FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W2793253873> ?p ?o ?g. }

• W2793253873 endingPage "1267" @default.
• W2793253873 startingPage "1264" @default.
• W2793253873 abstract "Chronic pain after surgery, referred to as persistent postsurgical pain (PPSP), is an important condition leading to significant symptom burden and reduced quality of life in affected individuals.1 PPSP is broadly recognized as pain lasting >3 months after surgery. The condition is common, with prevalence estimates ranging from 10% to 50% of all postsurgical patients.1 High-risk procedures include breast surgery, thoracotomy, limb amputation, and inguinal hernia repair.1 PPSP is relatively refractory to treatment and thus has generated interest in potential preventive strategies and treatments.2 A better understanding of which patients develop PPSP would help identify the subset of patients who are likely to require additional treatment to optimize their perioperative and postoperative pain management. However, the complexity and individuality of the human response to pain pose significant challenges in making accurate predictions of PPSP development. As Raja and Jensen3 pointed out, we are currently not better than the weather forecaster in predicting an individual patient’s postoperative pain experience. Although the differences between predicting PPSP and predicting the weather are very profound, we argue that several techniques used in weather prediction could actually serve as an example for attempts to better predict PPSP. We hypothesize that at least 3 techniques used in weather prediction may be relevant in an attempt to improve our ability to predict PPSP: (1) application of recent mathematical, statistical, and computational advances; (2) the construction of a model, taking into account the initial state of the system and the forecast range; and (3) ongoing improvement of algorithms by evaluating the success and failure of predictions. APPLICATION OF RECENT MATHEMATICAL, STATISTICAL, AND COMPUTATIONAL ADVANCES Weather forecasters were once the unfortunate subjects of countless jokes. Predicting the weather from multiple interacting meteorological factors that are greatly influenced by underlying geography seemed no better than extracting a forecast from a cloudy crystal ball. The complexity and individuality of the human response to the nociception of surgery pose similar hurdles to making accurate predictions of PPSP development. But access to huge amounts of data, refinement of mathematical models, and enhanced computing power have transformed predictive meteorology,4 and the same may apply to the predicting of PPSP. The development of PPSP has been associated with a wide range of factors including pain processing, demographic factors, treatment factors, psychosocial factors, and genetic/epigenetic factors.1,3 Conventional logistic regression approaches are unable to incorporate the rapidly expanding set of available data nowadays, let alone the genetic, proteomic, and metabolomic data expected to be available for analysis and prediction of PPSP in the near future. Machine-learning classifiers are algorithms that can autonomously integrate and learn from complex datasets with many hundreds of variables.5 Machine-learning techniques are used on a daily basis to solve prediction problems such as handwriting recognition, fraud detection, and e-mail spam filtering. These algorithms use a variety of mathematical approaches and are often more computationally efficient and accurate when using very large datasets (big data) with complex distributions that do not conform to the assumptions of parametric methods like logistic regression.5 Therefore, machine-learning classifiers may offer a solution to the vexing challenge of predicting PPSP. Tighe et al6 recently demonstrated the feasibility of this approach by using machine learning on readily available information in the electronic medical record to predict which individual patients are at risk of severe acute postoperative pain, which is a known risk factor for PPSP.1 The model used by Tighe et al6 was based on routinely collected clinical data but performed similarly in predicting acute moderate to severe postoperative pain to models from studies specifically gathering data to predict acute moderate to severe postoperative pain using traditional statistical methods.6 Another example of machine learning in the context of pain prediction is the successful application of machine learning to predict the response to postoperative opioid medication using preoperative electroencephalography (EEG) parameters.7,8 Gram et al7 recorded clinical parameters and EEG during rest and tonic pain in 81 patients the day before total hip replacement surgery. Postoperative pain treatment consisted of oral oxycodone and intravenous piritramide via patient-controlled analgesia. Patients were stratified into responders and nonresponders based on pain ratings during 24 hours postsurgery. Potential predictive preoperative parameters were analyzed using machine learning to predict the individual response to opioid analgesia. After surgery, they found a large interindividual variation in pain ratings, even though patients were treated using patient-controlled analgesia. The machine-learning prediction model found that preoperative EEG analysis during tonic pain was able to predict responders to opioid analgesia with an accuracy of 65%, indicating that it may be used as an objective preoperative biomarker of postoperative opioid response. In the field of diabetes research, Zeevi et al9 recently reported on a machine-learning approach to the complex problem of elevated postprandial glucose concentrations. In an 800-person cohort, they found markedly high variability in the glucose response to identical meals between individuals, suggesting that universal dietary recommendations have limited value. Using gut microbiome measurements, blood parameters, and other measurements of patients’ glucose physiology, they constructed a machine-learning algorithm to predict personalized glycemic responses to meals. Subsequently, they used this algorithm to provide individual food recommendations (such as chicken recommended for 1 person, but withheld from another) and demonstrated significantly lower postprandial glucose concentrations and concomitant changes in gut microbiota. Similarly, it is conceivable that machine-learning approaches applied to different, high-resolution, specifically for PPSP prediction gathered parameters such as data from EEG, treatment factors, psychosocial factors, and epigenetic and genetic factors, have the potential to create more accurate, individualized predictions of PPSP. Such models may even be used to tailor specific perioperative treatments based on individual patient characteristics. At present, there is moderate evidence for different “antihyperalgesic” drugs and regional anesthesia as preemptive analgesia in the context of specific procedures.10 However, most trials studying prevention of PPSP have been underpowered with inadequate patient follow-up.10 More data on the effect of preemptive analgesia on PPSP will impact the ultimate relevance of PPSP prediction. Although machine learning can improve the accuracy of prediction over the use of conventional regression models by capturing complex, nonlinear relationships in the data, it cannot squeeze out information that is not present in the data. Most big data efforts thus far have mostly focused on using readily available information from electronic health records or insurance data.5,6 These data are easy and inexpensive to obtain, but using these types of low-resolution data that were collected to serve a different purposes may lead to construct validity issues.5 It would be more worthwhile to apply these innovative analytics to data from traditionally high-resolution sources and combine them with new sources, as demonstrated in several of the aforementioned studies,7–9 although an obvious trade off in collecting more specific predication data is that it would likely require additional resources. Ultimately, the patient burden and resources required for specific high-resolution testing to predict PPSP need to be weighed against the treatment implications of accurate prediction of PPSP. In the context of procedures that are overall associated with a low risk of PPSP and/or no known effective preemptive treatments, it may at present be more feasible to restrict prediction models to readily available data. THE CONSTRUCTION OF A MODEL, TAKING INTO ACCOUNT THE INITIAL STATE OF THE SYSTEM AND THE FORECAST RANGE An improvement in the prediction of weather has been the realization that small perturbations to the initial conditions in unstable systems may result in vastly different forecasts on a long-term basis, limiting predictive performance. This concept, also known as “chaos theory,” and the recognition that forecasts have system state–dependent limits of predictability, has led to the development of models to estimate the uncertainty of forecasts.11 The chaos theory concept is often illustrated using the metaphorical example of the “butterfly effect,” stating that the path and time of a tornado formation can be influenced by minor perturbations such as the flapping of the wings of a distant butterfly several weeks earlier. Several recent systematic reviews have focused on the preoperative prediction of postoperative pain12–15 using psychophysical testing. Psychophysical testing, also called quantitative sensory testing, consists of recording the perceived pain to different standardized stimuli to obtain measures of pain processing. Patients prone to develop PPSP have been shown to exhibit a pronociceptive pain processing, as evidenced by higher perceived pain to the same standardized stimulus (hyperalgesia) and impaired pain inhibitory modulation, when compared to patients who will not develop PPSP.14 Preoperative psychophysical testing is currently indeed one of the most promising methods to predict PPSP12 and may predict 4%–54% of the variance in acute postoperative pain,14 which is an important risk factor for PPSP.1 However, from a clinical standpoint, such prediction is not sufficient to predict an individual patient’s PPSP development and justify invasive or medical preventative strategies.3 Although highly accurate preoperative prediction of long-term PPSP would indeed be the “silver bullet” in terms of initiating potential preventative perioperative treatment strategies, such prediction may not be feasible at present. In the context of PPSP, the significant perturbation that is applied to a patient’s pain processing system by the nociception of surgery may limit the predictability of PPSP using preoperative models. It is known that severe acute postoperative pain and specific acute postoperative pain trajectories are predictive of PPSP.1 Moreover, some studies have shown that treatment of acute postoperative pain with different analgesic regimes may prevent subsequent PPSP in a substantial fraction of patients.16 It may thus be more feasible to focus PPSP prediction models on the acute postoperative period, when the patient’s pain processing system has been affected by the perturbation of surgery and psychophysical measurements may show a shift in pain processing toward a pronociceptive pain response, which is likely associated with an increased risk of PPSP. Along these lines, we have previously reported lower pressure pain tolerance thresholds (Figure 1) and higher pain scores (Figure 2) very early after surgery in women who developed PPSP at 12 months after breast cancer surgery.17Figure 1.: Logistic probability plot of persistent postoperative pain 12 mo postoperatively based on change in pressure pain threshold at postoperative day 5 versus preoperative baseline. Change in pressure pain thresholds was measured in 94 women undergoing breast cancer surgery, and persistent pain was defined as a pain score of >30 mm on visual analog scale at 12 mo postoperatively. Eleven patients (12%) developed persistent postsurgical pain. This figure is a reanalysis from van Helmond et al.17Figure 2.: Logistic probability plot of persistent postoperative pain 12 mo postoperatively based on the clinical pain score reported during arm movement at postoperative day 5. VAS was measured in 94 women undergoing breast cancer surgery, and persistent pain was defined as a pain score of >30 mm on VAS at 12 mo postoperatively. Eleven patients (12%) developed persistent postsurgical pain. This figure is a reanalysis from van Helmond et al.17 VAS indicates visual analog scale.In agreement with the presented theory that the baseline pain system state of patients undergoing surgery may not allow for long-term predictions of PPSP, preoperative pain thresholds were not predictive of PPSP,17 which is consistent with a similar study on PPSP after breast cancer surgery.18 However, in the context of other types of surgery, a pronociceptive pain phenotype preoperatively14 has been reported to be predictive of PPSP. Most of these studies were conducted in the context of orthopedic joint surgery and represent a very different patient population. In the orthopedic population, patients have often suffered from ongoing pain and nociceptive input for a prolonged time preoperatively, which may have led to a sensitized pain system and enhanced pain message transmission even before surgery. Conversely, patients undergoing surgery for breast cancer are highly unlikely to have suffered from significant pain preoperatively and are consequently unlikely to express a change in their pain system preoperatively. Along these lines, it seems likely that a number of parameters have different predictive value for development of PPSP at different perioperative time points and these may again be variable for different surgical procedures. Additional and larger studies are needed to examine what perioperative time points and prediction models are appropriate for specific different surgical procedures. Similarly, a recent initiative has been taken by the Procedure-Specific Postoperative Pain Management group, consisting of experts analyzing existing evidence, to provide different specific recommendations for postoperative pain control for different surgical procedures.19 More evidence on treatments capable of preventing PPSP for different surgical procedures will help allocate adequate perioperative treatments to patients at high risk of developing PPSP, as identified by predication models. ONGOING IMPROVEMENT OF ALGORITHMS BY EVALUATING THE SUCCESS AND FAILURE OF PREDICTIONS A central component to the advancement of weather prediction over the past few decades has been that weather prediction performance is evaluated objectively daily and globally, so that success and failure of forecasts are accurately known and pathways to improve predictive performance can be effectively tested.20 In the context of predicting PPSP, most clinical studies focus on 1 individual factor or on a set of factors, in a single-center prospective or retrospective study design. The focus on individual significant factors, which contribute relatively little to the overall PPSP variation, is problematic from a clinical standpoint, since a small added risk cannot satisfactorily justify additional preventive treatments. Moreover, most individual factors that demonstrate explanatory statistical significance have not been used in actual dedicated subsequent prediction studies and most studies have not been validated in independent patient cohorts. It would be more valuable to take a comprehensive approach, combining as many clinical, demographic, psychosocial, genetic, and neurophysiological factors in a single prediction model using the aforementioned machine-learning approaches and to refine this model in an ongoing iterative manner. Improving a comprehensive model in a prospective iterative manner will likely allow for the identification of new pathways to increase predictive performance. It is conceivable that building such a model will concomitantly lead to more insight into the pathophysiological pathways through which PPSP arises, by generating testable hypotheses based on the found predictors. Because it is impossible to infer causality from such correlations, it will still be necessary to run conventional trials to test any strong prediction hypothesis that comes out of a machine-learning process. Applying innovative analysis approaches to all relevant data available will by no means abolish the need for controlled research on single factors but will likely catalyze its focus in a certain direction. The current lack of understanding of the pathophysiology of PPSP is in stark contrast with weather models, which are at the heart based on well-understood but nonlinear equations involving the conservation of mass, momentum, and energy. At this point in time, most clinicians would not want to emulate weather forecasters and be asked to predict an individual’s PPSP risk based on factors such as their pain processing profile, psychosocial factors, or genetic factors. However, when combined with a machine-learned algorithm focused at different postoperative time points, incorporating all individual factors, prediction may be less daunting. Such models could be used to provide individualized decision support to clinicians based on learning from trajectories in similar patients in the past. In the era of “big data” science, in which we can measure an enormous number of parameters, harnessing the most-predictive aspects of highly dimensional data will likely be very powerful. CONCLUSIONS In conclusion, prediction of PPSP may benefit from different techniques regularly used in weather prediction. First, prediction of PPSP may be enhanced by the application of recent mathematical, statistical, and computational advances. Second, prediction of PPSP should take into account the time scale and the stability of the initial pain processing system on which the model is developed. Third, comprehensive PPSP prediction models should be improved through an ongoing prospective, iterative manner by evaluating success and failure of predictions. Future studies need to test the hypothesis that PPSP prediction can be improved along the methods outlined in this article. DISCLOSURES Name: Noud van Helmond, MD. Contribution: This author helped manage the literature searches, summarize previous related work, and write the manuscript. Name: Søren S. Olesen, MD, PhD. Contribution: This author helped analyze the data, prepare the figures, and edit and revise the manuscript. Name: Oliver H. Wilder-Smith, MD, PhD. Contribution: This author helped edit and revise the manuscript. Name: Asbjørn M. Drewes, MD, PhD. Contribution: This author helped edit and revise the manuscript. Name: Monique A. Steegers, MD, PhD. Contribution: This author helped edit and revise the manuscript. Name: Kris C. Vissers, MD, PhD. Contribution: This author helped edit and revise the manuscript. This manuscript was handled by: Honorio T. Benzon, MD." @default.
• W2793253873 created "2018-03-29" @default.
• W2793253873 creator A5023802517 @default.
• W2793253873 creator A5027673859 @default.
• W2793253873 creator A5043932380 @default.
• W2793253873 creator A5049607244 @default.
• W2793253873 creator A5058714238 @default.
• W2793253873 creator A5062939154 @default.
• W2793253873 date "2018-11-01" @default.
• W2793253873 modified "2023-09-26" @default.
• W2793253873 title "Predicting Persistent Pain After Surgery: Can Predicting the Weather Serve as an Example?" @default.
• W2793253873 cites W1789155650 @default.
• W2793253873 cites W1978519915 @default.
• W2793253873 cites W2008952253 @default.
• W2793253873 cites W2021767175 @default.
• W2793253873 cites W2066042884 @default.
• W2793253873 cites W2110904274 @default.
• W2793253873 cites W2124884366 @default.
• W2793253873 cites W2129003580 @default.
• W2793253873 cites W2129568786 @default.
• W2793253873 cites W2130141996 @default.
• W2793253873 cites W2175465469 @default.
• W2793253873 cites W2188799301 @default.
• W2793253873 cites W2477891822 @default.
• W2793253873 cites W2559959729 @default.
• W2793253873 cites W2566478165 @default.
• W2793253873 cites W2726943764 @default.
• W2793253873 cites W2727650337 @default.
• W2793253873 doi "https://doi.org/10.1213/ane.0000000000003318" @default.
• W2793253873 hasPubMedId "https://pubmed.ncbi.nlm.nih.gov/29505448" @default.
• W2793253873 hasPublicationYear "2018" @default.
• W2793253873 type Work @default.
• W2793253873 sameAs 2793253873 @default.
• W2793253873 citedByCount "8" @default.
• W2793253873 countsByYear W27932538732019 @default.
• W2793253873 countsByYear W27932538732020 @default.
• W2793253873 countsByYear W27932538732021 @default.
• W2793253873 countsByYear W27932538732022 @default.
• W2793253873 crossrefType "journal-article" @default.
• W2793253873 hasAuthorship W2793253873A5023802517 @default.
• W2793253873 hasAuthorship W2793253873A5027673859 @default.
• W2793253873 hasAuthorship W2793253873A5043932380 @default.
• W2793253873 hasAuthorship W2793253873A5049607244 @default.
• W2793253873 hasAuthorship W2793253873A5058714238 @default.
• W2793253873 hasAuthorship W2793253873A5062939154 @default.
• W2793253873 hasBestOaLocation W27932538731 @default.
• W2793253873 hasConcept C42219234 @default.
• W2793253873 hasConcept C71924100 @default.
• W2793253873 hasConceptScore W2793253873C42219234 @default.
• W2793253873 hasConceptScore W2793253873C71924100 @default.
• W2793253873 hasIssue "5" @default.
• W2793253873 hasLocation W27932538731 @default.
• W2793253873 hasLocation W27932538732 @default.
• W2793253873 hasLocation W27932538733 @default.
• W2793253873 hasOpenAccess W2793253873 @default.
• W2793253873 hasPrimaryLocation W27932538731 @default.
• W2793253873 hasRelatedWork W1506200166 @default.
• W2793253873 hasRelatedWork W1995515455 @default.
• W2793253873 hasRelatedWork W2048182022 @default.
• W2793253873 hasRelatedWork W2080531066 @default.
• W2793253873 hasRelatedWork W2604872355 @default.
• W2793253873 hasRelatedWork W2748952813 @default.
• W2793253873 hasRelatedWork W2899084033 @default.
• W2793253873 hasRelatedWork W3031052312 @default.
• W2793253873 hasRelatedWork W3032375762 @default.
• W2793253873 hasRelatedWork W3108674512 @default.
• W2793253873 hasVolume "127" @default.
• W2793253873 isParatext "false" @default.
• W2793253873 isRetracted "false" @default.
• W2793253873 magId "2793253873" @default.
• W2793253873 workType "article" @default.