Clinically derived early postoperative pain trajectories differ by age, sex, and type of surgery

Abstract

The objective of this study was to determine the effects of age, sex, and type of surgery on postoperative pain trajectories derived in a clinical setting from pain assessments in the first 24 hours after surgery. This study is a retrospective cohort study using a large electronic medical records (EMR) system to collect and analyze surgical case data. The sample population included adult patients undergoing non-ambulatory, non-obstetric surgery in a single institution over a 1-year period. Analyses of postoperative pain trajectories were performed using a linear mixed effects model. Pain score observations (91,708) from 7,293 patients were included in the statistical analysis. On average, the pain score decreased about 0.042 [95% CI: (−0.044, −0.040)] points on the numerical rating scale (NRS) per hour following surgery for the first 24 postoperative hours. The pain score reported by male patients was about 0.27 [95% CI: (−0.380, −0.168)] NRS points lower than that reported by females. Pain scores significantly decreased over time in all age groups, with a slightly more rapid decrease for younger patients. Pain trajectories differed by anatomic location of surgery, ranging from −0.054 [95% CI: (−0.062, −0.046)] NRS units per hour for integumentary and nervous surgery to −0.104 [95% CI: (−0.110, −0.098)] NRS units per hour for digestive surgery, and a positive trajectory (0.02 [95% CI: (0.016, 0.024)] NRS units per hour) for musculoskeletal surgery. Our data support the important role of time after surgery in considering the influence of biopsychosocial and clinical factors on acute postoperative pain.

1. Introduction

Acute postoperative pain occurs in 80% of patients, with 86% of these patients experiencing moderate to severe pain [2]. Given the approximately 234 million surgeries performed worldwide each year, acute postoperative pain is an important public health problem [36]. Acute postoperative pain is often quantified using single pain intensity scores [8]. The “snapshot” nature of this assessment method may limit its usefulness, both managing daily pain and in determining optimal pain regimens for patients at a more systematic level [7,8,19,20,22,23]. Accurately characterizing the postoperative pain course, by characterizing temporal trends in pain scores, is important given uncontrolled acute postoperative pain has been linked to persistent postsurgical pain afflicting between 10% and 50% of surgical patients [24,27].

Pain trajectories have emerged as a novel way to characterize the postoperative pain experience [9]. Using linear models, patients have been categorized into one of three trajectories: those with negative slopes (pain decreases with time), flat trajectories, or positive slopes (pain increases with time). Indeed, certain pain trajectories have been linked to chronification of acute postoperative pain [1]. Prior work involving trajectory calculations has relied on daily averages of pain scores. However, such methodology fails to capture the daily fluctuations in pain with sufficient resolution to assist with clinical decision-making [15,26,29,32]. The increasingly widespread availability of EMR systems creates the opportunity to collect very large numbers of clinical postoperative pain intensity assessments at multiple time points [6]. By using pain assessments obtained from routine postoperative nursing inquiries, the results can be compared with previously reported findings from more tightly controlled research settings to determine the potential for widespread clinical implementation of research findings, as well as allow high-resolution analysis of how acute postoperative pain changes over time.

Prior work by our group has demonstrated that while early postoperative pain scores measured in the first and second hours after surgery correlate highly with each other, these pain scores correlate poorly with pain scores measured later in the recovery period on postoperative days 1 through 5 [34]. The aim of this project was to determine the effects of age, sex, and type of surgery on postoperative pain trajectories derived from a large collection of clinical pain assessments in a mixed surgical population within the first 24 hours after surgery to examine this initial hinge point in the correlation of early versus late pain assessment. We first hypothesized that patients would generally have negative pain trajectories, and that females would exhibit greater initial pain scores and less negative slopes in comparison with males, given prior data suggesting female sex is associated with higher postoperative pain scores [16]. Next, we hypothesized that younger patients would exhibit higher initial pain scores and smaller negative slopes in comparison with older patients given prior findings on pain perception and analgesic pharmacokinetics in older patients [13]. Our third hypothesis was that certain types of surgeries would be associated with pain trajectories with more positive slopes, given prior work demonstrating that spine and orthopedic procedures often result in significant postoperative [18].

2. Methods

2.1 Study design

The Institutional Review Board (IRB-01) at the University of Florida approved this study.

This study is a retrospective cohort study of surgical case data using a large EMR system that was designed to examine the roles of age, sex, and the type of surgical procedure on postoperative pain trajectory in an adult population undergoing a variety of surgical procedures. Results were reported in accordance with the STROBE criteria, a 22-point checklist of recommendations designed to guide researchers toward improved quality of reporting of observational studies in epidemiology (http://www.strobe-statement.org/fileadmin/Strobe/uploads/checklists/STROBE_checklist_v4_cohort.pdf).

2.2 Description of data

Surgical case data were obtained from the University of Florida’s Integrated Data Repository. The Integrated Data Repository is a large database of validated fields obtained from EMR systems used for patient tracking, billing, surgery, and hospital documentation purposes. We chose patients who were aged 21 and over undergoing non-ambulatory surgery at UF Health at the University of Florida over a 1-year period beginning May of 2011. Surgical case exclusion criteria included obstetric surgery, as well as those patients who received multiple separate surgeries within the study period in order to avoid contamination of pain scores from the effects of surgeries preceding or following the case of interest.

All pain scores were documented by clinical staff using the numeric rating scale (NRS) on an 11-point system ranging from 0 to 10. Pain scores were entered using the Epic electronic medical record (EMR) system (Epic Systems Corporation, Verona, WI); this particular implementation provides education on the administration of the NRS query at the point of data entry to improve the veracity of collected data, and constrains the options for data entry to avoid aberrant entries (e.g., pain score of 77 instead of 7). Pain scores were generally recorded every 15 min within the recovery room, and then every 4 h per nursing protocol, with a repeat query within 1 h following administration of analgesic medications for breakthrough pain. When the clinical staff documented a pain score as “patient asleep,” the pain score was converted to a missing value rather than 0/10 to account for the fact that some patients had received additional sedatives that may have facilitated sleep despite an otherwise large nociceptive load. All pain scores were recorded with a corresponding date/time stamp, as were the start and end times of the related surgical procedure. End of surgery times are defined in this system as the application of dressing over the surgical site following wound closure. For each pain score, the time in minutes following the end of surgery was calculated to enable a stable reference point from which to calculate pain trajectories. Pain scores were filtered to include only those obtained after the listed end-surgery time through 24 h after the documented end of surgery. In this system, analgesic protocols were not adjusted for adults based upon age or sex, and opioid therapies generally followed hospital-based protocols that were not surgery specific. The use of multimodal analgesics was surgeon-dependent with the exception of regional anesthesia, which was used in the vast majority of orthopedic surgical patients. Pain scores were restricted to the first 24 h after surgery to focus on early opportunities to modify the ensuing pain trajectories on postoperative days 2 through 7; this restriction had the salutary benefit of minimizing the effect of censuring within the mixed-model framework described below. Specific details on the type of anesthesia and analgesic medications were not available, although patients within each type of surgery were generally cared for within specific analgesic frameworks that included or precluded the use of neuraxial, para-neuraxial, and perineural catheters and/or single-injection nerve blocks on the basis of the type of surgery.

Types of surgery were identified using current procedural terminology (CPT) codes published by the American Medical Association. Given the large number of CPT codes, surgeries were grouped into separate categories based upon their anatomic location according to the first digits of the CPT code. For patients with multiple CPT codes, only the primary CPT code was used for definition of the type of surgery. The total number of CPT codes used to describe each surgery was included as an index of the complexity of the surgical procedure. Patient comorbidity was scored by first extracting up to 50 comorbid diagnoses as coded by the International Classification of Disease 9, Clinical Modification (ICD-9-CM) for each patient. The ICD-9-CM codes were then converted into a Charlson Comorbidity Score [10]. Sex was defined as male or female. Age was defined in separate models as a continuous variable.

Observations with missing pain scores or missing measurement time points for pain scores were removed from the analysis set. Additionally, anatomic types of surgery with less than 300 patients within the cohort were removed from the analysis set.

2.3 Statistical analysis

Analyses of postoperative pain trajectories were performed using the following linear mixed-effects model framework. A mixed-effects model was used because of the multiple continuous NRS pain intensity ratings reported from each patient, where the pain intensity ratings reported within each patient are correlated. This allowed us to examine within-patient changes of pain scores rather than simply examine changes in pain scores across the aggregate sample. We used the following model:

$${\mathrm{Y}}_{\mathrm{ij}}={\mathrm{\alpha }}_0+{{\mathrm{\alpha }}_1}^{\ast }{\mathrm{t}}_{\mathrm{ij}}+{\mathrm{\alpha }}_2{t}_{\mathrm{i}}+{{\mathbf{X}}_{\mathrm{ij}}}^{\ast }\mathbf{\alpha }+{\mathrm{u}}_{\mathrm{i}}+{\mathrm{\epsilon }}_{\mathrm{ij}}$$

where Y_ij represents the pain score measured for subject i at the jth measurement. t_ij is the actual time (in the unit of hours) after the surgery for the pain score being measured for subject i at the jth measurement. u_i ~ N(0,ϭ²) is the unobserved random cluster effect due to subject i representing a source of variation and the heterogeneity for the pain score from the patient i, i.e., each patient may have his/her specific feelings of the pain that follows a normal distribution with mean zero which is a common assumption made in a traditional mixed-effects model that is used to describe the correlated responses [25]. X_ij are all other observed covariates listed in Table 1. ε_ij ~ N(0,σ²) is the random measurement error and _i is the average time from end of surgery to NRS measurement (in the unit of hours) for subject i. This term is added to relax the independence assumption made in the traditional linear mixed effects model that the random effects (e.g., u_i) is independent of the fixed-effects covariates (e.g., t_ij in this study). This assumption rarely holds in practice. Adding this term makes the traditional mixed-effects model (i.e., the above model without _i) a special case that will allow one to obtain the consistent fixed effects parameter estimates regardless of whether the independence assumption holds or not. Starting with a grand full model by including all the observed covariates listed in Table 1, a model selection process is conducted via likelihood ratio test to obtain the optimal model deemed for the data. A residual Q-Q plot is obtained and it follows the theoretical normal distribution reasonably well as shown in the Supplementary Materials [14].

Table 1.

Variable	Category	Statistic
Gender	Male	3740 (51.28%)
	Female	3553 (48.72%)
Anatomic Type of Surgery	Cardiovascular	993 (13.62%)
	Digestive	1727 (23.68%)
	Integumentary	582 (7.98%)
	Musculoskeletal	1949 (26.72%)
	Nervous	1150 (15.77%)
	Pulmonary	281 (3.85%)
	Urinary	611 (8.38%)
BMI Group	Morbidly obese	505 (6.92%)
	Normal	1507 (20.66%)
	Obese	1766 (24.22%)
	Overweight	1664 (22.82%)
	Underweight	1007 (13.81%)
	Unknown	844 (11.57%)
Age Group	21–39	1266 (17.36%)
	40–64	3469 (47.57%)
	65–84	2378 (32.61%)
	>85	180 (2.47%)
No. of ICD9 Comorbidity Diagnoses	>20	596 (8.17%)
	10–19	2444 (33.51%)
	5–9	2523 (34.59%)
	<5	1730 (23.72%)
No. of CPT Codes to Describe Surgery	>5	109 (1.49%)
	3–5	1192 (16.34%)
	<3	5992 (82.16%)
Charlson Comorbidity Index	<3	6403 (87.80%)
	3–6	858 (11.76%)
	7–10	12 (0.16%)
	Unknown	20 (0.27%)

In addition to the consideration of the correlation among repeated measurements, our model relaxes the independence assumption between the random cluster effects and fixed-effects covariates by introducing the average time measured for each subject. This modeling scheme allows for the small and unequal spacing between repeated measurements, which are an expected feature of clinically acquired pain score observations. The need to include the extracted average time term per subject and/or cluster was tested via likelihood ratio statistics with a degree of freedom of one.

To investigate the association between the pain score and measurement time and other covariates, we used a linear mixed-effects model as mentioned before in which pain score was the outcome variable. We first fitted an oversaturated model (full model) that included time (t_ij), age, gender, average time (t_i,), Charlson comorbidity index, the total number of coded comorbidities, the total number of CPT codes, body mass index category, type of surgery by anatomic location, i.e. all the observed covariates summarized in Table 1, and their mutual interactions, and the random cluster effects of the subject. We then used the log-likelihood ratio test to select the optimal model deemed for the data. The final selected optimal model included the following covariates: intercept, time (t_ij), age, gender, average time (t_i,), Charlson comorbidity index, total number of ICD9-coded comorbidities, total number of CPT codes used to describe the surgical procedure, the anatomic type of surgery, and age and gender interaction. The need of average time term (t_i,) in the model was justified by likelihood ratio test with a P value of <10⁻¹⁰.

Given the large number of observations considered, P < 0.01 was chosen for statistical significance. All analyses were conducted using the open source statistical analysis software R [33] with the lme4 package for the mixed-effects model fitting (http://cran.r-project.org/web/packages/lme4/lme4.pdf).

3. Results

3.1 Demographics

Raw data included 389,617 records with 8,346 patients. However, after removing observations with missing pain scores or missing measurement time points for pain score and those surgical procedures with less than 300 cases per type of surgery, a linear mixed model of 91,708 pain score observations from 7,293 patients was made. Patient ages ranged from a minimum 21 years to a maximum 97 years (mean ± SD, 56.4 ± 16.4 years). Because there were only 180 patients in the 85 year and greater age group, we combined groups 65 to 84 years and >85 year into ≥65 years (Table 1).

3.2 Trajectory model

The model for postoperative pain trajectory suggests that after controlling for other covariates at the same level, on average, the pain score decreased about 0.042 points on the NRS per hour [95% CI: (−0.044, −0.040)] following the surgery (Figure 1). Similarly, after controlling for other covariates at the same level, including time after surgery, the pain score reported by male patients was about 0.27 NRS points [95% CI: (−0.380, −0.168)] lower than that reported by females (Table 2). With every year increase in age, the initial pain score decreased by 0.053 NRS points [95% CI: (−0.057, −0.049)] while controlling for other factors, including time after surgery at the same level (Figure 2).

Fig. 1.

Distribution of postoperative pain trajectory slopes for cohort of analyzed surgical patients. The distribution of slopes ranged from −6.3 NRS units per hour to 3.5 NRS units per hour, with a median of −0.023 NRS units per hour. Notably, of 7,293 subjects, only 22 had an NRS trajectory slope of less than −1 NRS units per hour, and only 15 had an NRS trajectory slope of greater than 1 NRS units per hour.

Table 2.

Parameter Estimates for Age, Sex, and Time after Surgery

Variable	Estimate	StdErr	P Value
Intercept	10.263	1.561	<10−10
tij	−0.042	0.001	<10−10
ageij	−0.053	0.002	<10−10
Genderi (male)	−0.274	0.054	3.3 × <10−7
ti	−0.054	0.008	<10−10
CCI Group (vs 0)	1: −0.246	1: 0.069	1: 3.56 × 10−4
	2: −0.299	2: 0.087	2: 6.26 × 10−4
	3: −0.534	3: 0.122	3: 1.18 × 10−5
	4: −0.317	4: 0.170	4: 0.062
	5: −0.429	5: 0.294	5: 0.145
	6: −1.018	6: 0.579	6: 0.079
	7: −1.466	7: 0.726	7: 0.043
	9: 0.472	9: 2.343	9: 0.840
TotalCptCount (vs 1)	2: 0.300	2: 0.068	2: 9.69 × 10−6
	3: 0.174	3: 0.089	3: 0.050
	4: 0.731	4: 0.145	4: 4.18 × 10−7
	5: 0.753	5: 0.205	5: 2.38 × 10−4
	6: 1.419	6: 0.264	6: 7.98 ×x 10−8
	7: 1.397	7: 0.540	7: 0.010
	8: 0.781	8: 0.746	8: 0.295
	9: −0.006	9: 1.138	9: 0.996
	10: 0.387	10: 1.564	10: 0.805
Anatos (vs Cardiovascular)	Digestive: 0.472	Digestive: 0.096	Digestive: 9.97 × 10−7
	Integumentary: 0.828	Integumentary: 0.124	Integumentary: 2.48 × 10−11
	Musculoskeletal: 0.540	Musculoskeletal: 0.094	Musculoskeletal: 1.09 × 10−8
	Nervous: 0.243	Nervous: 0.104	Nervous: 0.019
	Pulmonary: −0.155	Pulmonary: −0.156	Pulmonary: 0.320
	Urinary: 0.405	Urinary: 0.121	Urinary: 0.001

Fig. 2.

Overview of early postoperative pain trajectories by sex (A), age (B), and type of surgery (C). Group comparisons within each stratification for sex, age, and type of surgery visually depict differences in the y-intercept and slopes of early postoperative pain trajectories measured from clinical pain assessments within the first 24 hours after surgery. NRS = Numerical Rating Scale.

3.3 Stratification by sex

We conducted the same analysis using the same model without separation for male and female patients to better characterize the effects of sex on postoperative pain trajectories (Table 3). Our data reveal that for male and female surgical patients, the pain score decreased with time after surgery [−0.041 NRS points per hour [95% CI: (−0.045, −0.037)] for male surgical patients (P < 1 × 10E − 10); −0.043 NRS points per hour [95% CI: (−0.047, −0.039)] for female surgical patients (P < 1 × 10e – 10); Figure 3]. Furthermore, for both male and female surgical patients, increases in age were associated with more negative postoperative pain trajectories [−0.056 NRS points [95% CI: (−0.062, −0.050)] per year of age for males (p < 1 × 10E − 10); −0.05 NRS points [95% CI: (−0.056, −0.044)] per year of age for females (P < 1 × 10E − 10)]. This reflects how the pain score will change per year of age increase, at exactly the same time point, within the sex stratification.

Table 3.

Postoperative Pain Trajectories by Sex

Gender	Variable	Estimate	StdErr	P value
Male	Intercept	10.089	2.185	3.9 × 10−6
	tij	−0.041	0.002	<10−10
	ageij	−0.056	0.003	<10−10
	ti	−0.031	0.012	7.9 × 10ˆ−3
Female	Intercept	10.35	2.216	3.0 × 10−6
	tij	−0.043	0.002	<10−10
	ageij	−0.05	0.003	<10−10
	ti	−0.079	0.012	< 10−10

Fig. 3.

Differences in postoperative pain trajectory slopes for male and female patients. NRS/Hour = Change in Numerical Rating Scale units per hour after surgery. The mean and standard error are plotted for each group.

3.4 Stratification by age group

Similar to the approach of stratification by sex, we re-stratified the cohort according to the following age groups: 21 to 39 years, 40 to 64 years, and ≥65 years (Table 4). Pain scores significantly decreased over time in all age groups, with a slightly more rapid decrease for younger patients (pain trajectory slope of −0.046 NRS units per hour [95% CI: (−0.052, −0.040)] in the 21 to 39 year age group, −0.042 [95% CI: (−0.046, −0.038)] in the 40 to 64 year age group, and −0.041 [95% CI: (−0.045, −0.037)] in the ≥65 year age group, P < 1E − 10 for each group; Figure 4). When the effect of sex was considered within each age group, male sex was associated with lower pain scores compared with female sex only in the ≥65 year age group (−0.459 NRS units in males versus females, P = 4.8E − 7). This effect was roughly equivalent to the same decrement seen 11 h after surgery in this age group based upon the parameter estimates of time versus gender.

Table 4.

Postoperative Pain Trajectories by Age Group

Age Group	Variable	Estimate	StdErr	p-value
21–39	Intercept	8.174	2.086	8.9 × 10−5
	tij	−0.046	0.003	<10−10
	Genderi (male)	−0.143	0.121	0.236
	ti	−0.028	0.021	0.17
40–64	Intercept	4.986	0.28	<10−10
	tij	−0.042	0.002	<10−10
	Genderi (male)	−0.185	0.08	0.021
	ti	−0.062	0.012	6.3 × 10−7
≥65	Intercept	6.69	2.16	2.0 × 10−3
	tij	−0.041	0.002	<10−10
	Genderi (male)	−0.459	0.091	4.8 × 10−7
	ti	−0.05	0.013	1.4 × 10−4

Fig. 4.

Differences in postoperative pain trajectory slopes by age group. NRS/Hour = Change in Numerical Rating Scale units per hour after surgery. The mean and standard error are plotted for each group.

3.5 Stratification by type of surgery

Similar to the approach of stratification by sex, we re-stratified the cohort according to the following types of surgery: cardiovascular, digestive, integumentary, musculoskeletal, nervous, pulmonary, and urinary (Table 5). The effect of time after surgery was significant for all types of surgery (P < 1E − 10), and negative in slope ranging from −0.054 NRS units per hour [95% CI: (−0.062, −0.046)] for integumentary and nervous surgery to −0.104 NRS units per hour [95% CI: (−0.110, −0.098)] for digestive surgery. An important exception was musculoskeletal surgery, with a positive trajectory of 0.02 [95% CI: (0.016, 0.024)] units per hour (Figure 5). Although the effect of age, after controlling for postoperative pain trajectory, on postoperative pain scores was consistent across all types of surgery, the effects of sex were not. Male patients had statistically significant lower pain scores than female patients only for cardiovascular (−0.447 (95% CI: (−0.725, −0.169)), P = 1.6E − 3), digestive (−0.326 [95% CI: (−0.549, −0.103)], P = 4.4E − 3), and nervous (−0.357 [95% CI: (−0.626, −0.088)], P = 0.009) types of surgery. There were no types of surgery for which male patients had greater postoperative pain scores than female patients.

Table 5.

Postoperative Pain Trajectories by Type of Surgery

Type of Survey	Variable	Estimate	Lower 95% CI	Upper 95% CI	P value
Cardiovascular	Intercept	10.742	6.71028	14.77372	1.8 × 10−7
	tij	−0.055	−0.06088	−0.04912	<10−10
	ageij	−0.054	−0.0638	−0.0442	<10−10
	Genderi (male)	−0.447	−0.72532	−0.16868	1.6 × 10−3
	ti	0.007	−0.02632	0.04032	0.702
Digestive	Intercept	10.825	6.40324	15.24676	1.6 × 10−6
	tij	−0.104	−0.10988	−0.09812	<10−10
	ageij	−0.052	−0.05984	−0.04416	<10−10
	Genderi (male)	−0.326	−0.54944	−0.10256	4.4 × 10−3
	ti	−0.022	−0.05532	0.01132	0.196
Integumentary	Intercept	6.827	5.4844	8.1696	<10−10
	tij	−0.054	−0.06184	−0.04616	<10−10
	ageij	−0.041	−0.05276	−0.02924	<10−10
	Genderi (male)	−0.104	−0.45484	0.24684	0.561
	ti	0.005	−0.05772	0.06772	0.871
Musculoskeletal	Intercept	7.173	6.44976	7.89624	<10−10
	tij	0.02	0.01608	0.02392	<10−10
	ageij	−0.061	−0.06688	−0.05512	<10−10
	Genderi (male)	−0.187	−0.39084	0.01684	0.071
	ti	−0.072	−0.10924	−0.03476	1.5 × 10−4
Nervous	Intercept	7.658	6.79168	8.52432	< 10−10
	tij	−0.054	−0.05988	−0.04812	<10−10
	ageij	−0.046	−0.0558	−0.0362	<10−10
	Genderi (male)	−0.357	−0.62552	−0.08848	0.009
	ti	−0.058	−0.10504	−0.01096	0.015
Pulmonary	Intercept	8.201	6.31548	10.08652	<10−10
	tij	−0.094	−0.10576	−0.08224	<10−10
	ageij	−0.049	−0.06664	−0.03136	3.3 × 10−8
	Genderi (male)	0.347	−0.18416	0.87816	0.2
	ti	0.009	−0.06156	0.07956	0.791
Urinary	Intercept	9.223	8.29396	10.15204	<10−10
	tij	−0.062	−0.0718	−0.0522	<10−10
	ageij	−0.058	−0.06976	−0.04624	<10−10
	Genderi (male)	−0.127	−0.47196	0.21796	0.469
	ti	−0.074	−0.12888	−0.01912	0.008

Fig. 5.

Differences in postoperative pain trajectory by type of surgery. NRS/Hour = Change in Numerical Rating Scale units per hour after surgery. The mean and standard error are plotted for each group.

4. Discussion

Our results confirm prior findings that postoperative pain had a generally negative trajectory with a wide variance. However, our findings differed from prior reports by suggesting that while the effects of age, sex, and type of surgery were generally associated with postoperative pain after controlling for the effect of time after surgery, the effects of sex were quite variable within certain stratifications for age group and type of surgery.

Our results suggest the effect of sex on pain trajectories may be conditional depending on the age group and the type of surgery. In the whole-model analysis, male patients had a substantially more rapid decrease in postoperative pain compared with female patients, even after controlling for age, comorbidity status, type of surgery, and surgical complexity. These findings extend prior work on sex differences in static pain scores in both experimental and clinical settings [4,16,21,31,35]. However, in age-stratified analyses, the effect of sex on postoperative pain was significant only in those greater than or equal to 65 years of age. Likewise, when examining each type of surgery, the effect of sex on postoperative pain was negligible for integumentary, musculoskeletal, pulmonary, and urinary surgery. Such differences could conceivably point to an interaction between sex and type of surgery, which we were unable to explore despite the large number of observations available. Alternatively, it may be that for those surgeries where the effect of sex was negligible, the procedures resulted in nociceptive levels sufficiently high to abolish the effect of sex.

Our results did point to a consistent effect of age on postoperative pain trajectories as well as postoperative pain at any given time point after surgery. In our whole-model analysis, each year of increasing age leads to lower postoperative pain scores after controlling for the effects of time after surgery, sex, comorbidity status, type of surgery, and surgical complexity. These effects were maintained when considering male and female patients separately. However, when examining each age group, younger patients in the 21 to 39 year age group had slightly more negative postoperative pain trajectories than patients in the 40 to 64 and ≥65 year categories. Examining the marginal mean estimate of NRS at time zero, these data suggest that older patients may start with a lower postoperative pain score, but that their postoperative pain resolves at a slower rate. This finding may help explain the conflicting data on the effect of age on postoperative pain [3,5,17,28,30].

No association emerged between comorbidity and surgical complexity on postoperative pain trajectories, although specific comorbidities may have different effects on postoperative pain. An obvious example would be the presence of chronic pain conditions requiring the use of high-dose systemic opioids prior to surgery that would increase the likelihood of persistent postsurgical pain. It is quite possible that more specific, targeted measures of both pain-related comorbidities and surgical complexity would have offered improved insight into the role of these factors in postoperative pain trajectories by teasing out specific mechanistic attributes that were not diluted by a plethora of other comorbidity and complexity attributes that do not alter pain trajectories. Altogether, these results suggest the need to consider additional clinical and biopsychosocial factors, in a more targeted fashion, that are not routinely captured by the EMR.

Information on postoperative pain trajectories may offer a wealth of support in the clinical decision-making process. For instance, early postoperative trajectories that are non-negative may point to opportunities to use extended-release/long-acting formulations of opioid and non-opioid analgesics. Additionally, anesthesiologists may wish to include perioperative regional anesthetics, especially considering those options that include longer-acting local anesthetics and/or continuous catheter-based interventions. Patients in the highest risk categories for postoperative pain trajectories can have earlier disposition arrangements to rehabilitation centers rather than home care environments to assist with continued optimization of pain management and functional status in a supervised environment.

The use of pain trajectories to describe postoperative pain is an emerging and innovative way to characterize acute postoperative pain [1,9]. Chapman introduced the use of postoperative pain trajectories with a prospective observational approach using initial pain intensity (intercept) and rate of pain resolution (slope) to describe the course of pain over a period of 6 days. Patients provided daily reports of pain, and once discharged, provided pain scores at their convenience. Using completed NRS data for 502 patients, these authors noted a similar effect of age and gender on pain trajectories as in our study. Their surgical cases were categorized by surgical site instead of organ system, and they noted that chest surgery resulted in the highest initial pain intensity with the quickest decline in pain intensity over time. In our study, surgery of the digestive system resulted in the steepest decline in pain. An additional difference included the issue of preexisting chronic pain; patients with a history of treatment for chronic pain were excluded from Chapman’s report, whereas no such exclusion occurred in our study. The presence of chronic pain may influence pain scores and trajectories in that these patients may have lower thresholds for pain perception, increased tolerance to opioids, and opioid-induced hyperalgesia, all of which would result in higher initial pain scores and pain trajectories with positive slopes or less negative slopes [11,12]. This will need to be examined in greater detail in future studies, as our current model did not include information on home opioid use and/or histories of chronic pain.

Work by Althaus emphasized the use of acute pain trajectories in identifying patients at risk for chronic postsurgical surgical pain (CPSP) at 6 months, noting that higher initial pain intensity and slower resolution of acute pain were predictors for CPSP. The authors noted that patients with anxiety had higher initial pain scores and steeper declines in pain, whereas patients with depression had slightly lower initial pain scores with slower declines in pain. Neither anxiety nor depression was associated with CPSP at 6 months. The works of Chapman and Althaus highlight the important role that time plays in predicting postoperative pain, and how time can modulate the role of other important predictive factors.

Our study differed in many ways from the studies by Chapman and Althaus. Our sample size was significantly greater than either study, by an order of magnitude, through our use of a large clinical data repository. In addition, values for pain were clinically derived by routine nursing care at our institution; therefore, NRS data included a time stamp that allowed for accurate collection of pain scores as a function of time with the hours as the unit of measurement. In addition, our study examined the potential effects of the Charlson Comorbidity index, a measure of comorbidity, as well as the relative complexity of surgery on postoperative pain trajectories [10]. Our results offered a high-resolution analysis of how postoperative pain scores change within the earliest postoperative time epochs.

Although data gathered prospectively in the context of a clinical research protocol may allow more accuracy in data collection, such a resource-intensive approach can limit sample size and clinical generalizability of findings. However, clinical data may be inaccurate based on the heterogeneity of backgrounds of medical providers who are performing the pain assessments, as well as the patients’ understanding of the NRS scale. These additional sources of variability in existing clinical data point to a growing need to reconcile clinical observations with experimental findings on a more robust scale. Nevertheless, applied analytics for quality improvement will inevitably rely upon clinical rather than experimentally derived data, and the gap between clinical as opposed to experimentally derived data will need to be bridged in future research.

The limitations of our study include the shortened time frame of data acquisition, which is limited by the date of initiation of our electronic record system. Despite this, we were able to collect data at a higher resolution than in prior studies. However, there were an insufficient number of patients to allow assessment of more specific surgery types and to better delineate postoperative pain trajectories in the greatest age groups. In addition, our mixed model fails to fully capture differences due to pain assessments at irregular intervals, and it is possible that more advanced modeling methods, including Markov models and dynamic time warping, could offer additional insights into the observed time-domain variance. Our dataset did not include sufficient information on the dose, timing, and route of administration of perioperative analgesics, thus this information was not included in our analyses. This is an important limitation that will need to be investigated in greater detail with more complicated and dedicated analyses given that pain intensity ratings and analgesic medications have alternating causal relationships over multiple windows of time.

In conclusion, our data support the important role that time plays after surgery in considering the influence of biopsychosocial and clinical factors on acute postoperative pain. Future work is necessary to determine how these very early differences in postoperative pain trajectories lead to the resolution of acute postoperative pain beyond postoperative day 1. Furthermore, much work is necessary to examine the agreement between clinically and experimentally derived measures of postoperative pain trajectories.