A. Giordano, PhD, RN.

Contribution: This author helped with data interpretation and manuscript preparation.

Conflicts of Interest: None.

Name: Winifred Rojas, BSN, RN, CCRP.

Contribution: This author helped implement the study and review the manuscript.

Conflicts of Interest: None.

Name: Mary Jo Lindl, MS, RN.

Contribution: This author helped implement the study and review the manuscript.

Conflicts of Interest: None.

Name: Eugenio Lujan, MD.

Contribution: This author helped design the study, interpret the data, and prepare the manuscript.

Conflicts of Interest: None.

Name: Chester C. Buckenmaier III, MD.

Contribution: This author helped design the study, interpret the data, and prepare the manuscript.

Conflicts of Interest: None.

Name: Raymond Kroma, BS.

Contribution: This author helped implement the study and prepare the manuscript.

Conflicts of Interest: None.

Name: Krista B. Highland, PhD.

Contribution: This author helped with study design, study implementation, primary data analysis, data interpretation, and manuscript preparation.

Conflicts of Interest: None.

This manuscript was handled by: Honorio T. Benzon, MD.


    1. Pagé MG, Katz J, Curtis K, Lutzky-Cohen N, Escobar EM, Clarke HA. Acute pain trajectories and the persistence of post-surgical pain: a longitudinal study after total hip arthroplasty. J Anesth. 2016;30:568–577.
    2. Tracy LM. Psychosocial factors and their influence on the experience of pain. Pain Rep. 2017;2:e602.
    3. Weinrib AZ, Azam MA, Birnie KA, Burns LC, Clarke H, Katz J. The psychology of chronic post-surgical pain: new frontiers in risk factor identification, prevention and management. Br J Pain. 2017;11:169–177.
    4. Edwards RR, Dworkin RH, Sullivan MD, Turk DC, Wasan AD. The role of psychosocial processes in the development and maintenance of chronic pain. J Pain. 2016;17:T70–T92.
    5. Costa Lda C, Maher CG, McAuley JH, Hancock MJ, Smeets RJ. Self-efficacy is more important than fear of movement in mediating the relationship between pain and disability in chronic low back pain. Eur J Pain. 2011;15:213–219.
    6. van Boekel RLM, Vissers KCP, van der Sande R, Bronkhorst E, Lerou JGC, Steegers MAH. Moving beyond pain scores: multidimensional pain assessment is essential for adequate pain management after surgery. PLoS One. 2017;12:e0177345.
    7. Kent ML, Tighe PJ, Belfer I, et al. The ACTTION-APS-AAPM Pain Taxonomy (AAAPT) multidimensional approach to classifying acute pain conditions. J Pain. 2017;18:479–489.
    8. Gordon DB. Acute pain assessment tools: let us move beyond simple pain ratings. Curr Opin Anaesthesiol. 2015;28:565–569.
    9. Houle TT, Miller S, Lang JE, et al. Day-to-day experience in resolution of pain after surgery. Pain. 2017;158:2147–2154.
    10. Pagé MG, Katz J, Romero Escobar EM, et al. Distinguishing problematic from nonproblematic postsurgical pain: a pain trajectory analysis after total knee arthroplasty. Pain. 2015;156:460–468.
    11. Vasilopoulos T, Wardhan R, Rashidi P, et al.; Temporal Postoperative Pain Signatures (TEMPOS) Group. Patient and procedural determinants of postoperative pain trajectories. Anesthesiology. 2021;134:421–434.
    12. Hah JM, Cramer E, Hilmoe H, et al. Factors associated with acute pain estimation, postoperative pain resolution, opioid cessation, and recovery: secondary analysis of a randomized clinical trial. JAMA Netw Open. 2019;2:e190168.
    13. Abrishami A, Chan J, Chung F, Wong J. Preoperative pain sensitivity and its correlation with postoperative pain and analgesic consumption: a qualitative systematic review. Anesthesiology. 2011;114:445–457.
    14. Koorevaar RC, van ‘t Riet E, Gerritsen MJ, Madden K, Bulstra SK. The influence of preoperative and postoperative psychological symptoms on clinical outcome after shoulder surgery: a prospective longitudinal cohort study. PLoS One. 2016;11:e0166555.
    15. Pagé MG, Watt-Watson J, Choinière M. Do depression and anxiety profiles over time predict persistent post-surgical pain? A study in cardiac surgery patients. Eur J Pain. 2017;21:965–976.
    16. Giordano NA, Kane A, Jannace KC, et al. Discrete and dynamic postoperative pain catastrophizing trajectories across 6 months: a prospective observational study. Arch Phys Med Rehabil. 2020;101:1754–1762.
    17. Harris PA, Taylor R, Thielke R, Payne J, Gonzalez N, Conde JG. Research electronic data capture (REDCap)–a metadata-driven methodology and workflow process for providing translational research informatics support. J Biomed Inform. 2009;42:377–381.
    18. Cook KF, Kallen MA, Buckenmaier C III, et al. Evaluation of the validity and response burden of patient self-report measures of the pain assessment screening tool and outcomes registry (PASTOR). Mil Med. 2017;182:e1851–e1861.
    19. Cella D, Choi S, Garcia S, et al. Setting standards for severity of common symptoms in oncology using the PROMIS item banks and expert judgment. Qual Life Res. 2014;23:2651–2661.
    20. Hung M, Baumhauer JF, Licari FW, Voss MW, Bounsanga J, Saltzman CL. PROMIS and FAAM minimal clinically important differences in foot and ankle orthopedics. Foot Ankle Int. 2019;40:65–73.
    21. Kazmers NH, Qiu Y, Yoo M, Stephens AR, Tyser AR, Zhang Y. The minimal clinically important difference of the PROMIS and QuickDASH instruments in a nonshoulder hand and upper extremity patient population. J Hand Surg Am. 2020;45:399–e6.407
    22. Katz P, Kannowski CL, Sun L, Michaud K. Estimation of minimally important differences and patient acceptable symptom state scores for the patient-reported outcomes measurement information system pain interference short form in rheumatoid arthritis. ACR Open Rheumatol. 2020;2:320–329.
    23. Yost KJ, Eton DT, Garcia SF, Cella D. Minimally important differences were estimated for six patient-reported outcomes measurement information system-cancer scales in advanced-stage cancer patients. J Clin Epidemiol. 2011;64:507–516.
    24. Katz P, Pedro S, Alemao E, et al. Estimates of responsiveness, minimally important differences, and patient acceptable symptom state in five patient-reported outcomes measurement information system short forms in systemic lupus erythematosus. ACR Open Rheumatol. 2020;2:53–60.
    25. Lee AC, Driban JB, Price LL, Harvey WF, Rodday AM, Wang C. Responsiveness and minimally important differences for 4 patient-reported outcomes measurement information system short forms: physical function, pain interference, depression, and anxiety in knee osteoarthritis. J Pain. 2017;18:1096–1110.
    26. Chen CX, Kroenke K, Stump TE, et al. Estimating minimally important differences for the PROMIS pain interference scales: results from 3 randomized clinical trials. Pain. 2018;159:775–782.
    27. Purvis TE, Andreou E, Neuman BJ, Riley LH III, Skolasky RL. Concurrent validity and responsiveness of PROMIS health domains among patients presenting for anterior cervical spine surgery. Spine (Phila Pa 1976). 2017;42:E1357–E1365.
    28. Meaningful Change. Accessed June 16, 2021.
    29. Polomano RC, Galloway KT, Kent ML, et al. Psychometric testing of the defense and veterans pain rating Scale (DVPRS): a new pain scale for military population. Pain Med. 2016;17:1505–1519.
    30. Van Damme S, Crombez G, Bijttebier P, Goubert L, Van Houdenhove B. A confirmatory factor analysis of the pain catastrophizing scale: invariant factor structure across clinical and non-clinical populations. Pain. 2002;96:319–324.
    31. Genolini C, Alacoque X, Sentenac M, Arnaud C. kmlandkml3d:RPackages to cluster longitudinal data. J Stat Softw. 2015;65:1–34.
    32. Calinski T, Harabasz J. A dendrite method for cluster analysis. Commun Statistics Theory Methods. 1974;3:1–27.
    33. Singh JA, Lemay CA, Nobel L, et al. Association of early postoperative pain trajectories with longer-term pain outcome after primary total knee arthroplasty. JAMA Netw Open. 2019;2:e1915105.
    34. Okamoto A, Yamasaki M, Yokota I, et al. Classification of acute pain trajectory after breast cancer surgery identifies patients at risk for persistent pain: a prospective observational study. J Pain Res. 2018;11:2197–2206.
    35. Scott W, Wideman TH, Sullivan MJ. Clinically meaningful scores on pain catastrophizing before and after multidisciplinary rehabilitation: a prospective study of individuals with subacute pain after whiplash injury. Clin J Pain. 2014;30:183–190.
    36. Wheeler CHB, Williams ACC, Morley SJ. Meta-analysis of the psychometric properties of the pain catastrophizing scale and associations with participant characteristics. Pain. 2019;160:1946–1953.
    37. Muñoz-Leyva F, El-Boghdadly K, Chan V. Is the minimal clinically important difference (MCID) in acute pain a good measure of analgesic efficacy in regional anesthesia? Reg Anesth Pain Med. 2020;45:1000–1005.
    38. Anderson MR, Houck JR, Saltzman CL, et al. Validation and generalizability of preoperative PROMIS scores to predict postoperative success in foot and ankle patients. Foot Ankle Int. 2018;39:763–770.
    39. Sandvall B, Okoroafor UC, Gerull W, Guattery J, Calfee RP. Minimal clinically important difference for PROMIS physical function in patients with distal radius fractures. J Hand Surg Am. 2019;44:454–e1.459

    Supplemental Digital Content

    Copyright © 2021 International Anesthesia Research Society