title stringlengths 1 1.19k | keywords stringlengths 0 668 | concept stringlengths 0 909 | paragraph stringlengths 0 61.8k | PMID stringlengths 10 11 |
|---|---|---|---|---|
Supplementary Information | The online version contains supplementary material available at 10.1186/s12885-023-10945-9. | PMC10183678 | ||
Keywords | PMC10183678 | |||
Background | cancer | CANCER, EVENT | Since the beginning of 2020, COVID-19 has spread worldwide - including in Germany [These strategies affect all patients but particularly vulnerable patients such as cancer patiens [Such an unpredictable event as the COVID-19 pandemic may affect study conditions. The present survey investigates the influence of the pandemic on medical care in cancer patients. This supplementary survey was conducted as part of the intervention study “Patient information, communication and competence empowerment in oncology (PIKKO)” [ | PMC10183678 |
Methods | PMC10183678 | |||
Setting – the background of the PIKKO study | cancer | CANCER | PIKKO is a care concept funded by the Innovation Fund of the Federal Joint Committee in Germany (Gemeinsamer Bundesausschuss, funding number 01NVF17011) which supplements oncological care with an additional counseling and information pathway [But it was not only the surveys that were affected by the pandemic, but also the intervention itself. The use of PN and offers of the SCS was often linked to a visit to an on-site facility (clinic, practice, course rooms) and thus to movement in the public space. This can be a barrier for many cancer patients because they can no longer participate in the study due to external circumstances - such as lockdown strategies - or they do not want to participate due to their own fears. Fortunately, the PN could also be reached by telephone from the very beginning, so this was already a familiar alternative method of contact for the PIKKO patients. Beginning 30 Mar 2020, the SCS offered its psychological and psycho-social counseling during extended telephone hours and produced course videos to offer the content of the courses to patients online. Other SCS courses took place outdoors (Nordic Walking, QiGong). “My PIKKO” operated independent of location anyway. This meant that the entire intervention was still available.This supplementary survey, which was not included in the original study design of PIKKO, examines the impact of the pandemic on patients and the use of the intervention modules. Only the IG data collection took place under lockdown conditions and only the IG used the intervention modules. Therefore, all participants in this supplementary survey were from the IG. Participants were already informed about data privacy as part of the PIKKO study (The ethics committee of medical association of the Saarland approved the study protocol on 2 Nov 2017; approval number 114/17. The informed consent by study participants is obtained in a written way.) and the supplementary survey participation was voluntary. | PMC10183678 |
Participants and inclusion criteria | cancer disease | All participants of the supplementary survey met the inclusion criteria of the PIKKO study (age 18–90 years, diagnosis of any cancer disease, treatment by doctors from the Saarland, insured with one of the four statutory health insurance companies participating in this study) [ | PMC10183678 | |
Design | cancer | CANCER, GROUP B | In the present survey, we investigated two groups with a quasi experimental design. Group A was not affected (“affected” is related to participation in the PIKKO study) by the lockdown because the patients went through the PIKKO intervention as planned. The end of the PIKKO intervention or the voluntary exit from the PIKKO accompanying survey took place before the lockdown (16 Mar 2020). Group B was affected by the lockdown (in connection with participation in the PIKKO study). Part of the PIKKO intervention and/or survey was conducted during the lockdown, so the end of PIKKO was in or after the lockdown. Group B should still have regular contact (including face-to-face) with the PN, attend (on-site) courses or counseling sessions of the SCS, and complete surveys (which were mailed). All of these may require direct contact, which was limited by the pandemic containment strategies. In addition, as late participants in the PIKKO study, they are still more likely to be in active cancer treatment and are likely to be more frequent guests at medical facilities. Group assignment was not random; rather, but was determined by the timing of inclusion in the PIKKO study (Group A: early inclusion = already enrolled in PIKKO for an average of 358 days at the time of lockdown; Group B: late inclusion = already enrolled in PIKKO for an average of 167 days at the time of lockdown).The supplementary survey took place from 31 July 2020 to 31 Aug 2020. | PMC10183678 |
Variables | cancer | CANCER, DISEASE | To assess the impact of pandemic containment strategies on our cancer patients, we asked them questions about four aspects: Restrictions (“Have you had any restrictions with regard to your disease since 16 Mar 2020?: Yes/No If yes, which?”), influence on disease (“Do you think that the limitations due to the COVID-19 pandemic have an influence on the course of your disease?: Yes/No”), use of PIKKO (“Did you use parts of the PIKKO intervention during the limitations due to the COVID-19 pandemic?”), and their own sense of burden (“8 sub-questions on stressful situations to the assessment of the burden of the restrictions (CBS)”).Our self-designed COVID-19 conditional burden scale (CBS) questions covered the points of (1) hygienic strategies, (2) change of appointments, (3) movement in the public space, (4) cover mouth and nose, (5) no accompanying persons, (6) interaction with the medical staff, (7) interaction with the nursing staff, (8) ban on visits to the wards. Each of the questions could be rated on a 5-point Likert scale (0 = not stressful, 1 = a little stressful, 2 = moderately stressful, 3 = much stressful, 4 = very much stressful). Since not every patient was exposed to all eight stressful situations, the mean should be taken only for the questions answered, so missing (= item did not apply or was not ticked) enters the equation. The following formula was used to compute the CBS score (= the mean of all the patient’s responses): CBS-Score = Sum(CBS1, CBS2, CBS3, CBS4, CBS5, CBS6, CBS7, CBS8) / (8 – Sum(Missing)).The score ranges from 0 (no load) to 4 (heavy load). Cronbachs Alpha for the cases where all 8 items were completed is 0.9 (N = 52).Furthermore, previously collected utilization data from the main PIKKO study concerning the intervention modules were analyzed for the COVID-19 period. | PMC10183678 |
Data sources | The four questions concerning COVID-19 were collected using a two-page questionnaire.From our regular patient survey [ | PMC10183678 | ||
Bias | cancer | CANCER | Since the cancer patients interviewed were already part of the main PIKKO study, a selection bias can be assumed here. | PMC10183678 |
Study size | A full survey of all living PIKKO-IG participants was intended (n = 503). | PMC10183678 | ||
Statistical methods | cancer | REGRESSION, CANCER, DISEASE | First, selection effects were investigated by comparing participating and non-participating patients in the supplementary survey. To examine the selection bias, chi-square tests and F-tests were performed (independent variable: patient included versus excluded, depended variables: age, sex, etc.). Next, we compared the both groups of survey participants (not affected, A, and affected, B, by lockdown with regard to PIKKO) regarding to socio-demographic data, disease data, and treatment as well, using chi-square tests and F-tests.Then we compared both groups (not affected, A, and affected, B, by lockdown with regard to PIKKO) regarding health care related variables.To examine the restrictions in relation to the disease we conducted chi-square tests.To quantify the burden due to the pandemic containment strategies, we applied first a linear regression. All single items of the COVID-19 CBS and the sum score of the CBS (in separate calculations) were used as dependent variables. As independent variables we considered Group (0: not affacted, 1: affected), age (grouped by median, 0: under 60 years, 1: 60 + years), gender (0: female, 1: male), children in the household (0: no, 1: yes), financial burden (0: no, 1: yes), period of the most recent illness (dummy variable A: up to 1 year versus 6 + years, dummy variable B: 2–5 years versus 6 + years) and cancer treatment at baseline (dummy variable A: active treatment versus no active treatment, dummy variable B: only rehabilitation versus no active treatment). Based on the estimated regression coefficients we estimated adjusted group means and compared them with t-tests.The assumption of an influence on the course of the disease and using of parts of the PIKKO intervention during the lockdown was investigated with chi-square tests.Cramer-V [Missing values did not occur in the dichotomous questions. Missing answers in the CBS that occurred when an item/situation did not apply to the patient were marked as “not applicable” were included in the calculation of the CBS score as “8 - Sum(Missing)”. Only if all subitems were “not applicable”, these cases were excluded from the CBS analysis.In all analyses, the level of significance was α = 0.05.Data on the utilization of the PIKKO modules (Patient Navigator, SCS counseling and courses, knowledge database) were analyzed descriptively. | PMC10183678 |
Results | PMC10183678 | |||
Participants | GROUP B | A total of 503 (Group A = 241, Group B = 262) patients were contacted (the entire PIKKO intervention group). Of these, n = 356 returned a completed questionnaire of supplementary survey (n
Flow chart of the supplementary surveyFor unknown reasons 147 patients did not respond and could not be interviewed again in the main PIKKO study (126/147, 85.7%). Others were deceased (9/147, 6.1%), withdrew for health reasons (2/147, 1.4%), or completed the PIKKO surveys but no longer responded (10/147, 6.8%). | PMC10183678 | |
Descriptive data | First, selection effects were investigated, to identify whether the sample of participating (n = 356) and non-participating patients (n = 147) differed. Corresponding statistics are listed in Table
Description of the sample and determination of the differences between the subgroups non-participants and participants as well as groups A and B. Statistical differences (determined by F-test or chi-square test) are marked with asterisksYears of education (school + vocational)[m (sd)]
The amount of missing data in the burden data varied greatly from item to item: n | PMC10183678 | ||
Main results: comparison of “not affected” and “affected” by lockdown with regard to PIKKO | DISEASE | Comparisons of group A and group B in terms of socio-demographic data, disease data, and treatment (Table | PMC10183678 | |
Restrictions in relation to the disease | Out of all participants, 134 (37.6% of the sample) reported any restrictions. Group A and B differed significantly (χ²[ | PMC10183678 | ||
Burdens due to the pandemic containment strategies | REGRESSION | The most perceived burdens of the entire sample were: „Restriction or ban on visits while I was on the ward” (m = 1.92, sd = 1.542); “Wear a protective mouth-nose mask” (m = 1.33, sd = 1.282) and “Restrictions on accompanying persons” (m = 1.19, sd = 1.355). As the analyses show, there were no group differences between group A (not affected) and group B (affected), except for the burden of “restrictions on accompanying persons”. Other factors were more determinant in how strongly a burden is perceived (see Supplementary Material
Regression-adjusted means, standard deviation, t-statistic and significance (p-value) of the significant factors of the linear regression analyses (see additional file 1) of the burden values | PMC10183678 | |
Assumption of an influence on the course of the disease | DISEASE, SAID | Although 39% (140/356) said they fear that the restrictions will have an influence on the course of their disease, in most cases a psychological influence (124/140, 88.6%). The two groups did not differ on this point (χ²[ | PMC10183678 | |
Using of parts of the PIKKO intervention during the lockdown | 55.2% of the participants in group B (123/223) and 36.8% in group A (49/133) reported to the additional survey that they had used some PIKKO moduls during the lockdown. The two groups differed significantly on this point (χ²[ | PMC10183678 | ||
Other analyses: general use of the PIKKO modules during the lockdown | Beyond the information from the supplementary survey, the data from the regular patient questionnaires (contacts to PNs), the SCS data and the logfile data from the knowledge database provide information about the use of the PIKKO modules during the lockdown.The contacts to the PNs are shown in Fig.
The utilization of the PIKKO-intervention-moduls (Utilization of SCS psychological psycho-social counseling is shown in Fig. As with counseling, both the number of offerings (colored areas) and the number of participants (lines) for courses (Fig. The knowledge database (see Fig. | PMC10183678 | ||
Discussion | cancer, life-threatening diseases, PIKKO cancer | REGRESSION, CANCER | It was the aim of this supplementary survey to identify the restrictions and burdens of the pandemic containment strategies on the PIKKO cancer patients and thus on the PIKKO study itself.To this aim, we surveyed participants and analyzed these data using frequency analyses and linear regression analyses to determine which factors more strongly influenced patients’ perceived burden.Our first hypothesis that cancer patients in the PIKKO study felt restricted by the COVID-19 strategies, especially those affected by the lockdown (regarding PIKKO), can be confirmed. Fears of an impact on recovery were expected by the patients, which was similarly found by the online survey of the PRIO Working Group [Two of the greatest burdens of the COVID-19 lockdown were posed by the „ban on visits to the wards“ and „no accompanying persons“, two conditions that were a great imposition for patients with life-threatening diseases such as cancer with an increased need for assistance [Burdens due to change of appointments were also reported. Even though Fröhling et al. [The second hypothesis, that the restrictions caused a decrease in the use of person-centered intervention modules, can only be partially confirmed. The data showed that PN, as well as the SCS counseling and course service and the knowledge database, continued to be used, but in an adapted form. Briefly at the beginning of the lockdown, there were more contacts to the PN, especially live contacts. Many PNs were employed in clinics. This presumably responded to the ban on visits to the ward for inpatients. SCS also reported significantly more telephone consultations during the months of the lockdown. The SCS expanded this offer when face-to-face contact was restricted. Some offerings, such as nutrition courses, were limited, while outdoor Nordic Walking was offered continuously. These types of restrictions were also reported by Fröhling et al. in cancer care centers, in which there were restrictions on psycho-oncology, nutrition and exercise therapies, and social counseling for up to 12 weeks [The present survey revealed that the pandemic had negative effects on the medical care of cancer patients. Supportive offers such as counseling by PNs in clinics or by SCS counselors via phone, as well as adapting and maintaining other services such as exercise programs, are possible and necessary to mitigate the negative aspects of the pandemic restrictions. Others also reported good uptake for new telepsychological services during pandemic periods [ | PMC10183678 |
Strengths of the study | cancer | CANCER, DISEASE | We were able to provide insight into how an unforeseen disruption such as the COVID-19 crisis can impact an intervention study. This showed that a combination of flexibility (e.g., switching courses), diversity (e.g., face-to-face opportunities for patients), and location independence (e.g., web-based knowledge base) ensures that offerings for patients (here, cancer patients) can be maintained even when much else is no longer possible.Furthermore, it became apparent that burdens felt in addition to the disease (due to containment strategies and fears and worries) are not perceived with the same intensity by all groups of people, and that socio-demographic factors may be more decisive for this perception. | PMC10183678 |
Limitations | cancer | CANCER | Data on contacts to the PN are based on patient self-reports, some of whom were undergoing cancer treatment. There is evidence that chemotherapy can impair memory and recall [The supplementary survey took place in the summer of 2020, a few weeks after the first lockdown. Memories of something that happened a few weeks prior could be distorted as described above.The sample of this supplementary survey is not representative of the PIKKO intervention group in all respects. For example, patients were more likely not to participate if they were still working, younger, had children in the household, or reported financial worries before the pandemic. | PMC10183678 |
Acknowlegdements | Cancer | CANCER | We would like to thank Anna Bäcker, Katja Brenk-Franz, Heide Wenzel and Clara Korn from IPMPP Jena for their valuable support in planning and conducting the PIKKO study.Furthermore, we thank Robert Terbach from the German Cancer Society for providing the transfer data for the knowledge database.We would also like to thank all participants in the consortium management of the PIKKO study, the IKK Südwest, for their cooperation.Last but not least, we thank all patient navigators, psychooncologists, medical as well as non-medical staff and of course the participants for their support. | PMC10183678 |
Authors’ contributions | RECRUITMENT | BS, UA and JH conceptualized the PIKKO study. NS and UA developed the design of this supplementary survey and data collection. CK, FB, NS, SR, and UA were involved in patient recruitment, study conduct, and data collection. NS and UA analyzed the data of this supplementary survey. CK, FB, NS, SR, and UA interpreted the study data. NS wrote the basic version of this publication and prepared Figs. 1 and 2. BS, CK, FB and UA completed and corrected the manuscript. All authors read and approved the final manuscript. | PMC10183678 | |
Funding | PIKKO was funded by the Innovation Fund of the Federal Joint Committee (Gemeinsamer Bundesausschuss) (01NVF17011) in the period from 31 to 2017 to 31 Dec 2020.Open Access funding enabled and organized by Projekt DEAL. | PMC10183678 | ||
Data Availability | The datasets used and/or analysed during the current study available from the corresponding author on reasonable request. | PMC10183678 | ||
Declarations | PMC10183678 | |||
Competing interests | The authors declare no competing interests. | PMC10183678 | ||
Ethics approval and consent to participate | This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the ethics committee of medical association of the Saarland (2 Nov 2017/No. 114/17). The informed consent by study participants was obtained in a written way. | PMC10183678 | ||
Consent for publication | Not applicable. | PMC10183678 | ||
Abbreviations | Cancer | CANCER | COVID-19 conditional burden scalecontrol groupintervention grouponline knowledge databasePatient information, communication and competence empowerment in oncologypatient navigator (PN)Saarland Cancer Society | PMC10183678 |
References | PMC10183678 | |||
Subject terms | COVID-19 PNEUMONIA | Chest computed tomography (CT) has played a valuable, distinct role in the screening, diagnosis, and follow-up of COVID-19 patients. The quantification of COVID-19 pneumonia on CT has proven to be an important predictor of the treatment course and outcome of the patient although it remains heavily reliant on the radiologist's subjective perceptions. Here, we show that with the adoption of CT for COVID-19 management, a new type of psychophysical bias has emerged in radiology. A preliminary survey of 40 radiologists and a retrospective analysis of CT data from 109 patients from two hospitals revealed that radiologists overestimated the percentage of lung involvement by 10.23 ± 4.65% and 15.8 ± 6.6%, respectively. In the subsequent randomised controlled trial, artificial intelligence (AI) decision support reduced the absolute overestimation error (Trial registration: | PMC10039355 | |
Introduction | CAD | CAD, PULMONARY INVOLVEMENT, SECONDARY | The COVID-19 pandemic has created new ways in which existing and developing technologies are used in radiology. Although definitive diagnosis relies on real-time reverse-transcriptase-polymerase chain reaction (RT–PCR), CT still plays an essential role in the screening and monitoring of COVID-19 evolution, setting patient discharge criteriaRadiologists measure pulmonary involvement in COVID-19 using either a quantitative assessment of the overall involvementCOVID-19 quantification judgement process (left) compared to the circle size comparison problem (right). In both cases, the apparent proportion of COVID-19 involvement (red) in relation to total lung area (blue) appears greater in the bottom examples. This is due to the overinfluence of a primary linear dimension (1D segment ratio = 80%) that is often insufficiently adjusted to the 2D context (2D circle ratio = 60%). This leads to an overestimation of the perceived percentages.Area judgement is a century-old field of study in cognitive psychology, as human beings exhibit an acute lack of precision in visual geometric comparisonsDespite the wide adoption of lung involvement scores, the area judgement cognitive bias remains unaddressed in radiology. For the first time, radiologists are required to geometrically compare such irregular and disparate shapes, on such a large scale and all the available methods have one common denominator: the reliance on the reader's volume perception. This study analyses the perception risks of these measures and clinically tests an AI-based mitigation strategy.To statistically confirm if this problem exists, we first conducted two experiments that analyse the two-step thinking process of a radiologist analysing CTs for COVID-19. The first experiment isolated and investigated the estimation step in simulated data, then the second examined the bias’s effect over the whole process. The primary hypothesis (H1) is that this geometric ratio assessment is prone to an overestimation bias. Next, a randomised clinical trial was conducted to study whether the bias could be mitigated by using a commercial AI clinical support system. The secondary hypothesis (H2) is that the reader’s objectiveness can be improved with the use of computer-aided diagnosis (CAD). As the study does not propose to address the development process of a new AI CAD system, it employs Rayscape | PMC10039355 |
Results | PMC10039355 | |||
Bias validation in the synthetic experiment | The first experiment involved 40 radiologists who answered a survey regarding 18, nine intuitive and nine unintuitive, synthetically generated images (Fig. Synthetically crafted lung CT slices. The same starting slice was predelineated differently to simulate real involvement rates (18 slices and 9 involvement rates in total). Samples shown on the bottom row (unintuitive) should be more susceptible to overestimation, according to area perception theory.We compared the radiologists’ mean errors over the nine intuitive cases against the mean errors over the nine unintuitive (bias-susceptible) cases. We found that radiologists, on average, could be objective in judging the geometric ratios in the basic samples (mean difference = 1.193%), but they showed an overestimation bias in the unintuitive cases (COVID-19 involvement overestimation across synthetic cases and radiological seniority. (Next, we studied the radiologists’ bias tendency (i.e., mean difference of overestimation) with respect to their seniority (years of experience in radiology). Correlation analysis showed a moderate Pearson coefficient of − 0.405 ( | PMC10039355 | ||
Overestimation in retrospective analysis | Next, the bias was further studied in a retrospective analysis of the CT studies of 109 patients with RT–PCR-confirmed COVID-19 from HOSP-TM and EXMED. As part of the standard clinical practice, all radiological reports mentioned the total percentage of COVID-19 lung involvement.Similar to the previous experiment, a visible trend emerged in both hospitals (Fig. Retrospective overestimation analysis. ( | PMC10039355 | ||
CAD effectiveness in the clinical trial | CAD, fibrosis | PULMONARY INVOLVEMENT, PNEUMOTHORAX, FIBROSIS, CAD | To study the effect of CAD on the involvement assessment problem, we used the AI-PROBE protocol with A total of 85 enrolled patients were randomised between the control arm (AI intervention off) and the experimental arm (access to AI results). An additional analysis exclusion criterion eliminated 9 studies that failed to mention the involvement quantification marker from the final radiological report. Across all randomised patients, 76 CT studies (CAD access, n = 38; No CAD access, n = 38) were successfully analysed for the study outcome. This was a representative, consecutive sample of COVID-19 patients examined at HOSP-TM between 21 February 2022 and 15 March 2022 who met the inclusion criteria. The participants (37 females and 39 males) were aged 20–89 years (median = 72 years, interquartile range (IQR) = 66–81 years) and covered a wide spectrum of clinical conditions ranging from asymptomatic or milder to critical COVID-19 cases, in both inpatient and outpatient care. The results of the patient selection process are presented in Fig. CONSORT-AI flow diagram describing the patient selection process.The patients’ true lung involvement ratios (reference standards) fit an exponential distribution (λ = 0.032). The overall mean pulmonary involvement was 32.319% (CAD access, m = 31.381%; No CAD access, m = 33.157%), and the distribution tail ended at a maximum involvement percentage of 78.030% (CAD access, M = 70.748%; No CAD access, M = 78.030%). The number of patients with other reported clinically important findings that could influence the quantitative analysis (e.g., pneumothorax, fibrosis) was also evenly distributed, with nine (23.7%) in the control group and seven (18.9%) in the intervention group. The root mean square error of the AI outputs was 4.206, with no apparent skewness between severity subgroups, as shown in Extended Data Figure Measured estimation differences between the two arms. (AI intervention reduced the mean overestimation difference from 9.471 ± 6.561 (95% CI) in the control arm, to 0.983% ± 5.181 (95% CI) (Fig. | PMC10039355 |
Discussion | CAD, AI-induced | LUNG, CAD | This study analysed the CT quantification of lung involvement in COVID-19 in three ways. The first synthetic experiment validated the translation of a theoretical model from psychophysical science to radiology, demonstrating that radiologists are susceptible to a cognitive bias that leads to overestimating the level of involvement. The second experiment retrospectively revalidated that this cognitive bias occurs with data from real COVID-19 patients and further measured the extent of the overestimation. The results showed that the effect was even stronger in real investigations. Finally, a randomised clinical trial demonstrated that AI is a useful tool for reducing the area perception bias among radiologists.The impact of the observed overestimation errors in the last two experiments (15.829% ± 6.643 and 9.471% ± 6.561, respectively) is substantial in a clinical context. According to previous longitudinal studiesThe bias magnitude varied considerably across the three experiments and between the two sites. A bias jump was expected when switching from the 2D data of the first experiment to the 3D data of the following two, with multiple adjustments needing to be made by the reader. This is in line with the findings of the second experiment, although not with the clinical trial results. This might be due to the lack of patients displaying above 80% lung involvement, where the bias was most pronounced in the synthetic experiment. Moreover, the second analysis showed vast differences in perception between the two hospitals. There are various possible causes for this finding, such as interhospital protocol differences when analysing COVID-19 lesions, the different types of institutions or the fact that only HOSP-TM benefited from CAD analysis software. Consequently, clinical trial showed that the use of AI might be an influential factor, as it significantly and consistently reduced the perception errors.Lung involvement is an empirically developed measure that was previously shown to be not only predictive of patient outcomes but also decisive in establishing the patient treatment course. The study of interobserver agreement of COVID-19 scoring systems has led to varying, often positive resultsThe success of the AI arm in the clinical trial is accredited in part to the widespread adoption of the CAD system within HOSP-TM. The analysis was integrated into the hospital’s picture archiving and communication system (PACS). which allowed clinicians to use AI with minimal workflow adjustment. However, not every radiologist chose to use this assistance. The AI decision support was predominantly popular among the younger radiologists, who were also the ones that demonstrated the greatest bias susceptibility. However, the exact engagement of the radiologists could not be quantitatively followed, by design. This is important, to reduce any interference and thus influence on how the CAD is perceived and used throughout the radiology department. The pragmatic design of AI-PROBE allowed studying the effect even with partial engagement of the radiology department, similar to realistic expectations inside a hospital. The interarm difference might be even more pronounced with wider adoption, although the study strived to preserve the natural adoption extent of the software.The AI-powered CAD system was effective in mitigating perception errors. However, caution must be taken in accidentally trading the area perception bias for other AI-induced biases. AI inconsistencies in underserved patient populations are a known issueOur study demonstrated that quantification of the involvement of the lungs in COVID-19 on CT scans is a perception-sensitive process prone to cognitive overestimation bias. This is of key importance given the wide use of the marker, although it was shown to be controllable with an AI decision support system. This reinforces the benefits of human-AI synergy and strengthens the need to further study the adaptability of radiology to rapid technological and methodological changes. | PMC10039355 |
Methods and materials | EMERGENCY | All procedures were conducted in conformity with the Declaration of Helsinki and International Conference on Harmonisation Good Clinical Practice guidelines. The clinical trial received approval from the Ethical Committee for Scientific Research of Pius Brinzeu County Emergency Hospital (no. 282/01/02/2022). The informed consent was collected accordingly. The retrospective analysis of data originating from EXMED received exception from informed consent (no. 14/12/02/2022) from the same committee. The clinical trial was registered on 16/03/2022 (ClinicalTrials.gov number NCT05282056). | PMC10039355 | |
Preliminary analysis | pulmonary deterioration | To test H1 and facilitate power calculation before conducting a full prospective clinical trial, two preliminary experiments were carried out.The first experiment involved 40 voluntary diagnostic radiologists. A call for volunteers reached physicians from eleven Romanian medical institutions. The eligibility criteria consisted of practising diagnostic radiologists of any level of experience on thoracic CTs. The participants estimated the total percentage of pulmonary deterioration in simulated CT slices based on a predelineated involvement contour (Fig. For the second experiment, a random sample of 109 studies of patients with RT–PCR-confirmed COVID-19 from HOSP-TM and EXMED were analysed retrospectively. Each study contained at least one noncontrast pulmonary CT investigation, and its corresponding radiological report, acquired between August 2021 and January 2022. The lung involvement percentages were automatically extracted using regular expressions and manually reviewed to correct for any parsing mistakes. | PMC10039355 | |
Clinical trial design | CAD, lung volume deterioration | CAD | To test H2, we used the AI-PROBE-2 protocol with The clinical trial took place at HOSP-TM, between 21 February 2022 and 15 March 2022., where physicians were already using Rayscape, a commercial CAD system for COVID-19 volumetric quantification. The Rayscape CAD system is an existing medical device that complies with the European Economic Area regulations, it adheres to the quality management standard ISO 13485 and has obtained the CE mark. For the entire period of the study, Rayscape version v2-1.286-1.415-2.262, launched in January 2022, was used, which showed the AI analysis in the form of coloured delineated volumes along with a percentage of total lung volume deterioration, similar to standard clinical practice at the hospital. The AI analysed all CT studies in real-time and sent the analysis to the PACS. Half of the studies received the AI analysis, and the other half received the disclaimer. Aside from the disclaimer message of lacking the AI assistance in the controls during the development of the study, the radiologists did not receive any other instructions.The allocation process was performed programmatically, in real-time, by Rayscape’s Dicom Server using the default pseudorandom number generator of Python 3.8 | PMC10039355 |
Data collection | The enrolment inclusion criteria included an age of 16 or older (as per the Rayscape technical requirements), a non-contrast CT examination and positive RT-PCR results confirming COVID-19. The entire enrolment flow is illustrated in Fig. Chest CT investigations were performed using NeuViz 16 Essence (Neusoft Medical Systems), Revolution EVO (GE Healthcare) and MX16 (Philips Healthcare) scanners with slice thicknesses ranging from 1.25 to 1.5 mm.As the reference standard involvement percentages in the retrospective analysis and the clinical trial, two non-participating radiologists with at least seven years of experience in thoracic diagnostic radiology manually annotated all images at the pixel level using the ePAD | PMC10039355 | ||
Statistical analysis | REGRESSION | Based on the two preliminary experiments, we calculated that a sample size of 32 CT studies for each arm would be sufficient to detect a mean difference of 5% (alpha = 0.05, beta = 0.8) with regard to H2. To account for the risk of post-analysis exclusions, 20 extra patients were planned to be enrolled. We did not assume that the AI intervention would be noninferior in any setting; thus, two-tailed tests were used.A two-sided paired t test was used to analyse the mean differences between the two types of samples analysed by the radiologists in the first experiment. A two-sided two-sample t test was used to analyse the bias differences between the two arms of the trial. All differences were assessed for normality both visually (Q-Q plot) and numerically (Shapiro–Wilk test).Pearson correlation coefficient and significance value were calculated to validate the simple linear regression fitting. Despite the large residuals, the even spread of the outliers did not violate either the homoscedasticity or the normality assumptions of the regression analysis.Data management and analysis were conducted using SciPy 1.7.3 | PMC10039355 | |
Supplementary Information | The online version contains supplementary material available at 10.1038/s41598-023-31910-3. | PMC10039355 | ||
Author contributions | B.A.B. and A.B. designed the study concept and planned the study. A.B. originally observed and informally validated the cognitive bias presence in practice. B.A.B. wrote the first draft of the study. B.A.B. and A.B. had full access to the data in the trial and auxiliary experiments and take responsibility for the integrity of the data analysis. P.G.A., I.B. and A.B. conducted the ground-truth labelling process and the survey. C.A., A.T. and S.I. implemented the technical requirements of the AI-PROBE trial design, setup of the labelling infrastructure and data retrieval from PACS. M.B. and S.I. facilitated the administrative requirements for the smooth conduction of the study. C.R., D.C., A.S.B., M.M. and F.B., advised on the analysis or interpretation of the data. All authors commented on and revised the manuscript and approved its submission. | PMC10039355 | ||
Data availability | The raw data (CT studies, radiological reports, patient characteristics) are not publicly available, as consented by the ERBs and patients for research use only by the investigators of this study. If other authors are interested in additional experiments on the collected data, a request can be made to the corresponding author (B.A.B.) for the analyses to be made in collaboration with the current authors. | PMC10039355 | ||
Code availability | The Rayscape AI system is a publicly available medical device, available as a commercial software product and it’s also offered upon request for scientific enquiries. The code that generates the samples used in the first experiment (Fig. | PMC10039355 | ||
Competing interests | This study was organised and coordinated by Rayscape, a start-up company developing AI algorithms for medical images and producer of the mentioned AI system. B.A.B., A.B., P.G.A., I.B., M.M.B., C.A., A.T., and S.I. are employees of Rayscape that own stock in the company. The rest of the authors declare no competing interests. No participating radiologist in the clinical trial had any commercial or utilitarian incentive in using the software nor were they made aware of any endpoint of the study that could potentially influence their perception of the software. | PMC10039355 | ||
References | PMC10039355 | |||
Summary | Contributed equally to this work. | PMC10025757 | ||
Background | PWH, HIV infection, CVD, Inflammation, HIV (PWH | CARDIOVASCULAR DISEASE, CVD, INFLAMMATION, HIV INFECTION, INFLAMMATION | Persons with HIV (PWH) have an increased risk of cardiovascular disease (CVD) compared to HIV-seronegative individuals (SN). Inflammation contributes to this risk but the role of lipid mediators, with central roles in inflammation, in HIV infection remain to be established; further aspirin reduces CVD risk in the general population through production of some of these anti-inflammatory lipid mediators, but they have not been studied in PWH. | PMC10025757 |
Methods | PWH | We evaluated the relationship between plasma lipid mediators (i.e. 50 lipid mediators including classic eicosanoids and specialized pro-resolving mediators (SPMs)) and HIV status; and the impact of aspirin in PWH on regulating these autacoids. Plasma samples were obtained from 110 PWH receiving antiretroviral therapy (ART) from a randomized trial of aspirin (ACTG-A5331) and 107 matched SN samples (MACS-WIHS Combined Cohort). | PMC10025757 | |
Findings | PWH | PWH had lower levels of arachidonic acid-derived pro-inflammatory prostaglandins (PGs: PGE | PMC10025757 | |
Interpretation | inflammation, PWH, CVD | CVD, INFLAMMATION | Together these observations demonstrate that plasma lipid mediators profiles, some with links to systemic inflammation and CVD risk, become altered in PWH. Furthermore, aspirin intervention did not increase levels of aspirin-triggered pro-resolving lipid mediators, consistent with other reports of an impaired aspirin response in PWH. | PMC10025757 |
Keywords | PMC10025757 | |||
Research in context | PMC10025757 | |||
Evidence before this study | inflammation, PWH, CVD, HIV (PWH | CVD, INFLAMMATION, CARDIOVASCULAR DISEASE | Aspirin can reduce cardiovascular disease (CVD) risk in the general population, in part through its anti-inflammatory properties. For example, aspirin is known to reduce levels of pro-resolving (i.e. anti-inflammatory) lipid mediators of inflammation. Persons with HIV (PWH) have higher risk of CVD and aspirin administration could potentially reduce CVD risk in this population. We searched PubMed through August 20, 2022 for studies on “aspirin” AND “HIV” AND “lipid mediators of inflammation” (or variations of these terms), but could not find any study that comprehensively assessed pro-inflammatory and pro-resolving lipid mediators of inflammation in PWH before and after aspirin intervention. Further, we only found one small study (n < 20 per group) that assessed lipid mediator of inflammation profile in PWH. | PMC10025757 |
Added value of this study | inflammation, PWH | INFLAMMATION | In a randomized trial of aspirin intervention in 110 PWH, we comprehensively assessed the levels of pro-inflammatory and pro-resolving lipid mediators before and after aspirin intervention. Our results show that, in the interval tested, 12 weeks of aspirin administration did not increase levels of pro-resolving lipid mediators of inflammation in PWH. In addition, utilizing a well-characterized and well-matched cohort of HIV-seronegative individuals, our study also is the largest and most comprehensive study to characterize data on alterations of lipid mediators of inflammation by HIV status. | PMC10025757 |
Implications of all the available evidence | inflammation, PWH | CVD, INFLAMMATION | Our findings suggest that the lipid mediator of inflammation profile, some with important functions in CVD risk, is distinct by HIV status. Importantly, our data also suggest aspirin does not increase the production of aspirin triggered specialized pro-resolving mediators (AT-SPM) in PWH. These results are in line with other findings that inhibition of platelet activation in response to aspirin is also blunted in PWH. Given the potent platelet directed activities of AT-SPM, we believe that our present observations provide insights into mechanism behind the failure of aspirin to regulate platelet responses in these patients. Taken together, these results suggest that aspirin might potentially be less effective in PWH. The clinical implication of these results for CVD prevention in PWH needs to further addressed in future studies. | PMC10025757 |
Introduction | inflammation, PWH, CVD, HIV (PWH | CVD, INFLAMMATION, CARDIOVASCULAR DISEASE | Persons with HIV (PWH) have an increased risk of cardiovascular disease (CVD) compared to HIV-seronegative individuals (SN).Despite the critical role that inflammation plays in HIV outcomes such as CVD, the role of lipid mediators of inflammation (Bioactive lipid families. Illustration summarizing the bioactive mediator families evaluated in the present study together with their parent fatty acid and overall classification of their biological activities.It is now appreciated that lipid mediators mainly from the omega-3 essential fatty acids eicosapentaenoic acid, n−3 docosapentaenoic acid and docosahexaenoic acid and termed as specialized pro-resolving mediators (SPM) are central to the reprogramming of immune responses to limit inflammation.Studies investigating the mechanisms of aspirin in the general population, where it used to reduce CVD risk, demonstrate that, To address some of the gaps in knowledge related to the role of lipid mediators (and their biosynthetic pathways) and in particular SPM in PWH, we compared levels of lipid mediators of inflammation in peripheral blood of PWH on suppressive ART and SN from well-characterized study populations. | PMC10025757 |
Methods | PMC10025757 | |||
Study population | AIDS, inflammation, PWH | INFLAMMATION, AIDS | For this study, we assessed lipid mediators of inflammation from adult PWH on suppressive ART and from adult SN.All PWH with available baseline plasma samples (N = 110) from the AIDS Clinical Trials Group (ACTG) A5331 study (NCT02155985) were included in this analysis. The A5331 study, conducted in the United States from August 2014 to March 2015, has been detailed elsewhere.Matched adult SN (87 men, 20 women) were selected from a large ongoing cohort study in the United States: The Multicenter AIDS Cohort Study (MACS) | PMC10025757 |
Ethics approval | Study participants provided written informed consent and the study was approved by relevant ethics committees ( | PMC10025757 | ||
Laboratory assessment | PMC10025757 | |||
Targeted lipid mediator profiling | EDTA plasma samples were collected from participants as part of each study (details in | PMC10025757 | ||
Soluble and cellular markers of inflammation | Plasma levels of soluble CD14 (sCD14) was measured at baseline as part of the parent A5331 study using enzyme-linked immunosorbent assays (R&D Systems, Minneapolis, MN). | PMC10025757 | ||
Statistical analysis | inflammation, PWH, A5331 | REGRESSION, HEAT, INFLAMMATION | A total of 50 unique lipid mediators were measured in plasma samples, with 42 SPMs and 8 pro-inflammatory lipid mediators (The first objective of this study was to compare plasma levels of i) lipid mediators and ii) metabolomes between PWH (pre-intervention) and SN individuals. To visually examine how mediators/metabolomes and these two populations cluster together, hierarchical clustering heat maps were created for mediators/metabolomes (The second objective of this study was to determine the association of lipid mediators of inflammation with the monocyte activation marker sCD14 among PWH (i.e. A5331 study participants at baseline). Logistic regression was used to determine the association of high inflammation (outcome variable of “high” sCD14 defined as the highest quartile (Q4) compared to Q1–Q3) with i) individual metabolome, ii) individual principal components (derived from unsupervised principal components analysis) of the metabolome, and iii) individual principal components of the mediators (further details in The final objective of this study was to study whether and how the administration of aspirin changed levels of lipid mediators among PWH (i.e. among A5331 participants pre- and post-intervention). We conducted analysis of variance (ANOVA) to assess the effect of each aspirin arm in A5331 (100 mg and 300 mg arms) relative to placebo on 12-week change (i.e. change in lipid mediators/metabolomes from baseline to 12 weeks) in mediators and metabolomes (further details in With the exception of analyses utilizing FDR-adjusted p-values, all other exploratory analyses used unadjusted p-values conservatively assessed at the 1% significance level to limit the number of false positive findings. | PMC10025757 |
Role of the funding source | The study sponsors did have a role in the study design; in the collection, analysis, and interpretation of data; in the writing of the report; and in the decision to submit the paper for publication. | PMC10025757 | ||
Results | PMC10025757 | |||
Study population characteristics | PWH | Our study population (N = 217) included 110 PWH on suppressive ART and 107 SN. As expected with the matching of PWH and SN cohorts, study population characteristics including age, sex, race/ethnicity and body max index were similar by HIV status (Study population baseline characteristics by cohort.BMI: body mass index.Fisher's Exact Test.Wilcoxon Test. | PMC10025757 | |
Lipid mediator profile in PWH and SN individuals | PWH | HEAT | To understand the profile of pro-inflammatory and pro-resolving lipid mediators in PWH (i.e. baseline: pre-aspirin intervention) and SN individuals, we measured plasma levels of 50 lipid mediators belonging to 12 lipid mediator families derived from four major fatty acids namely docosahexaenoic acid (DHA), n−3 docosapentaenoic acid (n−3 DPA), eicosapentanoic acid (EPA) and arachidonic acid (AA) (Abundance heatmaps of the overall population (PWH and SN combined) at the lipid mediator family level, showed that while there were differences among individuals, there was no clear clustering by HIV status at either the metabolome or individual mediator level (Lipid metabolome and mediator abundance bar plot. A: Metabolome Abundance bar plot. B: Lipid mediator abundance bar plot. Abundance heat map for metabolome (A, left panel) and lipid mediator (B, right panel) for all individuals (N = 217) are shown as a stacked bar graph. Metabolome and individual mediators were logDifferences in metabolomes by HIV status.Analysis of covariance was conducted to assess the differences in metabolomes by HIV status. All models are adjusted for sex, race/ethnicity, age, body mass index, smoking, drinking and statin use. FDR: false discovery rate.PWH Geometric LS Mean/SN Geometric LS Mean.Based on ANCOVA analysis, adjusted for covariates, of 16 metabolites with VIP > 1, PWH had lower levels of 4S,14S-diHDHA (Δ = 0.77 (0.64, 0.93); unadjusted p | PMC10025757 |
Relationship of lipid mediators with protein mediators among PWH | inflammation, PWH | INFLAMMATION | As PWH with higher levels of the monocyte activation marker soluble CD14 (sCD14) have an increased risk of morbidity and mortality, another objective of this study was to determine the relationship of lipid mediators and sCD14 among PWH.At the family level, no statistically significant associations with ‘high sCD14’ (defined as those in the highest quartile) were observed after FDR adjustment (While our primary focus was on sCD14, we also explored the relationship of the lipid mediators of inflammation with additional markers including sCD163, IL-6 and CD4 T-cell activation (CD38+HLA-DR+). At the metabolome level, only PG had a strong inverse association with CD4 activation (OR: 0.29 (0.12, 0.71)). We did not observe statistically significant associations of other metabolomes or mediators, and their PCs with these markers ( | PMC10025757 |
Effect of aspirin treatment on lipid mediators among PWH | PWH | PWH were randomized to 12 weeks of either daily aspirin 300 mg, aspirin 100 mg or placebo and self-reported adherence to the intervention was high.Effect of aspirin on lipid mediator metabolomes among PWH.The mean fold-change and mean fold-change percent from baseline is shown for the three study arms for each lipid mediator metabolome. Analysis of variance were conducted to test the effect of the 100 mg and 300 mg dose of aspirin as compared to the placebo.We then repeated the above analyses for the individual lipid mediators. The difference noted above at the metabolome level for PG was mostly driven by PGE | PMC10025757 | |
Discussion | inflammation, PWH | CVD, INFLAMMATION | In this study, we assessed the relationship between circulating lipid mediators of inflammation and resolution with HIV status, monocyte activation and aspirin treatment. When compared with SN, PWH on suppressive ART had lower levels of Tx and various PGs, and higher MaRIn our comparison between PWH on ART and SN, we observed lower levels of AA-derived PG and Tx in PWH, with TxBOur findings related to sCD14 and lipid mediators showed relationships among PWH on suppressive ART. sCD14 had positive associations with various PGs and TxBAfter aspirin intervention, we observed expected large reductions in levels of PGs and Tx with both doses of aspirin. Aspirin binds and acetylates the cyclooxygenase (COX)-2 enzyme active site, leading to blocking of PG and Tx production.The limitations of our study are related to assessment of lipid mediators only at the beginning and end of treatment, merging samples from multiple studies (i.e. ACTG, MACS/WIHS) and limited data on gut integrity markers. The mechanisms behind the observed changes in lipid mediators of inflammation in PWH are not clear and future studies will need to address this. Despite these limitations, our study has multiple strengths. These include well-characterized and well-matched study populations, measurement of a comprehensive panel of pro- and anti-inflammatory lipid mediators along with paired soluble protein marker data, and a randomized controlled design to study the effect of aspirin treatment on lipid mediators.In conclusion, we observed distinct lipid mediator profiles among PWH compared to SN individuals. There were also relationships observed between lipid and markers of monocyte activation important in CVD. n−3 DPA-derived RvD5, linked with improved intestinal integrity, were inversely associated with sCD14 in PWH, suggesting intervention potential for this SPMs to improve intestinal integrity and reduce monocyte activation. We also noted that aspirin did not increase levels of aspirin-triggered SPMs in PWH, in line with other findings showing a defective aspirin response in PWH, and could help partly explain why aspirin was unable to reduce monocyte activation and systemic inflammation in PWH. | PMC10025757 |
Data sharing statement | The authors confirm that all data are available upon request ( | PMC10025757 | ||
Declaration of interests | RS, PWH, TTB | PCT | PCT: Merck has provided her institution with funding for her research; Gilead and Lilly have also provided her institution with funding for her to conduct industry-sponsored clinical trials. PWH: Gilead has provided funding to his institution, and Merck has provided donation of study drug for NIH-sponsored trial. He has also received consulting fees and other support from Viiv Healthcare, Biotron, Gilead and Longeveron. JD is an inventor on patents related to the composition of matter and/or use of pro-resolving mediators some of which are licensed by Brigham and Women's Hospital or Queen Mary University of London for clinical development. AG: NIH, UNITAID and CDC have provided funding to her institution. TTB has received consulting fees from ViiV Healthcare, Theratechnologies, Janssen, Merck and Gilead. JAA: Atea, Emergent Biosolutions, Frontier Technologies, Gilead Sciences, GSK, Janssen, Merck, Pfizer, Regeneron and Viiv Healthcare have provided her institution with funding. RS: funding for current work was provided to institution by NIH and the ACTG network. The remaining authors have no conflicts of interest to declare. | PMC10025757 |
References | PMC10025757 | |||
Supplementary data | PMC10025757 | |||
Supplement_Combined | PMC10025757 | |||
Acknowledgements | RS | The authors thank the study participants for their time and contributions in the A5331 and MWCCS sites. We thank E Turner Overton for help with the study design. We also thank Jonathan Kerman, Jennifer Cohen, Eryka Wentz, Gayle Springer and MWCCS Data Analysis and Coordinating Center for all their help with participant and sample selection. This research was supported by a sub-award (ACTG NWCS 448) to RS from the Supplementary data related to this article can be found at | PMC10025757 | |
Subject terms | human behavior, Pain, pain | Colors are an important factor that influences different aspects of people's lives. However, little is known about the effects of colors on pain. This preregistered study aimed to investigate whether the type of pain affects the impact of colors on pain intensity. 74 participants were randomly divided into 2 groups according to the type of pain: electrical or thermal. In both groups, pain stimuli of the same intensity were preceded by different colors. Participants rated the pain intensity induced by each pain stimulus. Additionally, pain expectations related to each color were rated at the beginning and the end of the procedure. A significant effect of color on pain intensity ratings was found. Pain was most intense in both groups after red, whereas the lowest ratings were given after white. A similar pattern of results was observed for pain expectations. Expectations also correlated with and were found to be a predictor of experienced pain for white, blue, and green. The study shows that white can reduce, while red can alter the experienced pain. Moreover, it shows that the effect of colors is affected to a greater extent by the pain expectations rather than the pain modality. We conclude that the way colors influence pain broadens the current knowledge on effects of colors on human behavior and could help in the future both patients and practitioners. | PMC10115883 | |
Introduction | Color affects cognitive functions, hyperalgesia, pain | Colors have been proven to influence many aspects of life. Color affects cognitive functions such as attentionWhat is more, colors have been proven to affect pain. Red has been found to increase pain intensity more than green and blue when electrical pain stimuli are used; however, of all colors examined (red, blue, green, orange, yellow and pink), only green was found not to induce hyperalgesia when compared to no color conditionThe differential impact of colors on various modalities of pain could be attributed to, next to the influence of innate and learned associations and expectations, the distinct processing mechanisms involved in pain induced by electrical or thermal stimulation. Although many nociceptors respond to different stimulus modalities, some have more specialized response propertiesConsequently, based on the results of previous studies | PMC10115883 | |
Results | ±, pain | There were no differences between groups (electrical and thermal) in terms of participants' age (Basic descriptive statistics of the study participants, expressed as the mean (± standard deviation) or percentage (number) and p-values of between group analyses.Means and standard errors of differences between pain ratings associated with black and other colorsMeans and standard errors of differences between expectation ratings between black and other colors. | PMC10115883 | |
Pain analyses | ’, pain | The two-way mixed-design ANOVA performed on the pain ratings revealed a statistically significant main effect for ‘color’ (The post-hoc tests were carried out without the modality division as there was a significant main effect for ‘color’ but not for the ‘color’ x ‘modality’ interaction. The results showed that all color comparisons which revealed significant differences consisted of either white or red (Table Post-hoc differences in the pain ratings for all colors (both modalities merged).The Bonferroni correction was used for all post-hoc tests. Significant comparisons are marked with *. | PMC10115883 | |
Expectation analyses | ’ | The three-way mixed-design ANOVA that was performed on the expectation ratings showed a significant main effect for ‘color’ (As only the main effect of ‘color’ and the ‘color’ x ‘block’ interaction effect were significant, the post-hoc tests were carried out without the modality division. The results for the first expectation block showed a lot of significant differences, mostly related to red, white, orange and pink. In the second expectation block, the list of statistically significant differences was shorter and mostly related to red, but also to some other colors (see Tables Post-hoc differences in the first expectation block for all colors (both modalities merged).The Bonferroni correction was used for all post-hoc tests. Significant comparisons are marked with *.Post-hoc differences in the second expectation block for all colors (both modalities merged).The Bonferroni correction was used for all post-hoc tests. Significant comparisons are marked with *. | PMC10115883 | |
Secondary analyses | pain, ’ | REGRESSION | Due to the non-significant ‘color’ x ‘modality’ interaction, the correlation and regression analyses were carried out without the division into modalities. Correlation analyses revealed that expectation ratings from the first expectation block were correlated with pain ratings for the following colors: blue (The post-study questionnaire revealed that 62% of participants figured out the real aim of the study (Q1). Moreover, 87.3% of participants believed that colors could influence pain perception (Q2), and 57.7% declared that colors had affected their pain sensation in this study (Q3). The percentage of participants who thought that colors could influence pain perception in general as well as in this study was 56.3%. In contrast, those who thought that colors could influence pain perception in general but had not affected it in this study was 31%. The additional analyses showed that participants’ awareness of the study aim (‘color’ x ‘Q1’ interaction ( | PMC10115883 |
Discussion | chronic pain, white reduced pain, pain | CHRONIC PAIN | This study aimed to examine the influence of pain modality, either thermal or electrical, on the effect of colors on pain perception. Consistent with previous researchPain intensity and pain expectations were highest for red, despite the pain modality. Furthermore, our data revealed that white reduced pain the most compared to the other colors, and this was the only color that reduced pain against the baseline. To the best of our knowledge, this is the first time that a hypoalgesic effect of white has been found. White is associated with morality, honesty, purity, and cleanlinessIt is suggested that colors convey specific meaning and information and influence behavior through entrenched associationsThe strengths of this study include the use of a mixed-method design, which enabled us to explore in depth the effect of colors on pain perception and test it against the two types of pain. As a result, we not only found that colors influence pain regardless of the modality of pain, but we were also able to replicate previous findings which showed that red increases pain more than all other colors. This study’s results clearly demonstrate that colors affect pain in humans, and the most novel finding is that white is the most hypoalgesic color. However, the findings of this study have to be seen in the light of some limitations. First, although we calibrated pain stimuli intensity to 5/10 on NRS, the average pain intensity ratings for the first baseline were 4.1 and 3.5 in the electrical and thermal groups, respectively. We observed a non-significant sensitization effect between the first and the second baselines in both groups, but the average pain intensity ratings were still lower than 5/10 (4.2 in the electrical group and 3.8 in the thermal group). Research on classical fear conditioning shows that the more painful an unconditioned stimulus, the faster an aversive emotional association is establishedFurther investigation and broadening of the participant pool to a patient or pain-susceptible population would enable a generalization or review of the current findings. However, our results already show that there is a need for careful design of experimental and clinical protocols, as well as the data interpretation. Our findings could facilitate the methodology of other pain studies which use colors as cues or as part of a procedure (e.g. investigating chronic pain interventions or placebo effects). Additional research is needed to understand how the effect of colors manifests in other conditions (e.g. general practice care, hospital care) and how it can be implemented in clinical practice, such as pain management therapies for either chronic or acute pain. There is also a need for identification and research on mechanisms behind the effect of colors on pain and whether they are similar to some extent to mechanisms related, e.g., to open-label placebo. | PMC10115883 |
Material and methods | POLAND | The study protocol was approved by the ethics committee at the Institute of Psychology, Jagiellonian University, Cracow, Poland (KE/25_2021) and was preregistered in the Open Science Framework [osf.io/xrznm]. All methods were carried out in accordance with the Declaration of Helsinki. | PMC10115883 | |
Participants | alcohol abuse, abnormal color vision, pain | DISORDERS | A total of 124 healthy participants aged 18 to 35 years were initially recruited through advertisements on social media and job portals, and by word of mouth. The exclusion criteria were previous participation in pain experiments; being a student (3 years or more) or graduate of psychology, cognitive science, or a medical major; the presence of chronic or acute pain; alcohol abuse; having unremovable metal objects in the forearms; diagnosed neurological, cardiovascular, metabolic, musculoskeletal, or psychological disorders in the preceding six months; current drug consumption; abnormal color vision. The exclusion criteria were evaluated with an online questionnaire, and only participants eligible for the study were invited for an experimental session.Participants were informed that the purpose of the study was to assess their pain sensitivity and how they respond to painful stimulation. They were required to provide informed consent and were financially compensated with 30PLN (~ 7USD) for completing the study. Fifty participants did not show up for the experimental session. They were compared with the 74 participants who did take part in the study in terms of BMI, education level, job situation, and sex, but no differences were found. Data from three participants were not assessed due to technical malfunctions (n = 2) and data file corruption (n = 1). Thus, the statistical analysis was performed on data set of 71 cases. The basic descriptive statistics of the tested participants are presented in Table | PMC10115883 |
Sample size | The sample size calculation was conducted using G*Power 3.1 software | PMC10115883 | ||
Stimuli and measures | NRS, Pain, pain | When carrying out an experiment focused on color, significant elements are the participants’ color perception and the screen which displays the colors. The former was assessed via the Ishihara test during the screening phaseTo measure the effect of colors on pain perception, we used 6 different color hues: red (RGB (255, 0, 0)), green (RGB (0, 255, 0)), orange (RGB (255, 128, 0)), blue (RGB (0, 128, 255)), yellow (RGB (255, 255, 0)), and pink (RGB (255, 0, 128)). The colors were selected in such a way to examine primary and complementary colors. The saturation level was set at 100%, because the hues are most distinct at this levelEach group received pain stimuli of one modality: either electrical or thermal. Pain stimuli were delivered to the volar side of the non-dominant forearm within C5 dermatome (participants self-reported handedness). Electrical stimuli (Digitimer DS8R; Digitimer; Welwyn, Garden City, England) were square pulses applied in a sequence of three pulses (200 μs) with an ISI of 100 μs. Thermal stimuli were applied using a contact thermode, sized 30 × 30 mm (Model ATS and TSA-II; Medoc Ltd Advanced Medical System; Israel), starting from a 32 °C baseline temperature with a 10 °C/s ramp-up/ramp-down ratio and 1 s plateau. To achieve the best possible similarity in pain sensation evoked with thermal and electrical stimulation, the physical characteristics of the stimuli (i.e. duration) were adjusted and calibration procedure was applied.Participants rated pain intensity (“How painful was the stimulation?”) and pain expectation (“How painful do you expect the stimulation to be after seeing this color?”) separately on an 11-point Numeric Rating Scale (NRS), from 0 = ‘no pain’ to 10 = ‘the most intense tolerable pain’. The NRSs were always presented on a color slide corresponding to the color used in the current trial.The procedure was programmed using Python 3 language and PsychoPy 2021 software | PMC10115883 | |
Experimental design | The experiment was based on a between-subject, repeated-measures design with between-factor modality (electrical, thermal) and within-factor color (black, blue, green, grey, orange, pink, red, white, yellow). Eligible participants were invited to an experimental session and randomly assigned to either the electrical or the thermal group. The onsite session comprised two parts: calibration and the main task. | PMC10115883 | ||
Calibration | pain | Tactile threshold, pain threshold, and pain intensity were assessed during calibration. Participants in the electrical group received stimuli in steps of 1 mA per 5 s, starting from 0 mA. In the thermal group, stimuli were delivered in steps of 0.5 °C per 5 s, the baseline temperature was set at 32 °C and the first applied stimulus was 38 °C. The 5 s interval reduces the risk of occurrence of temporal summation, since stimuli are delivered at frequency of 0.2Hz | PMC10115883 | |
Main task | pain | The main task consisted of ten blocks: two pain baseline blocks, two expectation color blocks, and six pain color blocks (see the flow of the task in Fig. Study design. The study involved two groups that differed in the modality of pain applied: thermal or electrical. Part (After finishing the main task, participants were asked to fill out a questionnaire to assess whether they had figured out the aim of the study and to collect their beliefs about the influence of colors on pain perception (questions in Table End of study questionnaire. | PMC10115883 | |
Statistical analysis | ’, handedness, pain | REGRESSION | First, descriptive statistics were calculated for age and body mass index (BMI), as well as the distribution of sex, education, job situation, and handedness. This was followed by analyses of the differences between the groups (electrical and thermal), calculated using the Student’s Then, the variables of interest were checked for outliers and prepared before the conduct of further statistical analyses: the pain ratings for each color and the baseline were aggregated into means. After, the first and the second pain baseline blocks were compared through the repeated-measures ANOVA, with ‘modality’ (electrical, thermal) as the between-subjects factor and ‘baseline’ (first, second) as the within-subjects factor. The baselines did not differ significantly, therefore they were combined into one. The baseline pain ratings were subtracted from the pain color ratings. Similarly, within both pain expectation blocks, the expectation baseline rating was subtracted from the expectation rating of every other color. Therefore, data analyses for pain and expectation were performed on the differences between the baseline (black) and the other colors.In the primary analyses on pain intensity, a two-way, mixed-design ANOVA was used with ‘modality’ (electrical, thermal) as the between-subjects factor and ‘color’ (differences between the baseline and each of the following: blue, green, grey, orange, pink, red, white, and yellow) as the within-subjects factor. Three additional ANOVAs of the same structure were conducted with respective additional factors: ‘Q1’ (yes, no), ‘Q2’ (yes, no), or ‘Q3’ (yes, no), in order to verify whether awareness of the study aim (Q1) and beliefs about the influence of colors on pain (Q2 and Q3) influenced the results.Also, a three-way, mixed-design ANOVA was performed on participants’ NRS expectation ratings. The ‘modality’ (electrical, thermal) was the between-subjects factor, while ‘color’ (differences between the black and each of the following: blue, green, grey, orange, pink, red, white and yellow) and ‘block ‘(first expectation block, second expectation block) served as the within-subject factors.The post-hoc tests were performed for the pain and expectation ratings to explore between-colors differences. Correlation and regression analyses were performed to explore the relationship between pain expectation related to colors at the beginning of the experiment and further pain intensity ratings.The alpha level was set at 0.05 for rejection of the null hypothesis. The Bonferroni correction was implemented for all analyses. The analyses were conducted using the IBM SPSS Statistics environment, version 26.0 (IBM Corp. 2019). | PMC10115883 |
Supplementary Information | The online version contains supplementary material available at 10.1038/s41598-023-33313-w. | PMC10115883 | ||
Author contributions | All authors discussed the results and commented on the manuscript. K.W.K. conceptualized and designed the study, collected and analyzed the data, and wrote the manuscript; J.B. designed the study, analyzed the data and wrote the manuscript; H.B. designed the study and wrote the manuscript; P.B. designed the study, supervised the study conduction and wrote the manuscript. | PMC10115883 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.