Explanatory Interactive Machine Learning: Establishing an Action Design Research Process for Machine Learning Projects

The most promising standard machine learning methods can deliver highly accurate classification results, often outperforming standard white-box methods. However, it is hardly possible for humans to fully understand the rationale behind the black-box results, and thus, these powerful methods hamper the creation of new knowledge on the part of humans and the broader acceptance of this technology. Explainable Artificial Intelligence attempts to overcome this problem by making the results more interpretable, while Interactive Machine Learning integrates humans into the process of insight discovery. The paper builds on recent successes in combining these two cutting-edge technologies and proposes how Explanatory Interactive Machine Learning (XIL) is embedded in a generalizable Action Design Research (ADR) process – called XIL-ADR. This approach can be used to analyze data, inspect models, and iteratively improve them. The paper shows the application of this process using the diagnosis of viral pneumonia, e.g., Covid-19, as an illustrative example. By these means, the paper also illustrates how XIL-ADR can help identify shortcomings of standard machine learning projects, gain new insights on the part of the human user, and thereby can help to unlock the full potential of AI-based systems for organizations and research. © 2023, The Author(s).

Changes in health-related quality of life in parents of adolescents, and the impact of gender, sociodemographic and psychological factors, and pain on the changes: A two-year longitudinal study

Aims: For many adults their role as a parent is a vital part of their lives. This role may influence their HRQOL and vary with the age of their child. The aim of the present study was to describe and compare sociodemographic and psychological factors, pain and HRQOL in parents of adolescents assessed at baseline and 2 years later, and to quantify possible impact of gender, sociodemographic and psychological factors and pain on changes in HRQOL over time. Method(s): We used data collected at T1 (baseline), when the adolescents were 14-15 years (November 2018-April 2019) and T2 (January-February 2021), when the adolescents were 16-17 years, and the COVID-19 pandemic ongoing. Data on sociodemographic, self-efficacy, self-esteem, pain, loneliness and stress were collected. HRQOL was assessed using RAND-36. Data were analyzed using McNemar tests, paired samples t-tests, and multiple linear regression analyses. Result(s): Among the 309 parents from the general Norwegian population with valid responses at both T1 and T2, 262 (82%) were mothers and 57 (18%) fathers. At T1 the mean age was 45 (SD = 5) years, 81% were married/cohabiting, 75% worked full-time and 57% had university education>4 years. From T1 to T2 the average pain score increased, 1.6 (1.8) vs 1.8 (1.9), p = 0.019, the pain interference emotion score increased (1.6 (1.9) vs. 1.8 (2.2), p = 0.007, a large proportion reported pain duration>3 months, 44% vs 50%, p = 0.014, the parents were more lonely, 12.8 (4.2) vs. 13.7 (4.3), p<0.001, and reported lower RAND-36 mental component summary (MCS) score, 52.2 (8.2) vs. 50.9 (9.7), p = 0.008. There were no significant changes in physical component summary (PCS) score. A positive change in MCS was associated with working part time (B = 5.22 (95% CI [1.05, 9.38)]) and full time (B = 3.64 (95% CI [-0.21, 7.48])) (ref no paid work) and older age (B = 0.24, (95% CI [-001, 0.42]), a negative change by stress (B = -17.39, (95% CI [-27.42, -7.51]). Conclusion(s): Over a 2-year period, and 1 year into the COVID-19 pandemic, parents of adolescents from the general population reported more pain and pain interference, were lonelier and experienced decreased mental HRQOL.

Studies on COVID-19 lethality: Causes and dynamics at German University Hospitals

Background: In the initial phase of the COVID-19 pandemic, a lower incidence and death rate was observed in Germany compared to its neighbouring countries, but some studies showed comparatively high death rates in ventilated COVID-19 patients. Methods: In this retrospective analysis, hospital stays of COVID-19 patients at 14 German university hospitals were analysed. For this purpose, local data integration centres of the German Medical Informatics Initiative (MII) combined their data to present death rates in different subgroups depending on gender, age, length of stay in the intensive care unit, ventilation and in combination with different comorbidities. Results: The total lethality rate in 1,318 COVID-19 patients was 18.8 %. In ventilated cases, the lethality rate was 38.8%. Common comorbidities were renal insufficiency (35.2 %), aplastic and other anaemia (26.0 %) diabetes mellitus (21.1 %). The average length of stay was 18 days, or 28 days in case of ventilated patients. Lethality decreased from 20.7 % to 12.7 % over the observation period. Conclusion: The observed decline in lethality rates may be explained with the continuous optimisation of COVID-19 treatment, increasing experience and improved therapy recommendations. The progress made so far by the MII allows cross-consortium analyses to be carried out just in time to better address the challenges of the COVID-19 pandemic.

Quality of life and fear of COVID-19 in 2600 baccalaureate nursing students at five universities: a cross-sectional study.

BACKGROUND: The COVID-19 pandemic has caused significant disruptions in the implementation of programs across educational institutions. Nursing students, being both young adults and by practical training, part of the health care system, may be particularly vulnerable during the COVID-19 pandemic. The purpose of this study was to explore the associations between self-reported fear of COVID-19, general health, psychological distress and overall quality of life (QoL) in a sample of Norwegian baccalaureate nursing students compared to reference data. METHODS: The survey targeted baccalaureate nursing students from five universities in February 2021. An electronic questionnaire consisted of the Fear of COVID-19 Scale (FCV-19S), the Hopkins Symptom Checklist 5 (SCL-5), one general health and one overall QoL question. The respondents' mean scores were compared to reference data. Hierarchical regression analyses were conducted, and effect sizes (Cohen's d) were evaluated. RESULTS: In total, 2605 out of 6088 (43%) students responded. Their FCV-19S scores (mean 2.45, CI 2.42, 2.48) were significantly higher than those of the reference population (mean 1.8, P < 0.001). Nursing students scores showed significantly lower general health (mean 3.50 ± 0.93 SD, population mean = 3.57, Cohen's d = 0.07), higher levels of psychological distress (mean 2.68 ± 1.03 SD, population mean = 2.12, Cohen's d = 0.55) and lower overall QoL (mean 5.50 ± 2.16 SD, population mean = 8.00, Cohen's d = 1.16) compared to pre-pandemic reference data. FCV-19S scores were significantly associated with levels of general health (Cohen's d = 0.26), psychological distress (Cohen's d = 0.76) and overall QoL (Cohen's d = 0.18). CONCLUSIONS: Baccalaureate nursing students reported worse outcomes during the Covid-19 pandemic on general health, psychological distress and overall QoL compared to the reference population. Level of fear of Covid-19, however, accounted for few of these differences. Other factors related to the pandemic may have reduced nursing students' overall QoL.

Use of r packages “revtools” and “litsearchr” for automated keyword extraction to support building systematic medline searches in covid-19

Rationale: COVID-19 has triggered significant research activities worldwide, leading to an immense number of scientific publications in the MEDLINE. Given the high publication volume, proper combinations of structured subject headings (MeSH) and keywords are required to narrow down the search within the COVID-19 literature to the most relevant subtopic. Manual keyword selection is time-consuming and may not always be feasible. The recent text mining algorithms permit automated extraction of keywords from publications identified during pilot reference searches. It is not clear, though, whether automated keyword extraction would be useful if carried out on the pilot reference sample without prior enrichment for the COVID-19 subtopic of interest. This was addressed in the present study. Methods: A non-comprehensive MEDLINE search on the subtopic “Digital telemedicine in COVID-19” was conducted to obtain a pilot reference sample. Without manual enrichment for pertinent publications, keywords from this reference sample were extracted using two R packages, “revtools” (utilizes topic models and weighed keyword ranking) and “litsearchr” (utilizes keyword co-occurrence networks). In parallel, a manual systematic MEDLINE search strategy (MeSH concepts and headings, and manually-selected keywords, the total of 75 terms) was designed on the above topic as per PICO criteria and expert discussions. The automatically extracted keywords were then compared with the terms in manual MEDLINE search strategy. Results: The “revtools” package extracted 150 keywords from the “non-enriched” pilot reference sample. Of those, 12 (8%) keywords (either individual words or phrases) overlapped with the terms already present in the manual MEDLINE search strategy. This extraction also yielded 3 unique keywords useful to augment the manual search strategy. The “litsearchr” package extracted 203 keywords. Of those, 4 (1.97%) overlapped with the terms in the manual MEDLINE search strategy. In addition, 4 unique phrases extracted by this package were found useful for the manual search strategy. The automatically extracted and useful keywords (respectively, 3 and 4) were unique for each package. Conclusions: Automated keyword extraction, despite parallel utilization of different algorithms, cannot yet fully replace an expert-built manual MEDLINE search strategy on COVID-19. Yet this extraction, even if conducted on a non-comprehensive and non-enriched reference sample, can augment the manual search. Automated extraction from the reference sample enriched for the COVID-19 subtopic of interest may further increase the yield of useful keywords.