Although the understanding of pain has advanced, managing postoperative pain remains a significant problem worldwide. Inadequate pain relief is common in postoperative patients, and only about one-fourth of surgical patients have their pain adequately controlled.1 In postoperative surgical patients, 41% of patients have moderate to severe pain on the day of their surgery, and 15% of patients still report moderate to severe pain on the fourth postoperative day.2 Improving pain management for postoperative patients is therefore a great challenge.
Among all the management approaches, intravenous patient-controlled analgesia (IV-PCA) is an effective and well-established approach to relieve acute postoperative pain.3 The simple route of administration and its safety makes IV-PCA a popular method of pain relief for most surgeries.4,5 However, this approach is used for a wide ranges of surgical procedures and results in significant differences of analgesic consumption when managing postoperative acute pain. Both the type and technique of surgical procedures induce different amounts of tissue damage and cause different intensities of pain.6,7 There are huge differences in postoperative analgesic requirement existing among surgical procedures. Treating the wide range of intensity and quality of pain created by different surgical procedures is vital in the PCA regimens. A comprehensive investigation of the pain intensity on the first postoperative day had shown that various pain intensities were caused by different surgical procedures.8 This result suggested the need for procedure-specific pain.9 However, most of the studies analyze the analgesic consumptions in the average level of the population; the analysis of the individual procedure level is still limited.10,11 To improve the postoperative pain management and develop procedure-specialized IV-PCA regimens by providing the information of individual procedure level of analgesic consumptions, we proposed a statistical approach to deal with the heterogeneity that comes from types of surgery.
The primary aim of this study was to quantify the procedure-specific total morphine consumption of IV-PCA. We also analyzed the effects of various surgical procedures on total morphine consumption and derived a corresponding formulation. Finally, we gave a rank of morphine consumption for surgeries to show the relative differences between surgeries and discussed the possible explanations.
MATERIALS AND METHODS
The data for model development were obtained from our previous retrospective observational study, which was investigating the serial analgesic consumptions of the IV-PCA.4 This study investigated postoperative patients using IV-PCA between January 2005 and December 2010. All patients in the study received the similar principal of IV-PCA management. IV-PCA was prepared with a standard solution of 1 mg/ml morphine in normal saline. A 0.05 mg/kg loading dose was first given for immediate acute postoperative pain management. IV-PCA was initially programmed to administer a bolus dose of 0.5 to 1.5 mg morphine with a lockout interval between 5 to 10 minutes when patients activated the demand button. For safety, we initially set a 4-hour limit to 6 to 12 mg according to the clinical conditions. Note that these PCA pump settings were not fixed, but could change on the basis of patients’ demand, evaluated by our acute pain service team during the therapeutic course.
The external validation was carried out using data from an independent retrospective observational study. This study investigated the correlation between serial pain scores and IV-PCA during the postoperative period. It included surgical patients with IV-PCA during the period spanning from January 2013 to December 2014. After approval by the Institutional Review Board of the Taipei Veterans General Hospital (IRB: 2016-04-003ACF), the data of total analgesic consumptions and bolus counts were retrieved from the log profile of infusion pumps (Aim plus system; Abbott Laboratories, North Chicago, IL). Patient demographic data were collected from the medical record and IV-PCA worksheet. Surgical procedures were coded according to the ICD-10 Procedure Coding System (ICD-10-PCS; Supplemental Digital Content 1, http://links.lww.com/CJP/A520).
All continuous data were reported as median and interquartile range, and categorical data were reported as proportion. Under the theory of the generalized linear mixed-effect model,12 we proposed 3 models for model construction. Fixed effects of the procedure parametrized the types of surgery by dummy variables, while random effect of the procedure was parametrized as a normal distribution. Total morphine consumption was checked, normality by quantile-quantile plot and histogram. Model was also assumed as a normal distribution or a student t distribution for dependent variables. Model fit used deviance information criterion (DIC) for measurement. A model with smaller DIC is preferred. Parameters estimation was carried out by Bayesian approach. Further statistical modeling and estimation were listed in the Supplemental Digital Content 1 (http://links.lww.com/CJP/A520). The predictive ability of the model was tested in an external validation data set and was qualified in terms of predictive squared correlation coefficient Q2.13 A P<0.05 was considered statistically significant. All statistical calculations and estimations were performed with SAS software (version 9.4; SAS Institute Inc., Cary, NC) and WinBUGS software (MRC Biostatistics Unit, Cambridge, UK).14
A total of 3284 patients who underwent 66 different surgical procedures were enrolled in the model development. The most common surgeries pertaining to IV-PCA were colon-rectal or general surgeries (36% and 29%, respectively). The median age of the population was 64 years, which was older than those monitored in Hong Kong.15 Moreover, our study patients were being treated for cancer and represented a group of patients at increased risk for surgery. Another data set of 500 surgical patients was used for external model validation. The demographic data are presented in Table 1.
The mean total morphine consumption was 64.5 mg, and the median was 60.2 mg in the model development. Both student t distribution and normal distribution were considered during the construction of the model (Fig. 1). However, high extreme values in the tail of the distribution were noted, suggesting that the distribution was skewed. The student t distribution was considered a better fit for the model according to the DIC value (the smaller, the better). Compared with the fixed-effect model, the random-effect model also improved the model fit. All of the model selection and parameter estimates are presented in Tables 2 and 3.
The final model included procedure effects as a random effect, which implies that the effects significantly varied from procedure to procedure. There were a total of 66 estimated procedure-specific morphine consumptions, with a wide range of variations (76.92 to 34.3 mg). The resection of the pancreas using the open approach resulted in the highest consumption of morphine to manage acute postoperative pain, while patients who underwent an excision of the uterus using the open approach consumed the lowest dose of morphine. Most abdominal area and open approach surgeries resulted in significantly greater morphine consumption. The quantitative estimations of 66 morphine consumption for surgeries are presented in Table 3. We also present the forest plot to show the wide variation of morphine consumption between surgeries (Fig. 2). The ranks of morphine consumption for surgeries are also listed for comparison, and 3 levels of surgical intensity was divided by the rank of the effect (Table 3). Some of the orthopedic surgeries, such as replacement of the knee joint (rank: 11/66) and fusion of the lumbar spine (rank: 16/66) were also painful procedures resulting in significantly greater morphine consumption.
In most cases, an endoscopic approach resulted in lower morphine consumption than the equivalent open approach. The relative ranks of procedures performed using both open and endoscopic approaches are as follows (open vs. endoscopic): resection of rectum (10 vs. 25); resection of gallbladder (15 vs. 26); resection of kidney and ureter (23 vs. 28); inspection of pleural cavity (8 vs. 34); resection of stomach (2 vs. 37); resection of ascending colon (18 vs. 48); and resection of spleen (45 vs. 52). Only resection of kidney (40 vs. 39) and resection of uterus (54 vs. 53) resulted in greater morphine consumption when the endoscopic approach was used, but these differences were very small.
In addition to the procedures’ effect in the model, age, sex, weight, and presence of cancer were also significant predictors for total morphine consumption. Younger age, male sex, presence of cancer, and heavier body weights were correlated with higher morphine consumption.
The IV-PCA regimens formulation for surgical procedure was also derived from the final model. The quantitative estimation of total morphine consumptions for specific surgery can be calculated by the following formulation:
Total morphine consumption (mg)=−0.90×age+8.98×sex (male=1, female=0)+3.06×cancer status (yes=1, no=0)+0.77×weight (kg)+procedure-specific factors (estimation values in Table 3).
For the ease of explanations, we present 2 clinical scenarios to estimate the total morphine consumption. The first is a 50-year-old male patient, who is 80 kg in weight, with rectum cancer. He will receive rectum resection by open method. Another is a 70-year-old female patient, who is 60 kg in weight, diagnosed as rectum cancer. She will receive surgery by laparoscopic approach. The estimated total morphine consumption for the 2 patients are:
- Patient 1: Total morphine consumption=−0.90×50+8.98×1+3.06×1+0.77×80+71.08=99.72 (mg);
- Patient 2: Total morphine consumption=−0.90×70+8.98×0+3.06×1+0.77×60+63.79=50.05 (mg).
The difference between the 2 estimated total morphine consumptions was 49.67 mg, which shows a substantial difference in morphine consumption postoperatively.
The performance of the model prediction in external validation showed moderated correlation (Q2=0.308). The relationship of the model prediction and data observation is shown in Figure 3.
It is of interest to investigate and quantify the effect of the specific surgical approach on the morphine consumption after surgery for acute pain management. Most of the studies have confirmed that the surgical procedure significantly correlated with morphine consumption for acute pain management. However, very few studies have attempted to quantify the morphine consumption for specific surgical procedures. The proposed model in our study was tailored for providing such a quantitative assessment for numerous surgical procedures; it can develop a useful formulation for clinical consultation for patients after surgery.
Acute postoperative pain is the result of nociception after surgery, and, traditionally, it has good response to opioid drugs. However, postoperative pain management is challenging and varies greatly between procedures and individuals. Different types of surgeries manipulate and damage the tissues and organs in different ways (through resection, fixation, traction, compression, etc.) and result in different levels of pain intensity. Studies support the idea that certain types of surgeries (abdominal, orthopedic, and thoracic) are significant predictors of postoperative pain.7,16 The type of surgery was also a significant predictor for analgesic consumption. On the basis of this evidence, procedure-specific pain management is suggested.17
IV-PCA has been used for decades as an effective and well-accepted management technique for acute postoperative pain. As it relies on the patients to administer their morphine themselves, total morphine consumption during IV-PCA represents the patient’s subjective requirements to ease acute pain, which has been used as a research tool in many studies.10,16,18,19 Most of them focused on the correlation between predictors and analgesic consumptions. In this study, we provided a quantitative estimation for the degree of variation in morphine consumptions among the various surgeries. In addition, the median and range of variations for those surgeries are now known. This information is useful for both clinical practice and research. It helps not only to develop procedure-specific IV-PCA regimens but also to compare the analgesic usage for different surgeries.
One difficulty in this study was surgical classification. Hundreds or thousands of existing surgical procedures are difficult to describe and classify into a few simple categories. A more comprehensive and organized coding system is thus needed. The ICD-10-PCS is an international system used for procedure coding. It was first released in 1998 and has since been updated annually. One advantage of this system is its classification of surgeries using 7 codes to describe the body system, main operation type, organs, and devices used, which allows for easier understanding and analysis.20 However, we did not use this hierarchical coding structure for further analysis, but simply used it to classify the surgeries for better presentation and communication. There are still some limitations of the ICD-10-PCS. First, some surgeries involving multiple body areas or surgery types can only be coded by the main surgery. Second, the ICD-10-PCS is mainly used for inpatient surgeries; outpatient surgeries are not included.
We analyzed the effect of the surgical procedure to capture interprocedure variation. The traditional analysis using a linear regression model treated the procedures as a fixed effect, involving coding the 66 procedures as the same number of categorical variables. However, the large number of surgical procedure parameters violates the parsimony principal of biostatistics, and models with large numbers of parameters are penalized in the model selection process. A benefit of using the random-effect model was that fewer parameters could be used to model a large number of surgeries, as evidenced by the improvement in the DIC value for both of the random-effect models, relative to the fixed-effect model. However, model fit was further improved after changing from a normal distribution to a student t distribution. This is quite common in real data, where data have more extreme values and deviate from the normal distribution. The t distribution was more robust and flexible than the normal distribution, and allowed coverage of the extreme data in the heavy tail of the distribution.
The surgical area seems to significantly affect the morphine consumption.21,22 Our results matched other investigations: surgeries involving an organ in the abdomen and pleural cavity used more morphine postoperatively.23–25 Abdominal surgeries, especially those involving the pancreas, stomach, or liver or colon resection, tended to have more total morphine consumption than surgeries involving other organs. Surgical procedures of the abdomen, using an open approach, were shown to result in higher morphine consumption than other surgeries. Surgery for cancer is one possible reason for this. Surgeries to remove cancers often resect more radically and extensively than other surgeries, causing a higher level of ischemic pain in the abdomen. Morphine consumption may also depend on the nature of the cancer involved. We did, however, include a cancer covariate in the model to compensate for this confounding factor.
The type of surgery also has influence on morphine consumption. Comparing the same surgery with different approaches (open and endoscopic method) resulted in different morphine consumptions. The endoscopic approach decreased morphine consumption, and therefore lowered the rank of morphine consumption in many surgeries. Many surgeries, such as resection of the stomach, resection of the ascending colon, and inspection of the pleural cavity were benefitted with the endoscopic approach. Although these abdominal surgeries with open approach were mostly in the top ranks of the morphine consumption, the surgeries with endoscopic approach were in the middle of the procedures included in this study. To facilitate the development of a procedure-specific IV-PCA regimen, we included a procedure-specific effect in the regression model. Thus, we assumed a random intercept for procedure level, which could model the variation in morphine consumption between procedures. Other predictors, such as age, sex, and weight, were treated as fixed effects (population-average effects). Both fixed and random effects were included in the model, resulting in a mixed-effect model.
When we sorted the estimated effect of surgery on total morphine consumption, we arrived at the rank statistics. It provides more robust comparisons among all surgeries and can be treated as an ordinal variable by different effect levels. Our previous study had concluded with 3 classifications of serial analgesic consumptions (high, middle, and low) by cluster analysis.4 The 3 levels of surgical effects can be considered in the future to present 3 intensity scales on analgesic consumptions for acute pain and is easier for explanation.
Although our study provided more information about morphine consumption for individual surgical procedures, there are several concerns that should be addressed. First, this was a retrospective observational study. There are possible unmeasured variables that can confound the morphine consumption, such as psychological predictors. However, this may have more influence at the individual level and not at the surgical level. Other demographic predictors, such as weight and sex, have been corrected in the regression model. Other unknown factors, such as preoperative medications or postoperative adjuncts not measured in the study, may have an effect in the study. Second, our population was characterized as older and with more cancer history, which may combine with more comorbidity and higher anesthetic risk. The concern over such a kind of population is that study samples may not be representative of the entire population of surgical patients but only a subset of the population. Although our study sample size was over 3000 patients, few types of surgery have a small sample size (<5). This will trouble the estimation of effect, because of a large confidence interval in flexed effect (Supplemental Digital Content 1, http://links.lww.com/CJP/A520). However, the random-effect model has the advantage of the parameter estimation by assuming a variation among surgeries. Third, we did not evaluate the preoperative medication factors, such as opioid and gabapentin administrations, or postoperative adjunct factors in the study. Although adjunctive medication may affect how much PCA is used by a patient, we think the detailed classification of surgical procedures in this study would more or less reduce the impact of this issue, because patients receiving the same type of surgery tended to have similar adjunctive medication. For example, patients receiving total knee arthroplasty due to osteoarthritis used to take non-steroidal anti-inflammatory drug before surgery but preoperative analgesic prescriptions are not common for those who have early-stage breast cancer. The detailed classification of surgical procedures would greatly diminish the heterogeneity resulting from the uncollected variables. Besides, the random-effect analytical approach would also decrease the unexplained variance caused by unobserved confounders from the statistical perspective of statistics. Therefore, we believe the lack of adjunctive medication before and during PCA administration would not be a serious issue in our study.
In this study, we have demonstrated the heterogeneity of surgical procedures with respect to IV-PCA morphine consumption. A random-effect approach was used to model interprocedure variation as a procedure-specific effect. We also ranked the surgical procedures according to their estimated effect on morphine consumption and discussed the possible reasons for variations among procedures. A procedure-specific IV-PCA formula is therefore strongly recommended for postoperative acute pain management.
1. Phillips DM. JCAHO pain
management standards are unveiled. Joint Commission on Accreditation of Healthcare Organizations. Jama. 2000;284:428–429.
2. Sommer M, deRijke JM, van Kleef M, et al. The prevalence of postoperative pain
in a sample of 1490 surgical inpatients. Eur J Anaesthesiol. 2008;25:267–274.
3. McNicol ED, Ferguson MC, Hudcova J. Patient controlled opioid analgesia
versus non-patient controlled opioid analgesia
for postoperative pain
. Cochrane Database Syst Rev. 2015;6.
4. Lin SP, Chang KY, Tsou MY, et al. Serial analgesic consumptions and predictors of intravenous patient-controlled analgesia
with cluster analysis. Clin J Pain
5. Yen CR, TSou MY, Mandell MS, et al. An analysis of patient variables that influence intravenous patient-controlled
analgesic use of morphine with quantile regression. Anesthesiology. 2010;112:688–695.
6. Kehlet H, Dahl JB. Anaesthesia, surgery, and challenges in postoperative
recovery. Lancet. 2003;362:1921–1928.
7. Sommer M, deRijke JM, van Kleef M, et al. Predictors of acute postoperative pain
after elective surgery. Clin J Pain
8. Gerbershagen HJ, Aduckathil S, van Wijck AJ, et al. Pain
intensity on the first day after surgery: a prospective cohort study comparing 179 surgical procedures
. Anesthesiology. 2013;118:934–944.
9. Gray A, Kehlet H, Bonnet F, et al. Predicting postoperative analgesia
outcomes: NNT league tables or procedure-specific evidence? Br J Anaesth. 2005;94:710–714.
10. De Cosmo G, Congedo E, Lai C, et al. Preoperative psychologic and demographic predictors of pain
perception and tramadol consumption using intravenous patient-controlled analgesia
. Clin J Pain
11. Gagliese L, Gauthier LR, Macpherson AK, et al. Correlates of postoperative pain
and intravenous patient-controlled analgesia
use in younger and older surgical patients. Pain
12. McCullagh P, Nelder JA. Generalized Linear Models Vol 37. Boca Raton, Florida: CRC Press; 1989.
13. Consonni V, Ballabio D, Todeschini R. Evaluation of model predictive ability by external validation techniques. J Chemometrics. 2010;24:194–201.
14. Lunn DJ, Thomas A, Best N, et al. WinBUGS-a Bayesian modelling framework: concepts, structure, and extensibility. Stat Comput. 2010;10:325–337.
15. Cheung CW, Ying CL, Lee LH, et al. An audit of postoperative
intravenous patient-controlled analgesia
with morphine: evolution over the last decade. Eur J Pain
16. Ip HY, Abrishami A, Peng PW, et al. Predictors of postoperative pain
and analgesic consumption: a qualitative systematic review. Anesthesiology. 2009;111:657–677.
17. Kehlet H, Wilkinson RC, Fischer HB, et al. PROSPECT: evidence-based, procedure-specific postoperative pain
management. Best Pract Res Clin Anaesthesiol. 2007;21:149–159.
18. Cohen L, Fouladi RT, Katz J. Preoperative coping strategies and distress predict postoperative pain
and morphine consumption in women undergoing abdominal gynecologic surgery. J Psychosom Res. 2005;58:201–209.
19. Voulgari A, Lykouras L, Papanikolaou M, et al. Influence of psychological and clinical factors on postoperative pain
and narcotic consumption. Psychother Psychosom. 1991;55:191–196.
20. Averill RF, Mullin RL, Steinbeck BA, et al. Development of the ICD-10 procedure coding system (ICD-10-PCS). Top Health Inf Manage. 2001;21:54–88.
21. Dahmani S, Dupont H, Mantz J, et al. Predictive factors of early morphine requirements in the post-anaesthesia care unit (PACU). Br J Anaesth. 2001;87:385–389.
22. Chang KY, Tsou MY, Chan KH, et al. Factors affecting patient-controlled analgesia
requirements. J Formos Med Assoc. 2006;105:918–925.
23. Kalkman CJ, Visser K, Moen J, et al. Preoperative prediction of severe postoperative pain
24. Chia YY, Chow LH, Hung CC, et al. Gender and pain
upon movement are associated with the requirements for postoperative patient-controlled
: a prospective survey of 2,298 Chinese patients. Can J Anaesth. 2002;49:249–255.
25. Chung F, Ritchie E, Su J. Postoperative pain
in ambulatory surgery. Anesth Analg. 1997;85:808–816.