Comparison between Analgesia Nociception Index (ANI) and self-reported measures for diagnosing pain in conscious individuals: a systematic review and meta-analysis

The Analgesia Nociception Index (ANI), an objective measure of pain based on heart rate variability (HRV), has its usefulness in awake patients still unclear. This systematic review and meta-analysis aimed to assess ANI's accuracy compared to self-reported pain measures in conscious individuals undergoing medical procedures or painful stimuli. PubMed, Ovid, Web of Science, Scopus, Embase, and grey literature were searched until March 2021. Of the 832 identified citations, 16 studies complied with the eligibility criteria. A meta-analysis including nine studies demonstrated a weak negative correlation between ANI and NRS for pain assessment in individuals in the post-anesthetic recovery room (r = − 0.0984, 95% CI = − 0.397 to 0.220, I2 = 95.82%), or in those submitted to electrical stimulus (r = − 0.089; 95% CI = − 0.390 to 0.228, I2 = 0%). The evidence to use ANI in conscious individuals is weak compared to self-report measures of pain, yet ANI explains a part of self-report. Therefore, some individuals may be benefited from the use of ANI during procedures or in the immediate postoperative period.

Information sources. Computerised searches across five electronic databases were conducted in October 2018. The databases used were PubMed (National Library of Medicine), Scopus (Elsevier), Web of Science (Clarivate Analytics), Ovid (Wolters Kluwer), and Embase (Elsevier). An update took place in March 2021. In addition, the reference lists of the included articles were also screened for references that might not have been retrieved during the computerised searches. Finally, searches for literature in Open Grey and Google Scholar were undertaken; the searches were limited to the first 300 most relevant hits 26 . Duplicate references were removed upon identification. The references were managed using EndNote software (Thomson Reuters, Toronto, Canada; https:// www. myend notew eb. com).
Search strategy. The search strategy in PubMed, Ovid, Embase, and Web of Science was analgesia nociception index OR analgesia-nociception index. A specific search strategy was tailored for Scopus: "analgesia nociception index" OR "analgesia-nociception index", and for Embase: 'analgesia nociception index'/exp OR 'analgesia nociception index' . Searches in Google Scholar and Open Grey were carried out with the "analgesia nociception index OR analgesia-nociception index" algorithm. Study selection. Study selection was conducted in two phases. In Phase 1, two review authors (DAB and LRC) read the titles/abstracts independently. The references whose titles/abstracts met the eligibility criteria were included straight away. In Phase 2, the same authors evaluated the full references with titles/abstracts containing insufficient information for a final decision. The references whose full texts met the eligibility criteria were also included. In both phases, divergences between authors were resolved by discussion until a consensus was reached.
Data extraction and data items. Two review authors (DAB, LRC) performed data extraction independently. Disagreements were resolved through discussion. If disagreements persisted, a third review author (LGA) decided. When additional or missing information was needed, the authors of the articles were contacted. The primary data were extracted and are cited in Table 1.

Risk of bias of individual studies.
Two review authors (DAB, LRC) independently carried out the quality assessment using the University of Adelaide critical appraisal checklists for diagnostic test accuracy studies 27 and analytical cross-sectional studies 28 . Any disagreement between the review authors over the risk of bias in individual studies was resolved by a third review author (LGA). For both tools, each item could be answered with yes (low risk of bias), no (high risk of bias), unclear (unclear risk of bias), or not applicable. Diagnostic accuracy measures. Accuracy and correlation measures were the outcomes. Sensitivity, specificity, positive predictive value, negative predictive value, area under the curve (AUC), and receiver operating characteristics (ROC) were used to measure diagnostic accuracy. Correlation between the ANI index and the self-reported pain diagnosis measures included r-values, p values, and confidence intervals (CI).

Synthesis of results and subgroup analysis.
Articles included that were methodologically homogeneous were incorporated into meta-analysis. Subgroup analyses were conducted considering the subjective pain measure used, medical procedures or electrical stimulus applied. Correlation analyses between ANI and subjective measures were conducted.
Statistical heterogeneity of the analyses was assessed using the I 2 statistics. In the meta-analysis with an I 2 higher than 40%, the random-effect model was used. In the meta-analysis with an I 2 lower than 40%, the fixedeffect model was used 29 (Table 1). Three articles indicated the calculation of the sample size [36][37][38] . One was a pilot study conducted with French women during labor 22 , and another was carried out with individuals being treated for burn wounds 34 . Three articles were on conscious and healthy volunteers 32,33,35 , ten about aware patients after procedures under general anaesthesia 16,30,31,[36][37][38][39][40][41][42] , and one in patients on sedo-analgesia (no premedication was administered before the procedure) 15 .
The self-reported subjective measures used to compare with ANI were the Visual Analogue Scale (VAS) 22,35,38,42 and the Numerical Rating Scale (NRS) 15,16,[30][31][32][33][34]36,37,[39][40][41] . Three studies 34,37,38 used objective measures other than ANI: cardiovascular depth of analgesia (CARDEAN) 34 , Surgical Plethysmographic Index (SPI) 37 , the pupillary light reflex (PLR) 37 , and the variation coefficient of pupillary diameter (VCPD) 38 .  Tables 2 and 3 (Supplementary Appendices B and C). The domain judged as having the highest risk of bias in accuracy studies was blinding the index test results concerning the reference standard because these results had not been cited or the test had not been performed. Three correlation studies presented a high risk of bias in identifying confounding factors and strategies to deal with them 15,38,41 , while one study 34  www.nature.com/scientificreports/ using volatile agents and opioid-based anaesthesia in another study 39 . In two other studies, the measure used in a similar scenario was VAS. Jeanne et al. 42 evidenced no correlation between ANI and VAS scores (Spearman rank test, r 2 = − 0.164, P=0.25) in total knee replacement orthopaedic surgery. Charier et al. 38 also found a similar weak negative correlation (Pearson correlation, r = − 0.15; P=0.006) in surgeries whose general anaesthesia and postoperative analgesia protocols had been left to the anaesthetist's discretion. ANI was strongly correlated with VAS in postpartum women (p < 0.0001), in particular before epidural analgesia 22 and presented a weak negative correlation (r = − 0.15). Two studies reported that ANI did not reflect different states of moderate to severe pain measured after sevoflurane-based general anaesthesia in adults 16,37 , revealing low sensitivity and specificity to detect the difference between NRS 0 and NRS 6-10 (AUC = 0.43) in one study 16 . In patients who had undergone colonoscopy under sedo-analgesia, ANI correlated significantly with NRS (r = − 0.402) 15 .

Risk of bias in individual studies. The methodological quality evaluation is summarised in
One study did not obtain satisfactory results when correlating ANI with NRS in three different groups for pain intensity (group I: NRS ≤ 3, group II: NRS 4-6, group III: NRS ≥ 7) in patients who had undergone laparoscopic cholecystectomies under sevoflurane/remifentanil anaesthesia 40 , and no correlation was observed between the postoperative NRS score and the postoperative ANIm values in elective supratentorial tumour surgery 36 .
Papaioannou et al. 34 , demonstrated considerable sensitivity (67%) and specificity (70%) of ANI in predicting pain. Furthermore, the accuracy increased when associated with another measure (CARDEAN). Thus, ANI was fit to measure nociception in a group of conscious burnt patients under analgesic effects during wound care procedures.
One study showed no correlation between ANI minima and NRS in individuals submitted to unexpected electrical pain or expected electrical pain, indicating that ANI was neither a specific nor a robust measure to assess pain intensity compared with NRS in conscious men. There was no correlation between minima ANI and NRS when assessing painful stimuli (rs = − 0.01, P = 0.97) 33 . Issa et al. 32 , showed a weak correlation between ANI and NRS (Pearson, − 0.089; 95% CI − 0.192 to − 0.014; P = 0.045), suggesting that ANI was not specific for the assessment of pain intensity in alert volunteers. Yan et al. 36 , evaluated conscious, healthy volunteers with a cold pressor simulator, showing that the correlation between ANI and VAS was negative and weak (r = − 0.27 and P = 0.017).

Synthesis of results and subgroup analysis.
Nine studies were incorporated into a meta-analysis. Two subgroup analyses of correlation between ANI and NRS were feasible: (1) data of studies assessing conscious individuals who had undergone medical procedures under general anaesthesia were pooled; (2) data of studies evaluating participants submitted to electrical stimulus were pooled.

Discussion
This systematic review and meta-analysis demonstrated that nociception assessed through ANI had a weak and negative correlation with subjective self-reported measures of pain in conscious individuals, i.e., those undergoing medical procedures or submitted to painful experimental stimuli. However, good accuracy of the ANI as compared with NRS was observed in some studies 30,31,41 .  2020) is: "An unpleasant sensory and emotional experience associated with actual or potential tissue damage, or described in terms of such damage" 43 . For nociception, the concept is "The neural process of encoding noxious stimuli. Note: Consequences of encoding may be autonomic (e.g., elevated blood pressure) or behavioural (motor withdrawal reflex or more complex nocifensive behaviour). Pain sensation is not necessarily implied" 44 . These definitions strongly underline the influence of stress and emotions in modifying the correlation of nociception and pain assessment in awake individuals after surgical procedures or painful stimuli. The nature of pain is multifactorial 45 . Nociception depends on the trigger, and pain is clearly defined as a subjective experience 46 .
The satisfactory ANI's accuracy reported by Boselli et al. 30 and Boselli et al. 31 in the post-operative period should be interpreted with attention. Such findings may be clinically relevant because their results suggest that ANI can support practitioners to assess pain in the surgical setting and consequently allow a more reliable prescription of medications during and after surgical procedures. Thus, the use of ANI in the PACU may be potentially appropriate since inadequate management of pain in the postoperative period leads to undesirable results during the patient's recovery 47 .
On the other hand, the large variability in the results of correlation between ANI (objective measure) and NRS (subjective measure) in participants in the postoperative period of general anaesthesia proved by meta-analysis provides power to self-report measures of pain as the "gold standard" for deciding on analgesic complementation in conscious patients.
Our findings underline the influence of anaesthetic agents on ANI scores. However, there is no consensus on which anaesthetic agent would improve the correlation of ANI with subjective pain measures, possibly because the studies have compared different types of anaesthetics beside their relevant heterogeneity 16,30,31,36,37,39,41 . The anaesthetic agent and the drug consumption for pain control during the surgical procedure (remifentanil, fentanyl, sevoflurane, propofol, halogenated) [36][37][38][39][40][41][42][47][48][49] or the technique (spinal, regional, or general anaesthesia) 22,50 may influence the ANS regulation and alter the response of HRV to nociception.
Another vital point is whether the patients were conscious when answering about their pain. Factors such as the patient's level of awareness and perception of the situation may also impact the final result of pain assessment 39 . Different surgical procedures 36,37,39,42 and drugs' residual effect 36 should also be taken into account in assessing pain. It is worth mentioning the negative correlation found in one included study, in which the patients exhibited spontaneous breathing during labour 22 .
As evidenced by studies included in the second subgroup analysis, the subjectivity of pain may impair the ANI assessment in individuals exposed to electrical stimuli 32,33 . The ANS is an essential regulator of heart rate. The transition in the time between two heartbeats is designated as HRV. It provides reliable information about the interaction of the sympathetic and parasympathetic nervous systems 51 . Parasympathetic activity is dominant in resting conditions, such as relaxation and sleep, whereas sympathetic activity increases heart rate and blood pressure in situations requiring energy expenditure. Their interaction is known as the sympathetic-vagal balance of the ANS. ANS balance and its conditions are reflected in HRV, which refers to short-and long-term heart rate variations due to several states, including emotional issues [52][53][54][55][56] . The sympathovagal balance is influenced by arousal, emotions, medications, and drugs used intraoperatively. Some of these factors, such as arousal and feelings, are more evident in conscious patients.
A weak correlation has made us reflect on the statistical analysis of the included studies whose authors had evaluated the agreement between two tests. Correlation analysis, the measure used by the majority, may be a powerless statistic. Therefore, in future research, accuracy measures or the Bland-Altman plot test should be used 57 .  www.nature.com/scientificreports/ According to a narrative review, measuring nociception in clinical settings is practically unfeasible. Still, it would be desirable for patients under general anaesthesia or unable to communicate to prevent acute postoperative pain. The authors conclude that no device has its usefulness justified in practice 5 . However, a recent systematised review described the validity of ANI for nociception assessment in anesthetised patients undergoing surgery and reported considerable changes in ANI values were found in response to nociceptive stimuli at different opioid concentrations and higher ANI values were noted during nociceptive stimuli 58 .
The studies included in our review used two Patient-Reported Outcome Measures (PROMS): VAS 22,35,38,42 and NRS 16,[30][31][32][33][34]36,37,[39][40][41] . According to the FDA (Food and Drug Administration), a "PRO (Patient-Reported Outcome) is any report of the status of a patient's health condition that comes directly from the patient, without interpretation of the patient's response by a clinician or anyone else". Therefore, the accuracy of ANI was assessed according to PROs 59 .
The present study has limitations. Although individuals in the included studies were conscious and reported their pain, the pain stimuli assessed were quite different. Women in labour, patients treated for burns, various elective surgeries, and patients who had received electrical stimuli may exhibit different responses to pain, taking into account the subjective pattern of pain and the influence of nociceptive stimuli. This study provided two subgroups of meta-analysis, but it is necessary to consider that, individually, some included studies 30,31,41 demonstrated adequate accuracy and correlation of ANI with subjective measures of pain.
The strengths of this systematic review and meta-analysis include a comprehensive literature search through major electronic databases and grey literature, adherence to the PRISMA guidelines, and the inclusion of the highest number of studies published on this topic. Finally, data extraction, evaluation of outcomes, and the risk of bias assessment were all performed in duplicate. One limitation is the methodological heterogeneity among the included studies with different designs, precluding additional aggregated analyses. Among the included studies, only nine showed homogeneity. Due to differences regarding the setting where the studies had been conducted, a unique meta-analysis was unfeasible. Analyses were conducted in two subgroups; in one subgroup, data of only two studies were aggregated. According to the literature, quantitative analyses (even those with a few studies) represent a powerful tool to summarise data and increase sample size, allowing the researchers to obtain more reliable estimates. Nevertheless, the findings of those quantitative analyses should be interpreted with caution due to shortcomings of data that have been aggregated, such as studies' risk of bias, publication bias, and small-study effect 60,61 .

Conclusion
There was a weak correlation between the subjective pain scales and the Analgesia and Nociception Index, i.e., a part of pain self-report is explained by nociception assessed through ANI. Therefore, in the perioperative period, fully or partially conscious children or other individuals, who cannot self-report their pain, might benefit from using ANI during health procedures.

Data availability
All data generated or analysed during this study are included in this published article (and its Supplementary Information files).