Acute postoperatieve pijn is een van de meest gemelde postoperatieve klachten. Identificatie van de prognostische factoren voor postoperatieve pijn kan resulteren in een snellere en betere pijnbestrijding als deze prognostische factoren geïdentificeerd kunnen worden.

Description of studies

1. Acute postoperative pain

Berghmans (2018), Rabbits (2015) and Rabbitts (2020) presented multivariable models describing associations of demographics, psychological variables, preoperative pain and surgery-related factors with acute postoperative pain. Details of the studies are shown in Table 1.

Table 1. Description of studies reporting unvalidated models of acute postoperative pain

CBCL, Child Behavior Checklist; HRQOL, Health-related quality of life; PPPM, parents postoperative pain measure

2. Chronic postoperative pain

Julien-Marsollier (2017), Mimura (2019), Rabbits (2020) and Siemer (2020) presented multivariable models describing associations of demographics, psychological variables, preoperative pain and surgery-related factors with postoperative pain. Details of the studies are shown in Table 2.

Table 2. Description of studies reporting unvalidated models of chronic postoperative pain

HRQOL, health-related quality of life; PedsQL, Pediatric Quality of Life; PROMIS, Pediatric patient-reported outcome measurement system

Results

1. Acute Postoperative pain

1.1         Model performance

Rabbitts (2020), Rabbits (2015) and Berghmans (2018) presented multivariable models describing associations of demographics, psychological variables, preoperative pain and surgery-related factors with acute postoperative pain. Details of the presented models are summarized in Table 3.

Berghmans (2018) constructed two multivariate logistic regression models to assess whether internalizing and externalizing problems explained pain at home during first 3 days after surgery. In the first model, parental state anxiety (β 0.209; P=0.017) and parental need for information (β 0.209; P=0.046) were statistically significant predictors of pain scores during the first 3 days after surgery. In the second model, internalizing and externalizing problems were entered in the regression model. In this second model, parental need for information (β 0.207; P=0.01) and internalizing problems (β 0.343; P=0.001) were statistically significant predictors of pain scores during the first 3 days after surgery. Overall, model 1 explained 9.7% (P=0.03) of the variance of pain score at home and model 2 explained 22.5% (P=0.000) of the variance of pain scores at home.

Rabbitts (2015) conducted a multivariate logistic regression model to assess whether psychosocial and behavioral factors before surgery predict acute postoperative pain 2 weeks after surgery. Baseline pain intensity (β = 0.44; P<0.001), parental pain catastrophizing (β = 0.28, P<0.05) and presurgery sleep duration (β = −0.26; P<0.05) were statistically significantly associated with mean pain intensity reported 2 weeks after surgery. Overall, the model explained 33.8% of the variance of mean pain intensity 2 weeks after surgery.

Rabbitts (2020) examined associations between demographic factors, baseline pain, adolescent psychological/behavioral, and parent/family factors before surgery with presence of acute postoperative pain 2 weeks after surgery. The base model (Model 1) included baseline demographic variables only. Then baseline pain intensity and pain distribution (Model 2), adolescent emotional and behavioral factors (Model 3), and parent and family factors (Model 4) were added into the models sequentially, leading to four nested regression models. Baseline pain intensity was the only statistically significant predictor of acute postoperative pain in the final model (OR model 4: 1.96; 95% CI 1.32 to 2.90). This means that for each 1 point higher mean pain intensity rating during the week preceding surgery, the odds of acute postoperative pain was 96% higher at 2-weeks following surgery (P<0.001). Unfortunately, the overall performance of the models was not reported. Only Likelihood Ratio test statistic of the models were reported, whereby each successive model was compared to the previous model (i.e. M2 vs. M1, M3 vs M2, and M4 vs. M3). However, this outcome does not reflect the overall performance of the models to predict postoperative pain.

Table 3. Unvalidated models predicting acute postoperative pain in children undergoing surgery

APAIS, Amsterdam Perioperative Anxiety Information scale-information part; LR, Likelihood Ratio test statistic; mYPAS, modified Yale Preoperative Anxiety scale

*P<0.05

**P<0.01

2. Chronic postoperative pain

2.1         Model performance

Julien-Marsollier (2017), Mimura (2019), Rabbits (2020) and Siemer (2020) presented multivariable models describing associations of demographics, psychological variables, preoperative pain and surgery-related factors with postoperative pain. Details of the presented models are summarized in Table 4.

Julien-Marsollier (2017) identified risk factors for neuropathic pain one year after scoliosis children. Only the factors independently associated with the occurrence of neuropathic pain at 1 year after surgery in the univariate model were entered in a multivariate model. In the multivariate model, preoperative pain during the 3 months preceding surgery (β 2.5; SD 1.2; P=0.003) and total morphine at day 1 ≥0.5 mg kg−1 (β 3.7; SD 1.2; P=0.002) were statistically significant predictors of neuropathic pain at 1 year after surgery. The area under the curve of the model was 0.89 (95% CI 0.78–0.99).

Mimura (2019) identified risk factors for postoperative pain two year after surgery for adolescent idiopathic scoliosis. Only the factors independently associated with the postoperative pain at 2 year after surgery in the univariate model were entered in a multivariate general linear model. In the multivariate model, higher preoperative pain (β 0.30; SD 0.09; P < 0.01) and higher postoperative T5-12 kyphotic angle (β -0.12; SD 0.05; P = 0.02) were statistically significant predictors of chronic postoperative pain at 2 year after surgery. Overall, the model explained 17% of the variance of pain 2 years after surgery.

Rabbits (2020) examined associations between demographic factors, baseline pain, adolescent psychological/behavioral, and parent/family factors before surgery with chronic postsurgical pain 4 months after surgery. The base model (Model 1) included baseline demographic variables only. Then baseline pain intensity and pain distribution (Model 2), adolescent emotional and behavioral factors (Model 3), and parent and family factors (Model 4) were added into the models sequentially, leading to four nested regression models. Only addition of adolescent psychological and behavioral factors increased model fit (M3 vs M2 LR=17.02). In the final model (M4), depressive symptoms (OR 1.22; 95% CI 1.01 to 1.47) and sleep quality (OR 0.26; 95% CI 0.08 to 0.83) were the only statistically significant predictors of chronic postoperative pain. This means that for each 1 point higher level of depressive symptoms on the Revised Child Anxiety and Depression Scale (RCADS) before surgery, the odds of chronic postoperative pain was 22% higher at 4-months following surgery (P=0.03). Each 1 point higher sleep quality score on the Adolescent Sleep Wake Scale (ASWS) before surgery was associated with 74% lower odds of developing chronic postoperative pain at 4-months following surgery (P=0.02).

Siemer (2020) identified the associations between parent factors and child symptomatology and long-term pain interference 1 year after surgery. A 2-step model was performed, whereby in the 1^st step only parent related factors were included in a multivariate model, and in the 2^nd step children related factors were added to this model. In the first model, parent catastrophizing (β 0.18; 95% CI 0.04 to 0.32; P=0.14) and parent pain relief preference (β 0.51; 95% CI 0.02 to 0.99; P=0.04) were statistically significant associated with child’s long-term pain interference score. This model explained 20% of the variance of pain 1 year after surgery. In the second model, only child high symptom profile (β 7.98; 95% CI 4.72 to 11.19; P<0.001) was statistically significant associated with child’s long-term pain interference score. This model explained 43% of the variance of pain 1 year after surgery.

Table 4. Unvalidated models predicting chronic postoperative pain in children undergoing surgery

LR, Likelihood Ratio test statistic; PROMIS, Pediatric patient-reported outcome measurement system

*P<0.05

**P<0.01

Level of evidence of the literature

The level of evidence was not graded, because none of the included studies reported validated models.

Preferably a systematic review of the literature measuring the effect of using a clinical decision model to identify children undergoing surgery with a high risk for developing (severe) postoperative pain, compared to standard care or chance. As such research is very rare and the working group did not expect to find such studies, a systematic review of the literature was performed to answer the following questions:

1. What is the prognostic value of a prediction model for developing acute postoperative pain in children undergoing surgery?

P: Children 0 to 18 years old undergoing surgery

I:  Prediction model:

Outcome: acute postoperative pain (for each type of surgery, pain within 2 weeks after surgery)

Factors in 4 different domains:

1. Demographics: age, gender, education level, ethnic group
2. Psychological: anxiety (patient and caregiver preoperative anxiety state), psychological distress (patient’s and caregiver’s mood, affect, personality trait, anxiety trait), coping strategies of patient and caregiver, history of physical, emotional, or sexual abuse.
3. Preoperative pain: preoperative pain/analgesic experience, patient’s perception regarding pain or analgesia, pain threshold.
4. Surgery-related factors: type of surgery, previous surgery

C: Absence of prognostic model/ chance

O: Model performance (predictive value)

T: Pre-operative

S: In hospital, outpatient clinic

2. What is the prognostic value of a prediction model for developing chronic postoperative pain in children undergoing surgery?

P: Children 0 to 18 years old undergoing surgery

I: Prediction model:

Outcome: chronic postoperative pain (for each type of surgery, pain lasting longer than 2 months after surgery)

Factors in 4 different domains:

1. Demographics: age, gender, education level, ethnic group
2. Psychological: anxiety (patient and caregiver preoperative anxiety state), psychological distress (patient’s and caregiver’s mood, affect, personality trait, anxiety trait), coping strategies of patient and caregiver, history of physical, emotional, or sexual abuse.
3. Preoperative pain: preoperative pain/analgesic experience, patient’s perception regarding pain or analgesia, pain threshold.
4. Surgery-related factors: type of surgery, previous surgery

C: Absence of prognostic model/ chance

O: Model performance (predictive value)

T: Pre-operative

S: In hospital, outpatient clinic

Relevant outcome measures

The guideline development group considered model performance (predictive value) as a critical outcome measure for decision making.

Acute postoperative pain was defined as pain within 2 weeks after surgery.

Chronic postoperative pain was defined as pain lasting longer than 2 months after surgery.

The working group defined the performance of the included models as follows (Cohen, 1988):

R² < 0.02: very weak;

0.02 ≤ R² < 0.13: weak;

0.13 ≤ R² < 0.26: moderate;

R² ≥ 0.26: substantial

0.7≤AUC<0.8: acceptable;

0.8≤AUC<0.9: excellent;

AUC≥0.9: outstanding.

Prognostic research: Study design and hierarchy

When reviewing literature, there is a hierarchy in quality of individual studies. Preferably, the effectiveness of a clinical decision model is evaluated in a clinical trial. Unfortunately, these studies are very rare. If not available, studies in which prediction models are developed and validated in other samples of the target population (external validation) are preferred as there is more confidence in the results of these studies compared to studies that are not externally validated. Most samples do not completely reflect the characteristics of the total population, resulting in deviated associations, possibly having consequences for conclusions. Studies validating prediction models internally (e.g. bootstrapping or cross validation) can be used to answer the first research question as well, but downgrading the level of evidence is obvious due to risk of bias and/or indirectness as it is not clear whether models perform sufficiently in target populations. The confidence in the results of unvalidated prediction models is very low. Therefore, such models will not be graded. This is also applicable for association models. The risk factors identified from such models can be used to inform patients about the elevated risk on developing postoperative pain after surgery, however they are less suitable to be used in clinical decision making.

Search and select (Methods)

The databases Medline (via OVID) and Embase (via Embase.com) were searched with relevant search terms from 2007 until 9 November 2021. The detailed search strategy is depicted under the tab Methods. The systematic literature search resulted in 506 hits. Studies were selected based on the following criteria:

• Being a systematic review, randomized controlled trial (RCT) or observational study (cohort study).
• Reporting multivariable longitudinal association model or prediction model with outcome acute or chronic postoperative pain as dependent variable and independent variables (demographics, psychological variables, preoperative pain, surgery-related factors) determined before the surgery.
• Models do not take independent variables into account that were determined after the start of the surgery.

40 studies were initially selected based on title and abstract screening. After reading the full text, 30 studies were excluded based on study design and PICO (see the table with reasons for exclusion under the tab Methods). Additionally, four studies (Ferland, 2017; Martin, 2020; Voepel-Lewis, 2018; Connely, 2014) were excluded because no overall performance of the multivariate model was reported. Finally, six studies were included in the analysis of the literature. None of these studies reported validated models.

Results

Three studies were included in the analysis of the literature for acute postoperative pain. Four studies were included in the analysis of the literature for chronic postoperative pain. Because none of the studies reported validated models, no evidence and risk of bias tables were created, and no grading was performed.