Three Epochs of Artificial Intelligence in Health Care

Importance

Interest in artificial intelligence (AI) has reached an all-time high, and health care leaders across the ecosystem are faced with questions about where, when, and how to deploy AI and how to understand its risks, problems, and possibilities.

Observations

While AI as a concept has existed since the 1950s, all AI is not the same. Capabilities and risks of various kinds of AI differ markedly, and on examination 3 epochs of AI emerge. AI 1.0 includes symbolic AI, which attempts to encode human knowledge into computational rules, as well as probabilistic models. The era of AI 2.0 began with deep learning, in which models learn from examples labeled with ground truth. This era brought about many advances both in people's daily lives and in health care. Deep learning models are task-specific, meaning they do one thing at a time, and they primarily focus on classification and prediction. AI 3.0 is the era of foundation models and generative AI. Models in AI 3.0 have fundamentally new (and potentially transformative) capabilities, as well as new kinds of risks, such as hallucinations. These models can do many different kinds of tasks without being retrained on a new dataset. For example, a simple text instruction will change the model's behavior. Prompts such as "Write this note for a specialist consultant" and "Write this note for the patient's mother" will produce markedly different content.

Conclusions and Relevance

Foundation models and generative AI represent a major revolution in AI's capabilities, ffering tremendous potential to improve care. Health care leaders are making decisions about AI today. While any heuristic omits details and loses nuance, the framework of AI 1.0, 2.0, and 3.0 may be helpful to decision-makers because each epoch has fundamentally different capabilities and risks.

Baseline Levels of Circulating Inflammatory Biomarkers Stratify Patients with Vitiligo Who Significantly Repigment after Treatment with Ruxolitinib Cream

BACKGROUND

Efficacy of ruxolitinib cream, a topical Jak1/Jak2 inhibitor, was demonstrated in a phase 2 trial in patients with vitiligo.

OBJECTIVE

This study aimed to characterize circulating inflammatory biomarker profiles in patients who demonstrated ≥50% improvement in facial Vitiligo Area Scoring Index scores by week 24 (group 1) and those who did not (group 2).

DESIGN

This was a posthoc analysis of a multicenter, randomized, double-blind, vehicle-controlled, phase 2 study in which screening was conducted between June 7, 2017 and March 21, 2018.

POPULATION

Patients aged between 18 and 75 years with vitiligo, including depigmentation affecting ≥0.5% of body surface area on the face and ≥3% of body surface area on nonfacial areas, were eligible.

INTERVENTION

Patients applied 1.5% ruxolitinib cream to lesions once or twice daily for 52 weeks.

MAIN OUTCOMES AND MEASURES

Patients were grouped by achievement of ≥50% improvement in facial Vitiligo Area Scoring Index at week 24. Proteomic analysis was performed on baseline serum samples.

RESULTS

Mean ± standard error facial Vitiligo Area Scoring Index in group 1 (n = 30) versus group 2 (n = 27) improved by 79.9 ± 4.0% versus 1.1 ± 7.3% and 91.9 ± 1.5% versus 25.1 ± 13.4% at weeks 24 and 52, respectively. Broad proteomic analysis revealed 76 proteins (of 1,104 tested) that were differentially expressed between groups 1 and 2 at baseline (P < 0.05). Ten distinct proteins were upregulated in group 1; 64 were elevated in group 2.

CONCLUSION

This analysis identified potential differences between patients who achieved ≥50% improvement in facial Vitiligo Area Scoring Index at 24 weeks and those who did not that require deeper scientific interrogation and may be important in stratifying therapeutic benefit for patients with vitiligo.

TRIAL REGISTRATION

The original study was registered at ClinicalTrials.gov, NCT03099304.

A biomarker signature to predict complete response to itacitinib and corticosteroids in acute graft-versus-host disease

A broad proteomic analysis was conducted to identify and evaluate candidate biomarkers potentially predictive of response to treatment with an oral selective Janus kinase 1 (JAK1) inhibitor, itacitinib, in acute graft‐versus‐host disease (GVHD). Plasma samples from 25 participants (identification cohort; NCT02614612) were used to identify novel biomarkers that were tested in a validation cohort from a placebo‐controlled, randomised trial (n = 210; NCT03139604). The identification cohort received corticosteroids plus 200 or 300 mg itacitinib once daily. The validation cohort received corticosteroids plus 200 mg itacitinib once daily or placebo. A broad proteomic analysis was conducted using a proximity extension assay. Baseline and longitudinal comparisons were performed with unpaired t‐test and one‐way analysis of variance used to evaluate biomarker level changes. Seven candidate biomarkers were identified. Monocyte‐chemotactic protein (MCP)3, pro‐calcitonin/calcitonin (ProCALCA/CALCA), together with a previously identified prognostic acute GVHD biomarker, regenerating islet‐derived protein (REG)3A, stratified complete responders from non‐responders (participants with progressive disease) to itacitinib, but not placebo, potentially representing predictive biomarkers of itacitinib in acute GVHD. ProCALCA/CALCA, suppressor of tumorigenicity (ST)2, and tumour necrosis factor receptor (TNFR)1 were significantly reduced over time by itacitinib in responders, potentially representing response‐to‐treatment biomarkers. Novel biomarkers have the potential to identify patients with acute GVHD that may respond to itacitinib plus corticosteroid treatment (NCT02614612; NCT03139604).

Patient and Clinician Perceptions of Factors Relevant to Ideal Specialty Consultations

IMPORTANCE

Inpatient subspecialty consultations, a common and expensive practice within inpatient medicine, do not always go well; however, little is known about the failure modes of consultation, thus making it difficult to identify interventions to improve consultation quality.

OBJECTIVE

To understand how stakeholders envision the ideal inpatient consultation and identify how and why consultations commonly fall short of this ideal.

DESIGN, SETTING, AND PARTICIPANTS

This qualitative study used in-depth, semistructured interviews collected from April to October 2017 and analyzed from January 2018 to February 2020 using conventional content analysis. The setting was a single academic medical center in Boston, Massachusetts. Participants were hospitalists and specialists who had requested or performed a consultation for a non-intensive care unit patient in the previous 4 months, patients who had received a consultation while hospitalized at the medical center in the previous 15 months, and family members of such patients.

MAIN OUTCOMES AND MEASURES

Consultation experiences reported by participants. Clinicians were asked about characteristics of the ideal consultation, positive and negative consultation experiences, costs and benefits, and suggested improvements. Patients and family members were asked about their consultation experience, changes in care, communication preferences, and suggested improvements.

RESULTS

The study included 38 participants: 17 specialists, 13 hospitalists, 4 patients, and 4 family members. More than half (21 of 38) of the participants were female. There were 11 key information exchanges identified that occur among the specialist team, primary team, and patient/family during an ideal consultation. These exchanges are time sensitive and primarily carried out through unwritten protocols. We also identified 6 defects (process failures) that commonly derail information exchanges (complete omission, exclusion of a key stakeholder, poor timing, incomplete or inaccurate information, and misinterpretation) and 5 contextual factors (roles and boundaries, professionalism, team hierarchy, availability, and operational know-how) that influence how information exchange unfolds, making some consultations more prone to defects.

CONCLUSIONS AND RELEVANCE

Successful inpatient consultation requires a complicated, sequenced series of time-sensitive information exchanges that are highly vulnerable to failure. Maximizing the benefit of consultations will likely entail not only minimizing low-value consultations but also actively preventing defects, such as information inaccuracies and misinterpretation, that commonly derail the consultation process.

Janus kinase 1 inhibitor INCB054707 for patients with moderate-to-severe hidradenitis suppurativa: results from two phase 2 studies

Background

Janus kinase (JAK)‐mediated cytokine signalling contributes to local and systemic inflammation in hidradenitis suppurativa (HS).

Objectives

To describe the safety and efficacy results from two multicentre phase II trials of the JAK1 inhibitor INCB054707 in patients with moderate‐to‐severe HS.

Methods

Patients received open‐label INCB054707 15 mg once daily (QD; Study 1) or were randomized to INCB054707 30, 60 or 90 mg QD or placebo (3 : 1 within each cohort; Study 2) for 8 weeks. Eligible patients were aged 18–75 years and had moderate‐to‐severe HS (Hurley stage II/III disease), lesions present in at least two anatomical locations, and a total abscess and inflammatory nodule count ≥ 3. The primary endpoint for both studies was safety and tolerability. Secondary endpoints included HS Clinical Response (HiSCR) and other efficacy measures.

Results

Ten patients were enrolled in Study 1 (15 mg INCB054707) and 35 in Study 2 (INCB054707: 30 mg, n = 9; 60 mg, n = 9; 90 mg, n = 8; placebo, n = 9). Overall, 70% of patients in Study 1 and 81% of patients receiving INCB054707 in Study 2 experienced at least one treatment‐emergent adverse event; 30% and 42% of patients, respectively, had at least one treatment‐related adverse event. Among the evaluable patients, three (43%) in Study 1 and 17 (65% overall: 30 mg, 56%; 60 mg, 56%; 90 mg, 88%) receiving INCB054707 vs. 4 patients (57%) receiving placebo in Study 2 achieved HiSCR at week 8.

Conclusions

INCB054707 was well tolerated, with responses observed in patients with moderate‐to‐severe HS. The safety and efficacy findings from these studies demonstrate proof of concept for JAK1 inhibition in HS. The studies are registered on ClinicalTrials.gov (NCT03569371 and NCT03607487).

Privacy-first health research with federated learning

Privacy protection is paramount in conducting health research. However, studies often rely on data stored in a centralized repository, where analysis is done with full access to the sensitive underlying content. Recent advances in federated learning enable building complex machine-learned models that are trained in a distributed fashion. These techniques facilitate the calculation of research study endpoints such that private data never leaves a given device or healthcare system. We show-on a diverse set of single and multi-site health studies-that federated models can achieve similar accuracy, precision, and generalizability, and lead to the same interpretation as standard centralized statistical models while achieving considerably stronger privacy protections and without significantly raising computational costs. This work is the first to apply modern and general federated learning methods that explicitly incorporate differential privacy to clinical and epidemiological research-across a spectrum of units of federation, model architectures, complexity of learning tasks and diseases. As a result, it enables health research participants to remain in control of their data and still contribute to advancing science-aspects that used to be at odds with each other.

Modeling Skin Inflammation Using Human In Vitro Models

Skin is made up of an epidermis and, dermis which serve as a barrier against physical and environmental threats. Keratinocytes make up greater than 95% of the epidermis and form different layers based on their level of differentiation. Millions of individuals suffer from skin diseases, which are characterized by significant barrier disruption and inflammation. Investigators previously relied on animal models to investigate inflammatory skin diseases; however, technological advances have enabled the use of physiologically human skin models to investigate the effects of inflammatory mediators on the structure and function of skin cells. In this article, we describe two protocols using keratinocytes to investigate tumor necrosis factor alpha (TNF-α) and interferon gamma (IFN-γ) driven skin inflammation as a surrogate for psoriasis, vitiligo, and other autoimmune skin diseases driven by these cytokines. © 2021 Wiley Periodicals LLC. Basic Protocol 1: Preparing a HaCaT keratinocyte culture Basic Protocol 2: 3-Dimensional organotypic skin cultures to assess TNF-α and IFN-γ driven skin inflammation.

Interplay between Humoral and CLA+ T Cell Response against Candida albicans in Psoriasis

Candida albicans (CA) infections have been associated with psoriasis onset or disease flares. However, the integrated immune response against this fungus is still poorly characterized in psoriasis. We studied specific immunoglobulins in plasma and the CA response in cocultures of circulating memory CD45RA^- cutaneous lymphocyte antigen (CLA)^+/- T cell with autologous epidermal cells from plaque and guttate psoriasis patients (cohort 1, n = 52), and also healthy individuals (n = 17). A complete proteomic profile was also evaluated in plaque psoriasis patients (cohort 2, n = 114) regarding their anti-CA IgA levels. Increased anti-CA IgA and IgG levels are present in the plasma from plaque but not guttate psoriasis compared to healthy controls. CA cellular response is confined to CLA⁺ T cells and is primarily Th17. The levels of anti-CA IgA are directly associated with CLA⁺ Th17 response in plaque psoriasis. Proteomic analysis revealed distinct profiles in psoriasis patients with high anti-CA IgA. C-C motif chemokine ligand 18, chitinase-3-like protein 1 and azurocidin were significantly elevated in the plasma from plaque psoriasis patients with high anti-CA levels and severe disease. Our results indicate a mechanism by which Candida albicans exposure can trigger a clinically relevant IL-17 response in psoriasis. Assessing anti-CA IgA levels may be useful in order to evaluate chronic psoriasis patients.

Ruxolitinib cream for treatment of vitiligo: a randomised, controlled, phase 2 trial

BACKGROUND

Vitiligo is a chronic autoimmune disease resulting in skin depigmentation and reduced quality of life. There is no approved treatment for vitiligo repigmentation and current off-label therapies have limited efficacy, emphasising the need for improved treatment options. We investigated the therapeutic potential of ruxolitinib cream in patients with vitiligo and report the efficacy and safety results up to 52 weeks of double-blind treatment.

METHODS

We did a multicentre, randomised, double-blind, phase 2 study for adult patients with vitiligo in 26 US hospitals and medical centres in 18 states. Patients with depigmentation of 0·5% or more of their facial body surface area (BSA) and 3% or more of their non-facial BSA were randomly assigned (1:1:1:1:1) by use of an interactive response technology system to receive ruxolitinib cream (1·5% twice daily, 1·5% once daily, 0·5% once daily, or 0·15% once daily) or vehicle (control group) twice daily on lesions constituting 20% or less of their total BSA for 24 weeks. Patients in the control group in addition to patients in the 0·15% once daily group who did not show a 25% or higher improvement from baseline in facial Vitiligo Area Scoring Index (F-VASI) at week 24 were re-randomised to one of three higher ruxolitinib cream doses (0·5% once daily, 1·5% once daily, 1·5% twice daily). Patients in the 0·5% once daily, 1·5% once daily, or 1·5% twice daily groups remained at their original dose up to week 52. Patients, investigators, and the study sponsor (except members of the interim analysis and primary endpoint analysis data monitoring teams) remained masked to treatment assignment throughout the study. The primary endpoint was the proportion of patients achieving a 50% or higher improvement from baseline in F-VASI (F-VASI50) at week 24, assessed in the intention-to-treat population. The study is registered with ClinicalTrials.gov, NCT03099304.

FINDINGS

Between June 7, 2017, and March 21, 2018, 205 patients were screened for eligibility, 48 were excluded and 157 patients (mean age, 48·3 years [SD 12·9]; 73 [46%] male and 84 [54%] female) were randomly assigned to either an intervention group or the control group. 32 (20%) of 157 were assigned to the control group, 31 (20%) to the 0·15% once daily group, 31 (20%) to the 0·5% once daily group, 30 (19%) to the 1·5% once daily group, and 33 (21%) to the 1·5% twice daily group. F-VASI50 at week 24 was reached by significantly more patients given ruxolitinib cream at 1·5% twice daily (15 [45%] of 33) and 1·5% once daily (15 [50%] of 30) than were treated with vehicle (one [3%] of 32). Four patients had serious treatment-emergent adverse events (one patient in the 1·5% twice daily group developed subdural haematoma; one patient in the 1·5% once daily group had a seizure; one patient in the 0·5% once daily group had coronary artery occlusion; and one patient in the 0·5% once daily group had oesophageal achalasia), all of which were unrelated to study treatment. Application site pruritus was the most common treatment-related adverse event among patients given ruxolitinib cream (one [3%] of 33 in the 1·5% twice daily group; three [10%] of 30 in the 1·5% once daily group; three [10%] of 31 in the 0·5% once daily group; and six [19%] of 31 in the 0·15% once daily group)with three [9%] of 32 patients showing application site pruritis in the control group. Acne was noted as a treatment-related adverse event in 13 (10%) of 125 patients who received ruxolitinib cream and one (3%) of 32 patients who received vehicle cream. All treatment-related adverse events were mild or moderate in severity and similar across treatment groups.

INTERPRETATION

Treatment with ruxolitinib cream was associated with substantial repigmentation of vitiligo lesions up to 52 weeks of treatment, and all doses were well tolerated. These data suggest that ruxolitinib cream might be an effective treatment option for patients with vitiligo.

FUNDING

Incyte.

Bundled Consent in US Intensive Care Units

BACKGROUND

Bundled consent, the practice of obtaining anticipatory consent for a predefined set of intensive care unit procedures, increases the rate of informed consent conversations and incorporation of patients' wishes into medical decision-making without sacrificing patients' or surrogates' understanding. However, the adoption rate for this practice in academic and nonacademic centers in the United States is unknown.

OBJECTIVE

To determine the national prevalence of use of bundled consent in adult intensive care units and opinions related to bundled consent.

METHODS

A random sample of US hospitals with medical/surgical intensive care units was selected from the AHA [American Hospital Association] Guide. One intensive care unit provider (bedside nurse, nurse manager, or physician) from each hospital was asked to self-reportuse of per-procedure consent versus bundled consent, consent rate for intensive care unit procedures, and opinions about bundled consent.

RESULTS

Of the 238 hospitals contacted, respondents from 100 (42%) completed the survey; 94% of respondents were nurses. The prevalence of bundled consent use was 15% (95% CI, 9%-24%). Respondents using per-procedure consent were more likely than those using bundled consent to self-report performing invasive procedures without consent. Users of bundled consent unanimously recommended the practice, and 49% of respondents using per-procedure consent reported interest in implementing bundled consent.

RESULTS

Bundled consent use is uncommon in academic and nonacademic intensive care units, most likely because of conflicting evidence about the effect on patients and surrogate decision makers. Future work is needed to determine if patients, family members, and providers prefer bundled consent over per-procedure consent.

Predicting Inpatient Medication Orders From Electronic Health Record Data

In a general inpatient population, we predicted patient‐specific medication orders based on structured information in the electronic health record (EHR). Data on over three million medication orders from an academic medical center were used to train two machine‐learning models: A deep learning sequence model and a logistic regression model. Both were compared with a baseline that ranked the most frequently ordered medications based on a patient’s discharge hospital service and amount of time since admission. Models were trained to predict from 990 possible medications at the time of order entry. Fifty‐five percent of medications ordered by physicians were ranked in the sequence model’s top‐10 predictions (logistic model: 49%) and 75% ranked in the top‐25 (logistic model: 69%). Ninety‐three percent of the sequence model’s top‐10 prediction sets contained at least one medication that physicians ordered within the next day. These findings demonstrate that medication orders can be predicted from information present in the EHR.

Customization scenarios for de-identification of clinical notes

BACKGROUND

Automated machine-learning systems are able to de-identify electronic medical records, including free-text clinical notes. Use of such systems would greatly boost the amount of data available to researchers, yet their deployment has been limited due to uncertainty about their performance when applied to new datasets.

OBJECTIVE

We present practical options for clinical note de-identification, assessing performance of machine learning systems ranging from off-the-shelf to fully customized.

METHODS

We implement a state-of-the-art machine learning de-identification system, training and testing on pairs of datasets that match the deployment scenarios. We use clinical notes from two i2b2 competition corpora, the Physionet Gold Standard corpus, and parts of the MIMIC-III dataset.

RESULTS

Fully customized systems remove 97-99% of personally identifying information. Performance of off-the-shelf systems varies by dataset, with performance mostly above 90%. Providing a small labeled dataset or large unlabeled dataset allows for fine-tuning that improves performance over off-the-shelf systems.

CONCLUSION

Health organizations should be aware of the levels of customization available when selecting a de-identification deployment solution, in order to choose the one that best matches their resources and target performance level.

Targeting the Janus Kinase Family in Autoimmune Skin Diseases

Autoimmune skin diseases are characterized by significant local and systemic inflammation that is largely mediated by the Janus kinase (JAK)-signal transducer and activator of transcription (STAT) pathway. Advanced understanding of this pathway has led to the development of targeted inhibitors of Janus kinases (JAKinibs). As a class, JAK inhibitors effectively treat a multitude of hematologic and inflammatory diseases. Growing evidence suggests that JAK inhibitors are efficacious in atopic dermatitis, alopecia areata, psoriasis, and vitiligo. Additional evidence suggests that JAK inhibition might be broadly useful in dermatology, with early reports of efficacy in several other conditions. JAK inhibitors can be administered orally or used topically and represent a promising new class of medications. Here we review the evolving data on the role of the JAK-STAT pathway in inflammatory dermatoses and the potential therapeutic benefit of JAK-STAT antagonism.

The Practice of Respect in the ICU

Although "respect" and "dignity" are intuitive concepts, little formal work has addressed their systematic application in the ICU setting. After convening a multidisciplinary group of relevant experts, we undertook a review of relevant literature and collaborative discussions focused on the practice of respect in the ICU. We report the output of this process, including a summary of current knowledge, a conceptual framework, and a research program for understanding and improving the practice of respect and dignity in the ICU. We separate our report into findings and proposals. Findings include the following: 1) dignity and respect are interrelated; 2) ICU patients and families are vulnerable to disrespect; 3) violations of respect and dignity appear to be common in the ICU and overlap substantially with dehumanization; 4) disrespect may be associated with both primary and secondary harms; and 5) systemic barriers complicate understanding and the reliable practice of respect in the ICU. Proposals include: 1) initiating and/or expanding a field of research on the practice of respect in the ICU; 2) treating "failures of respect" as analogous to patient safety events and using existing quality and safety mechanisms for improvement; and 3) identifying both benefits and potential unintended consequences of efforts to improve the practice of respect. Respect and dignity are important considerations in the ICU, even as substantial additional research remains to be done.