Moral Engagement and Disengagement in Health Care AI Development

BACKGROUND

Machine learning (ML) is utilized increasingly in health care, and can pose harms to patients, clinicians, health systems, and the public. In response, regulators have proposed an approach that would shift more responsibility to ML developers for mitigating potential harms. To be effective, this approach requires ML developers to recognize, accept, and act on responsibility for mitigating harms. However, little is known regarding the perspectives of developers themselves regarding their obligations to mitigate harms.

METHODS

We conducted 40 semi-structured interviews with developers of ML predictive analytics applications for health care in the United States.

RESULTS

Participants varied widely in their perspectives on personal responsibility and included examples of both moral engagement and disengagement, albeit in a variety of forms. While most (70%) of participants made a statement indicative of moral engagement, most of these statements reflected an awareness of moral issues, while only a subset of these included additional elements of engagement such as recognizing responsibility, alignment with personal values, addressing conflicts of interests, and opportunities for action. Further, we identified eight distinct categories of moral disengagement reflecting efforts to minimize potential harms or deflect personal responsibility for preventing or mitigating harms.

CONCLUSIONS

These findings suggest possible facilitators and barriers to the development of ethical ML that could act by encouraging moral engagement or discouraging moral disengagement. Regulatory approaches that depend on the ability of ML developers to recognize, accept, and act on responsibility for mitigating harms might have limited success without education and guidance for ML developers about the extent of their responsibilities and how to implement them.

Preclinical efficacy in therapeutic area guidelines from the U.S. Food and Drug Administration and the European Medicines Agency: a cross-sectional study

BACKGROUND AND PURPOSE

Therapeutic area guidelines (TAGs) published by the EMA and the FDA offer guidance in planning the launch of a trial in a certain indication. We assessed and compared the guidance on preclinical efficacy of all available TAGs from EMA and FDA.

EXPERIMENTAL APPROACH

EMA and FDA websites and databases were searched for all TAGs. A mixed deductive and inductive approach was applied to analyse and cluster content for preclinical efficacy.

KEY RESULTS

A total of 114 EMA and 120 FDA TAGs were identified, covering 126 indications. Our core finding is that 75% of EMA TAGs and 58% from the FDA TAGs do not offer any guidance on preclinical efficacy. TAGs varied widely on the extent, nature and detail of guidance.

CONCLUSIONS AND IMPLICATIONS

Guidance on preclinical efficacy in a consistent, comprehensive and explicit way that still allows for justified deviations is an important but neglected aspect of transparency for drug development. This transparency would help sponsors in designing preclinical studies and in negotiating more efficiently with regulators.

Preclinical efficacy studies in investigator brochures: Do they enable risk-benefit assessment?

Human protection policies require favorable risk–benefit judgments prior to launch of clinical trials. For phase I and II trials, evidence for such judgment often stems from preclinical efficacy studies (PCESs). We undertook a systematic investigation of application materials (investigator brochures [IBs]) presented for ethics review for phase I and II trials to assess the content and properties of PCESs contained in them. Using a sample of 109 IBs most recently approved at 3 institutional review boards based at German Medical Faculties between the years 2010–2016, we identified 708 unique PCESs. We then rated all identified PCESs for their reporting on study elements that help to address validity threats, whether they referenced published reports, and the direction of their results. Altogether, the 109 IBs reported on 708 PCESs. Less than 5% of all PCESs described elements essential for reducing validity threats such as randomization, sample size calculation, and blinded outcome assessment. For most PCESs (89%), no reference to a published report was provided. Only 6% of all PCESs reported an outcome demonstrating no effect. For the majority of IBs (82%), all PCESs were described as reporting positive findings. Our results show that most IBs for phase I/II studies did not allow evaluators to systematically appraise the strength of the supporting preclinical findings. The very rare reporting of PCESs that demonstrated no effect raises concerns about potential design or reporting biases. Poor PCES design and reporting thwart risk–benefit evaluation during ethical review of phase I/II studies.

To make a clinical trial ethical, regulatory agencies and institutional review boards have to judge whether the trial-related benefits (the knowledge gain) outweigh the trial-inherent risks. For early-phase human research, these risk–benefit assessments are often based on evidence from preclinical animal studies reported in so-called “investigator brochures.” However, our analysis shows that the vast majority of such investigator brochures lack sufficient information to systematically appraise the strength of the supporting preclinical findings. Furthermore, the very rare reporting of preclinical efficacy studies that demonstrated no effect raises concerns about potential design and/or reporting biases. The poor preclinical study design and reporting thwarts risk–benefit evaluation during ethical review of early human research. Regulators should develop standards for the design and reporting of preclinical efficacy studies in order to support the conduct of ethical clinical trials.

Cometary science. The organic-rich surface of comet 67P/Churyumov-Gerasimenko as seen by VIRTIS/Rosetta

The VIRTIS (Visible, Infrared and Thermal Imaging Spectrometer) instrument on board the Rosetta spacecraft has provided evidence of carbon-bearing compounds on the nucleus of the comet 67P/Churyumov-Gerasimenko. The very low reflectance of the nucleus (normal albedo of 0.060 ± 0.003 at 0.55 micrometers), the spectral slopes in visible and infrared ranges (5 to 25 and 1.5 to 5% kÅ(-1)), and the broad absorption feature in the 2.9-to-3.6-micrometer range present across the entire illuminated surface are compatible with opaque minerals associated with nonvolatile organic macromolecular materials: a complex mixture of various types of carbon-hydrogen and/or oxygen-hydrogen chemical groups, with little contribution of nitrogen-hydrogen groups. In active areas, the changes in spectral slope and absorption feature width may suggest small amounts of water-ice. However, no ice-rich patches are observed, indicating a generally dehydrated nature for the surface currently illuminated by the Sun.

More education, less administration: reflections of neuroimagers' attitudes to ethics through the qualitative looking glass

In follow-up to a large-scale ethics survey of neuroscientists whose research involves neuroimaging, brain stimulation and imaging genetics, we conducted focus groups and interviews to explore their sense of responsibility about integrating ethics into neuroimaging and readiness to adopt new ethics strategies as part of their research. Safety, trust and virtue were key motivators for incorporating ethics into neuroimaging research. Managing incidental findings emerged as a predominant daily challenge for faculty, while student reports focused on the malleability of neuroimaging data and scientific integrity. The most frequently cited barrier was time and administrative burden associated with the ethics review process. Lack of scholarly training in ethics also emerged as a major barrier. Participants constructively offered remedies to these challenges: development and dissemination of best practices and standardized ethics review for minimally invasive neuroimaging protocols. Students in particular, urged changes to curricula to include early, focused training in ethics.

[Comparative compositional mapping of chicken and quail chromosomes]

The distribution of various isochore families on mitotic chromosomes of domestic chicken and Japanese quail was studied by the method of fluorescence in situ DNA--DNA hybridization (FISH). DNA of various isochore families was shown to be distributed irregularly and similarly on chromosomes of domestic chicken and Japanese quail. The GC-rich isochore families (H2, H3, and H4) hybridized mainly to microchromosomes and a majority of macrochromosome telomeric regions. In chicken, an intense fluorescence was also in a structural heterochromatin region of the Z chromosome long arm. In some regions of the quail macrochromosome arms, hybridization was also with isochore families H3 and H4. On macrochromosomes of both species, the pattern of hybridization with isochores of the H2 and H3 families resembled R-banding. The light isochores (L1 and L2 families) are mostly detected within macrochromosome internal regions corresponding to G bands, whereas microchromosomes lack light isochores. Although mammalian and avian karyotypes differ significantly in organization, the isochore distribution in genomes of these two lineages of the warm-blooded animals is similar in principle. On macrochromosomes of the two avian species studied, a pattern of isochore distribution resembled that of mammalian chromosomes. The main specific feature of the avian genome, a great number of microchromosomes (about 30% of the genome), determines a compositional specialization of the latter. This suggests the existence of not only structural but also functional compartmentalization of the avian genome.

Compositional mapping of chicken chromosomes and identification of the gene-richest regions

'Compositional chromosomal mapping', namely the assessment of the GC level of chromosomal bands, led to the identification, in the human chromosomes, of the GC-richest H3+ bands and of the GC-poorest L1+ bands, which were so called on the basis of the isochore family predominantly present in the bands. The isochore organization of the avian genome is very similar to those of most mammals, the only difference being the presence of an additional, GC-richest, H4 isochore family. In contrast, the avian karyotypes are very different from those of mammals, being characterized, in most species, by few macrochromosomes and by a large number of microchromosomes. The 'compositional mapping' of chicken mitotic and meiotic chromosomes by in-situ hybridization of isochore families showed that the chicken GC-richest isochores are localized not only on a large number of microchromosomes but also on almost all telomeric bands of macrochromosomes. On the other hand, the GC-poorest isochores are generally localized on the internal regions of macrochromosomes and are almost absent in microchromosomes. Thus, the distinct localization of the GC-richest and the GC-poorest bands observed on human chromosomes appears to be a general feature of chromosomes from warm-blooded vertebrates.

Genes, isochores and bands in human chromosomes 21 and 22

The recently available DNA sequences from chromosomes 21 and 22 enabled us to define the relationships of different band types with isochores and with gene concentration and to compare these relationships with previous results. We showed that chromosomal bands appear as Giemsa or Reverse bands depending not on their absolute GC level, but on the composition GC level relative to those of adjacent contiguous bands. We also demonstrated that the GC-richest, and gene-richest H3+ bands are characterized by a lower DNA compaction compared with the GC-poorest, gene-poorest L1+ bands. Moreover, our results indicate that the human genome contains about 30,000 genes.

Purification of the aldehyde oxidase homolog 1 (AOH1) protein and cloning of the AOH1 and aldehyde oxidase homolog 2 (AOH2) genes. Identification of a novel molybdo-flavoprotein gene cluster on mouse chromosome 1

We report the cloning of the AOH1 and AOH2 genes, which encode two novel mammalian molybdo-flavoproteins. We have purified the AOH1 protein to homogeneity in its catalytically active form from mouse liver. Twenty tryptic peptides, identified or directly sequenced by mass spectrometry, confirm the primary structure of the polypeptide deduced from the AOH1 gene. The enzyme contains one molecule of FAD, one atom of molybdenum, and four atoms of iron per subunit and shows spectroscopic features similar to those of the prototypic molybdo-flavoprotein xanthine oxidoreductase. The AOH1 and AOH2 genes are 98 and 60 kilobases long, respectively, and consist of 35 coding exons. The AOH1 gene has the potential to transcribe an extra leader non-coding exon, which is located downstream of exon 26, and is transcribed in the opposite orientation relative to all the other exons. AOH1 and AOH2 map to chromosome 1 in close proximity to each other and to the aldehyde oxidase gene, forming a molybdo-flavoenzyme gene cluster. Conservation in the position of exon/intron junctions among the mouse AOH1, AOH2, aldehyde oxidase, and xanthine oxidoreductase loci indicates that these genes are derived from the duplication of an ancestral precursor.

Gene density in the Giemsa bands of human chromosomes

The human genome is formed by isochores belonging to five families, L1, L2, H1, H2 and H3, that are characterized by increasing GC levels and gene concentrations. In-situ hybridization of DNA from different isochore families provides, therefore, information not only on the correlation between isochores and chromosomal bands, but also on the distribution of genes in chromosomes. Three subsets of R(everse) bands were identified: H3+, H3* and H3-, that contain large, moderate, and no detectable amounts, respectively, of the gene-richest H2 and H3 isochores, and replicate very early and early, respectively, in S phase of the cell cycle. Here, we investigated the GC levels, replication timings and DNA compaction of G(iemsa) bands. We showed that G bands comprise two different subsets of bands, one of which is predominantly composed of L1 isochores, replicates at the end of the S phase, has a higher DNA compaction relative to H3+ bands and corresponds to the darkest G bands of Francke (1994). In contrast, the other subset is composed of L2 and H1 isochores, has less-extreme properties in replication and composition and corresponds to the less-dark G bands of Francke.

The gene-richest bands of human chromosomes replicate at the onset of the S-phase

Previous investigations on the correlations between isochore organization and human chromosomal bands have identified three sets of R(everse) bands: H3+, H3* and H3-, endowed with large, moderate, and no detectable amounts of the gene-richest H3 isochores, respectively. In the present work we compared the replication timing of these three sets of bands and showed that the chromosomal bands containing H3 isochores replicate almost entirely (in the case of H3+ bands) or largely (in the case of H3* bands) at the onset of S phase, whereas chromosomal bands not containing H3 isochores (H3- bands) replicate later. The existence, at a resolution of 400 bands per haploid genome, of at least three distinct subsets of R bands is, therefore, not only supported by their GC and gene concentration but also by their replication times.

Biochemical changes in prostanoids and cerebral expression of cyclooxygenase (COX)-1 and COX-2 during morphine sulfate infusion in the newborn piglet

To examine the biochemical regulation of morphine sulfate (MS) on prostanoid synthesis, conscious newborn piglets received a bolus dose of 100 microg/kg followed by a continuous infusion dose of 100 microg/kg/h. The control group received equivalent volume bolus and continuous infusion of 5% dextrose. Blood samples were drawn from the femoral artery and sagittal sinus vein before, during and after infusion for measurement of prostanoids. The expression of mRNAs encoding cyclooxygenases (COX)-1 and -2 in the brainstem, thalamus, cortex, and cerebellum of the newborn piglets were also examined. Systemic PGE2 levels declined substantially during and post MS infusion (p < 0.01), whereas sagittal sinus vein PGE2 levels increased following the bolus dose (p < 0.01) and at 4 h of continuous infusion (p < 0.01). MS infusion did not affect systemic 6-ketoPGF1alpha levels, however, in the cerebral circulation 6-ketoPGF1alpha levels increased 146% (p < 0.01) following the bolus dose and remained elevated throughout the infusion and post infusion times. Systemic TxB2 levels increased transiently at 4 h (p < 0.01) and sagittal sinus vein TxB2 increased at 0.5 and 1 h (p < 0.01) during continuous infusion. RT-PCR assays revealed a 1.5- (p < 0.001) to 4-fold (p < 0.001) increased expression of COX-1 mRNA in the MS-infused brain samples. In contrast, no differences in COX-2 mRNA were detected between the groups. These data imply that MS may have significant effects on prostanoid synthesis in the newborn. The data further show that the MS-induced prostanoid responses appear to be mediated via COX-1.

Identification of the gene-richest bands in human prometaphase chromosomes

The human genome is a mosaic of long, compositionally homogeneous DNA segments, the isochores, that can be partitioned into five families, two GC-poor families (L1 and L2), representing 63% of the genome, and three GC-rich families (H1, H2 and H3), representing 24%, 7.5% and 4-5% of the genome, respectively. Gene concentration increases with increasing GC levels, reaching a level 20-fold higher in H3 compared with L isochores. In-situ hybridization of DNA from different isochore families provides, therefore, information on the chromosomal distribution of genes. Using this approach, three subsets of reverse or Giemsa-negative bands, H3+, H3* and H3-, containing large, moderate, and no detectable amounts, respectively, of the gene-richest H3 isochores were identified at a resolution of 400 bands. H3+ bands largely coincide with the most heat-denaturation-resistant bands, the chromomycin-A3-positive, DAPI-negative bands, the bands with the highest CpG island concentrations, and the earliest replicating bands. Here, we have defined the H3+ bands at a 850-band resolution, and have thus identified the human genome regions, having an average size of 4 Mb, that are endowed with the highest gene density.

cDNA characterization and chromosome mapping of the human GAS2 gene

Murine Gas2 is a microfilament-associated protein whose expression is increased at growth arrest in mammalian cells. During apoptosis, Gas2 is specifically cleaved at its C-terminus by a still unknown ICE-like protease, and the processed protein induces dramatic rearrangements in the cytoskeleton when overexpressed in several cell types. Here we report the characterization of a cDNA encoding the human homologue of Gas2, showing high conservation with the murine counterpart at the protein level. Fluorescence in situ hybridization analysis and radiation hybrid mapping localized the GAS2 gene on human chromosome 11p14.3-p15.2, in a region homologous to the gas2 region on mouse chromosome 7.

Magnetic resonance angiographic evaluation of circulus arteriosus cerebri (circle of Willis): a morphologic study in 100 human healthy subjects

The Circle of Willis was studied by Magnetic Resonance Angiography in 100 healthy subjects. In 41% of these cases, the arteries were arranged in the classically described way. In 21%, hypoplasia of the posterior communicating as. was noted, while in 13% the posterior cerebral as were found to originate from the internal carotid a.. In 9% of these cases, three anterior cerebral as. were present. In 3% the anterior communicating artery could not be identified, while the left posterior communicating artery was hypoplastic. In 2% the absence of a posterior communicating artery was associated with the origin of a posterior cerebral a. from the internal carotid. In another 2% the anterior cerebral as. were partially fused, and in yet another 2%, hypoplasia of both an anterior and a posterior cerebral a. was present. The remaining seven cases (7%), all different from each other, represented combinations of the above described variations. Statistical analysis indicated that anomalies occurred more commonly on the left than on the right side. The morphology of the Circle of Willis could not be correlated with either sex or Body Index.

Double-blind controlled trial of flumazenil in patients who underwent upper gastrointestinal endoscopy

The antisedative effect of flumazenil, a benzodiazepine antagonist, was studied in a double-blind placebo controlled trial in 61 patients who underwent upper gastrointestinal endoscopy and sedation with benzodiazepines. The efficacy of flumazenil in reversing the effect of both benzodiazepines, diazepam and midazolam, was significantly higher than placebo (p less than 0.0001). The effect of flumazenil was prompt and was clearly noticed at the first assessment, 5 min after its administration. In none of the patients was a relapse of the sedative effect of the benzodiazepines noticed. The administration of flumazenil was free of major side effects. Flumazenil administration permits an earlier discharge of patients following endoscopy. Its availability in the endoscopy suite may improve the outcome of serious but rare side effects related to benzodiazepines.