Electronic Health Record Usability, Satisfaction, and Burnout for Family Physicians.

Importance: Electronic health record (EHR) work has been associated with decreased physician well-being. Understanding the association between EHR usability and physician satisfaction and burnout, and whether team and technology strategies moderate this association, is critical to informing efforts to address EHR-associated physician burnout. Objectives: To measure family physician satisfaction with their EHR and EHR usability across functions and evaluate the association of EHR usability with satisfaction and burnout, as well as the moderating association of 4 team and technology EHR efficiency strategies. Design, Setting, and Participants: This study uses data from a cross-sectional survey conducted from December 12, 2021, to October 17, 2022, of all family physicians seeking American Board of Family Medicine recertification in 2022. Exposure: Physicians perceived EHR usability across 6 domains, as well as adoption of 4 EHR efficiency strategies: scribes, support from other staff, templated text, and voice recognition or transcription. Main Outcomes and Measures: Physician EHR satisfaction and frequency of experiencing burnout measured with a single survey item ("I feel burned out from my work"), with answers ranging from "never" to "every day." Results: Of the 2067 physicians (1246 [60.3%] younger than 50 years; 1051 men [50.9%]; and 1729 [86.0%] practicing in an urban area) who responded to the survey, 562 (27.2%) were very satisfied and 775 (37.5%) were somewhat satisfied, while 346 (16.7%) were somewhat dissatisfied and 198 (9.6%) were very dissatisfied with their EHR. Readability of information had the highest usability, with 543 physicians (26.3%) rating it as excellent, while usefulness of alerts had the lowest usability, with 262 physicians (12.7%) rating it as excellent. In multivariable models, good or excellent usability for entering data (ß = 0.09 [95% CI, 0.05-0.14]; P < .001), alignment with workflow processes (ß = 0.11 [95% CI, 0.06-0.16]; P < .001), ease of finding information (ß = 0.14 [95% CI, 0.09-0.19]; P < .001), and usefulness of alerts (ß = 0.11 [95% CI, 0.06-0.16]; P < .001) were associated with physicians being very satisfied with their EHR. In addition, being very satisfied with the EHR was associated with reduced frequency of burnout (ß = -0.64 [95% CI, -1.06 to -0.22]; P < .001). In moderation analysis, only physicians with highly usable EHRs saw improvements in satisfaction from adopting efficiency strategies. Conclusions and Relevance: In this survey study of physician EHR usability and satisfaction, approximately one-fourth of family physicians reported being very satisfied with their EHR, while another one-fourth reported being somewhat or very dissatisfied, a concerning finding amplified by the inverse association between EHR satisfaction and burnout. Electronic health record-based alerts had the lowest reported usability, suggesting EHR vendors should focus their efforts on improving alerts. Electronic health record efficiency strategies were broadly adopted, but only physicians with highly usable EHRs realized gains in EHR satisfaction from using these strategies, suggesting that EHR burden-reduction interventions are likely to have heterogenous associations across physicians with different EHRs.

Estimating immunization coverage at the district level: A case study of measles and diphtheria-pertussis-tetanus-Hib-HepB vaccines in Ethiopia.

Ethiopia has made significant progress in the last two decades in improving the availability and coverage of essential maternal and child health services including childhood immunizations. As Ethiopia keeps momentum towards achieving national immunization goals, methods must be developed to analyze routinely collected health facility data and generate localized coverage estimates. This study leverages the District Health Information Software (DHIS2) platform to estimate immunization coverage for the first dose of measles vaccine (MCV1) and the third dose of diphtheria-pertussis-tetanus-Hib-HepB vaccine (Penta3) across Ethiopian districts ("woredas"). Monthly reported numbers of administered MCV1 and Penta3 immunizations were extracted from public facilities from DHIS2 for 2017/2018-2021/2022 and corrected for quality based on completeness and consistency across time and districts. We then utilized three sources for the target population (infants) to compute administrative coverage estimates: Central Statistical Agency, DHIS2, and WorldPop. The Ethiopian Demographic and Health Surveys were used as benchmarks to which administrative estimates were adjusted at the regional level. Administrative vaccine coverage was estimated for all woredas, and, after adjustments, was bounded within 0-100%. In regions with the highest immunization coverage, MCV1 coverage would range from 83 to 100% and Penta3 coverage from 88 to 100% (Addis Ababa, 2021/2022); MCV1 from 8 to 100% and Penta3 from 4 to 100% (Tigray, 2019/2020). Nationally, the Gini index for MCV1 was 0.37, from 0.13 (Harari) to 0.37 (Somali); for Penta3, it was 0.36, from 0.16 (Harari) to 0.36 (Somali). The use of routine health information systems, such as DHIS2, combined with household surveys permits the generation of local health services coverage estimates. This enables the design of tailored health policies with the capacity to measure progress towards achieving national targets, especially in terms of inequality reductions.

Impact of response bias in three surveys on primary care providers' experiences with electronic health records.

OBJECTIVE: To identify impacts of different survey methodologies assessing primary care physicians' (PCPs') experiences with electronic health records (EHRs), we compared three surveys: the 2022 Continuous Certification Questionnaire (CCQ) from the American Board of Family Medicine, the 2022 University of California San Francisco (UCSF) Physician Health IT Survey, and the 2021 National Electronic Health Records Survey (NEHRS). MATERIALS AND METHODS: We evaluated differences between survey pairs using Rao-Scott corrected chi-square tests, which account for weighting. RESULTS: CCQ received 3991 responses from PCPs (100% response rate), UCSF received 1375 (3.6% response rate), and NEHRS received 858 (18.2% response rate). Substantial, statistically significant differences in demographics were detected across the surveys. CCQ respondents were younger and more likely to work in a health system; NEHRS respondents were more likely to work in private practice; and UCSF respondents disproportionately practiced in larger academic settings. Many EHR experience indicators were similar between CCQ and NEHRS, but CCQ respondents reported higher documentation burden. DISCUSSION: The UCSF approach is unlikely to supply reliable data. Significant demographic differences between CCQ and NEHRS raise response bias concerns, and while there were similarities in some reported EHR experiences, there were important, significant differences. CONCLUSION: Federal EHR policy monitoring and maintenance require reliable data. This test of existing and alternative sources suggest that diversified data sources are necessary to understand physicians' experiences with EHRs and interoperability. Comprehensive surveys administered by specialty boards have the potential to contribute to these efforts, since they are likely to be free of response bias.

Primary Care Physicians' Satisfaction With Interoperable Health Information Technology.

Importance: Enabling widespread interoperability-the ability of health information technology systems to exchange information and to use that information without special effort-is a primary focus of public policy on health information technology. More information on clinicians' experience using that technology can serve as one measure of the impact of that policy. Objective: To assess primary care physician perspectives on the state of interoperability. Design, Setting, and Participants: A cross-sectional survey of family medicine physicians in the US was conducted from December 12, 2021, to October 12, 2022. A sample of family medicine physicians who completed the Continuous Certification Questionnaire (CCQ), a required part of the American Board of Family Medicine certification process, which has a 100% response rate, were invited to participate. Main Outcomes and Measures: Eighteen items on the CCQ assessed experience accessing and using various information from outside organizations, including medications, immunizations, and allergies. Results: A total of 2088 physicians (1053 women [50%]; age reported categorically as either ≥50 years or <50 years) completed the CCQ interoperability questions in 2022. Of these respondents, 35% practiced in hospital or health system-owned practices, while 27% practiced in independently owned practices. Eleven percent were very satisfied with their ability to electronically access all 10 types of information from outside organizations included on the questionnaire, and a mean of 70% were at least somewhat satisfied. A total of 23% of family medicine physicians reported information from outside organizations was very easy to use, and an additional 65% reported that information was somewhat easy to use. Only 8% reported that information from different electronic health record (EHR) developers' products was very easy to use compared with 38% who reported information from the same EHR developer's product was very easy to use. Conclusions and Relevance: This survey study of family medicine physicians found modest and uneven improvement in physicians' experience with interoperability. These findings suggest that substantial heterogeneity in satisfaction by information type, source of information, EHR, practice type, ownership, and patient population necessitates diverse policy and strategies to improve interoperability.

Influence of Prior SARS-CoV-2 Infection on COVID-19 Severity: Evidence from the National COVID Cohort Collaborative.

Background: A large share of SARS-CoV-2 infections now occur among previously infected individuals. In this study, we sought to determine whether prior infection modifies disease severity relative to no prior infection. Methods: We used data from first and second COVID-19 episodes in the National COVID Cohort Collaborative, a nationwide collection of de-identified electronic health records. We used nested logistic regressions of monthly cohorts weighted on the inverse probability of prior infection to assess risk of hospitalization, death, and increased severity in the first versus second infection cohorts. Results: We included a total of 2,058,274 individuals in the analysis, 147,592 of whom had two recorded infections. The impact of prior infection differed meaningfully between months. Prior infection was largely protective prior to March 2022, with odds ratios (ORs) as low as 0.66 (95% confidence interval: 0.51 to 0.86) in November 2021 for hospitalization. and as low as 0.23 (0.06 to 0.86) in June 2021 for death. However, prior infection was associated with an increased risk of hospitalization and death, mostly after March 2022 when the ORs were as high as 1.87 (1.26 to 2.80) and 2.99 (1.65 to 5.41) in April 2022, respectively. The overall OR for more severe disease was 1.06 (1.03 to 1.10) among previously infected individuals. Conclusion: In the pandemic's first two years, previously infected patients generally had less severe disease than people without prior infection. During the Omicron era, however, previously infected patients had the same or worse severity of disease as patients without prior infection.

How Do Family Physicians Document Patients' Social Needs in Electronic Health Records?

Social needs are critical determinants of patient health, but their capture in clinical records began recently. A representative survey of family physicians showed that 61% of respondents document social needs using notes, with fewer using diagnosis codes or electronic forms. This preference for unstructured documentation may make it difficult to connect patients across organizations or for policymakers and planners to identify geographic variation in needs.

Variation in Family Physicians' Experiences Across Different Electronic Health Record Platforms: a Descriptive Study.

BACKGROUND: Electronic health records (EHRs) have been connected to excessive workload and physician burnout. Little is known about variation in physician experience with different EHRs, however. OBJECTIVE: To analyze variation in reported usability and satisfaction across EHRs. DESIGN: Internet-based survey available between December 2021 and October 2022 integrated into American Board of Family Medicine (ABFM) certification process. PARTICIPANTS: ABFM-certified family physicians who use an EHR with at least 50 total responding physicians. MEASUREMENTS: Self-reported experience of EHR usability and satisfaction. KEY RESULTS: We analyzed the responses of 3358 physicians who used one of nine EHRs. Epic, athenahealth, and Practice Fusion were rated significantly higher across six measures of usability. Overall, between 10 and 30% reported being very satisfied with their EHR, and another 32 to 40% report being somewhat satisfied. Physicians who use athenahealth or Epic were most likely to be very satisfied, while physicians using Allscripts, Cerner, or Greenway were the least likely to be very satisfied. EHR-specific factors were the greatest overall influence on variation in satisfaction: they explained 48% of variation in the probability of being very satisfied with Epic, 46% with eClinical Works, 14% with athenahealth, and 49% with Cerner. CONCLUSIONS: Meaningful differences exist in physician-reported usability and overall satisfaction with EHRs, largely explained by EHR-specific factors. User-centric design and implementation, and robust ongoing evaluation are needed to reduce physician burden and ensure excellent experience with EHRs.

Estimated travel time and staffing constraints to accessing the Ethiopian health care system: A two-step floating catchment area analysis.

Background: Despite large investments in the public health care system, disparities in health outcomes persist between lower- and upper-income individuals, as well as rural vs urban dwellers in Ethiopia. Evidence from Ethiopia and other low- and middle-income countries suggests that challenges in health care access may contribute to poverty in these settings. Methods: We employed a two-step floating catchment area to estimate variations in spatial access to health care and in staffing levels at health care facilities. We estimated the average travel time from the population centers of administrative areas and adjusted them with provider-to-population ratios. To test hypotheses about the role of travel time vs staffing, we applied Spearman's rank tests to these two variables against the access score to assess the significance of observed variations. Results: Among Ethiopia's 11 first-level administrative units, Addis Ababa, Dire Dawa, and Harari had the best access scores. Regions with the lowest access scores were generally poorer and more rural/pastoral. Approximately 18% of the country did not have access to a public health care facility within a two-hour walk. Our results suggest that spatial access and staffing issues both contribute to access challenges. Conclusion: Investments both in new health facilities and staffing in existing facilities will be necessary to improve health care access within Ethiopia. Because rural and low-income areas are more likely to have poor access, future strategies for expanding and strengthening the health care system should strongly emphasize equity and the role of improved access in reducing poverty.

Equitable Prioritization of Health Interventions by Incorporating Financial Risk Protection Weights Into Economic Evaluations.

OBJECTIVES: Financial risk protection (FRP), or the prevention of medical impoverishment, is a major objective of health systems, particularly in low- and middle-income countries where the extent of out-of-pocket (OOP) health expenditures can be substantial. We sought to develop a method that allows decision makers to explicitly integrate FRP outcomes into their priority-setting activities. METHODS: We used literature review to identify 31 interventions in low- and middle-income countries, each of which provided measures of health outcomes, costs, OOP health expenditures averted, and FRP (proxied by OOP health expenditures averted as a percentage of income), all disaggregated by income quintile. We developed weights drawn from the Z-score of each quintile-intervention pair based on the distribution of FRP of all quintile-intervention pairs. We next ranked the interventions by unweighted and weighted health outcomes for each income quintile. We also evaluated how pro-poor they were by, first, ordering the interventions by cost-effectiveness for each quintile and, next, calculating the proportion of interventions each income quintile would be targeted for a given random budget. A ranking was said to be pro-poor if each quintile received the same or higher proportion of interventions than richer quintiles. RESULTS: Using FRP weights produced a more pro-poor priority setting than unweighted outcomes. Most of the reordering produced by the inclusion of FRP weights occurred in interventions of moderate cost-effectiveness, suggesting that these weights would be most useful as a way of distinguishing moderately cost-effective interventions with relatively high potential FRP. CONCLUSIONS: This preliminary method of integrating FRP into priority-setting would likely be most suitable to deciding between health interventions with intermediate cost-effectiveness.

Residual Disease Burden After Neoadjuvant Therapy Among US Patients With High-Risk HER2-positive Early-Stage Breast Cancer: A Population Effectiveness Model.

BACKGROUND: Approximately half of patients with high-risk HER2-positive early-stage breast cancer (ESBC) do not have pathologic complete response (pCR) after neoadjuvant therapy. The residual burden of disease among this population has not been previously quantified. MATERIALS AND METHODS: We used decision-modeling techniques to simulate recurrence, progression from locoregional to distant cancer, breast cancer-related mortality, and mortality from other causes over a 10-year period in a hypothetical cohort. We derived progression probabilities primarily from the KATHERINE trial of T-DM1 (ado-trastuzumab emtansine) and mortality outcomes from the published literature. Modeled outcomes included recurrences, breast cancer deaths, deaths from other causes, direct medical costs, and costs due to lost productivity. To estimate the residual disease burden, we compared outcomes from a cohort of patients treated with T-DM1 versus a hypothetical cohort with no disease recurrence. RESULTS: We estimated that 9,300 people would experience incident high-risk HER2-positive ESBC in the United States in 2021 based on cancer surveillance databases, clinical trial data, and expert opinion. We estimated that, in this group, 2,118 would experience disease recurrence, including 1,576 distant recurrences, and 1,358 would experience breast cancer deaths. This residual disease burden resulted in 6,435 life-years lost versus the recurrence-free cohort, and healthcare-related costs totaling $644 million, primarily associated with treating distant cancers. CONCLUSION: Patients with HER2-positive ESBC who do not achieve pCR after neoadjuvant therapy are at ongoing risk of recurrence despite the effectiveness of neoadjuvant treatment. There is substantial clinical and economic value in further reducing the residual disease burden in this population.

Radiologist Preferences for Artificial Intelligence-Based Decision Support During Screening Mammography Interpretation.

BACKGROUND: Artificial intelligence (AI) may improve cancer detection and risk prediction during mammography screening, but radiologists' preferences regarding its characteristics and implementation are unknown. PURPOSE: To quantify how different attributes of AI-based cancer detection and risk prediction tools affect radiologists' intentions to use AI during screening mammography interpretation. MATERIALS AND METHODS: Through qualitative interviews with radiologists, we identified five primary attributes for AI-based breast cancer detection and four for breast cancer risk prediction. We developed a discrete choice experiment based on these attributes and invited 150 US-based radiologists to participate. Each respondent made eight choices for each tool between three alternatives: two hypothetical AI-based tools versus screening without AI. We analyzed samplewide preferences using random parameters logit models and identified subgroups with latent class models. RESULTS: Respondents (n = 66; 44% response rate) were from six diverse practice settings across eight states. Radiologists were more interested in AI for cancer detection when sensitivity and specificity were balanced (94% sensitivity with <25% of examinations marked) and AI markup appeared at the end of the hanging protocol after radiologists complete their independent review. For AI-based risk prediction, radiologists preferred AI models using both mammography images and clinical data. Overall, 46% to 60% intended to adopt any of the AI tools presented in the study; 26% to 33% approached AI enthusiastically but were deterred if the features did not align with their preferences. CONCLUSION: Although most radiologists want to use AI-based decision support, short-term uptake may be maximized by implementing tools that meet the preferences of dissuadable users.

Implementation of pharmacogenomic clinical decision support for health systems: a cost-utility analysis.

We constructed a cost-effectiveness model to assess the clinical and economic value of a CDS alert program that provides pharmacogenomic (PGx) testing results, compared to no alert program in acute coronary syndrome (ACS) and atrial fibrillation (AF), from a health system perspective. We defaulted that 20% of 500,000 health-system members between the ages of 55 and 65 received PGx testing for CYP2C19 (ACS-clopidogrel) and CYP2C9, CYP4F2 and VKORC1 (AF-warfarin) annually. Clinical events, costs, and quality-adjusted life years (QALYs) were calculated over 20 years with an annual discount rate of 3%. In total, 3169 alerts would be fired. The CDS alert program would help avoid 16 major clinical events and 6 deaths for ACS; and 2 clinical events and 0.9 deaths for AF. The incremental cost-effectiveness ratio was $39,477/QALY. A PGx-CDS alert program was cost-effective, under a willingness-to-pay threshold of $100,000/QALY gained, compared to no alert program.

Economic evaluations of health system strengthening activities in low-income and middle-income country settings: a methodological systematic review.

OBJECTIVE: Health system strengthening (HSS) activities should accompany disease-targeting interventions in low/middle-income countries (LMICs). Economic evaluations provide information on how these types of investment might best be balanced but can be challenging. We conducted a systematic review to evaluate how researchers address these economic evaluation challenges. METHODS: We identified studies about economic evaluation of HSS activities in LMICs using a two-stage approach. First, we conducted a broad search to identify areas where economic evaluations of HSS activities were being conducted. Next, we selected specific interventions for more targeted literature review. We extracted study characteristics using the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) checklist. Finally, we summarised authors' modelling decisions using a framework that examines how models are developed to emphasise generalisability, precision, or realism. FINDINGS: Our searches produced 1978 studies, out of which we included 36. Most studies used data from prospective trials and calculated cost-effectiveness directly from these trial inputs, rather than using simulation methods. As a group, these studies primarily emphasised precision and realism over generalisability, meaning that their results were best suited to specific settings. CONCLUSIONS: The number of included studies was small. Our findings suggest that most economic evaluations of HSS do not leverage methods like sensitivity analyses or inputs from literature review that would produce more generalisable (but potentially less precise) results. More research into how decision-makers would use economic evaluations to define the expansion path to strengthening health systems would allow for conceptualising impactful work on the economic value of HSS.

Assessing the Economic Value of Clinical Artificial Intelligence: Challenges and Opportunities.

OBJECTIVES: Clinical artificial intelligence (AI) is a novel technology, and few economic evaluations have focused on it to date. Before its wider implementation, it is important to highlight the aspects of AI that challenge traditional health technology assessment methods. METHODS: We used an existing broad value framework to assess potential ways AI can provide good value for money. We also developed a rubric of how economic evaluations of AI should vary depending on the case of its use. RESULTS: We found that the measurement of core elements of value-health outcomes and cost-are complicated by AI because its generalizability across different populations is often unclear and because its use may necessitate reconfigured clinical processes. Clinicians' productivity may improve when AI is used. If poorly implemented though, AI may also cause clinicians' workload to increase. Some AI has been found to exacerbate health disparities. Nevertheless, AI may promote equity by expanding access to medical care and, when properly trained, providing unbiased diagnoses and prognoses. The approach to assessment of AI should vary based on its use case: AI that creates new clinical possibilities can improve outcomes, but regulation and evidence collection may be difficult; AI that extends clinical expertise can reduce disparities and lower costs but may result in overuse; and AI that automates clinicians' work can improve productivity but may reduce skills. CONCLUSIONS: The potential uses of clinical AI create challenges for health technology assessment methods originally developed for pharmaceuticals and medical devices. Health economists should be prepared to examine data collection and methods used to train AI, as these may impact its future value.

Economic Evaluation of Urine-Based or Magnetic Resonance Imaging Reflex Tests in Men With Intermediate Prostate-Specific Antigen Levels in the United States.

OBJECTIVES: For men with intermediate prostate-specific antigen (PSA) levels (4-10 ng/mL), urine-based biomarkers and multiparametric magnetic resonance imaging (MRI) are increasingly used as reflex tests before prostate biopsy. We assessed the cost effectiveness of these reflex tests in the United States. METHODS: We used an existing microsimulation model of prostate cancer (PCa) progression and survival to predict lifetime outcomes for a hypothetical cohort of 55-year-old men with intermediate PSA levels. Urine-based biomarkers-PCa antigen (PCA3), TMPRSS2:ERG gene fusion (T2:ERG), and the MyProstateScore (MPS) for any PCa and for high-grade (Gleason score ≥7) PCa (MPShg)-were generated using biomarker data from 1112 men presenting for biopsy at 10 United States institutions. MRI results were based on published sensitivity and specificity for high-grade PCa. Costs and utilities were sourced from literature and Medicare reimbursement schedules. Outcome measures included life years, quality-adjusted life years (QALYs), and lifetime medical costs per patient. Incremental cost-effectiveness ratios were empirically calculated on the basis of simulated life histories under different reflex testing strategies. RESULTS: Biopsying all men provided the most life years and QALYs, followed by reflex testing using MPShg, MPS, MRI, T2:ERG, PCA3, and biopsying no men (QALY range across strategies 15.98-16.09). Accounting for costs, MRI and MPShg were dominated by other strategies. PCA3, T2:ERG, and MPS were likely to be the most cost-effective strategy at willingness-to-pay thresholds of $100 000/QALY, $125 000/QALY, and $150 000/QALY, respectively. CONCLUSIONS: Using PCA3, T2:ERG, or MPS as reflex tests has greater economic value than MRI, biopsying all men, or biopsying no men with intermediate PSA levels.

Clarifying the Trade-Offs of Risk-Stratified Screening for Prostate Cancer: A Cost-Effectiveness Study.

Cancer risk prediction is necessary for precision early detection, which matches screening intensity to risk. However, practical steps for translating risk predictions to risk-stratified screening policies are not well established. We used a validated population prostate-cancer model to simulate the outcomes of strategies that increase intensity for men at high risk and reduce intensity for men at low risk. We defined risk by the Prompt Prostate Genetic Score (PGS) (Stratify Genomics, San Diego, California), a germline genetic test. We first recalibrated the model to reflect the disease incidence observed within risk strata using data from a large prevention trial where some participants were tested with Prompt PGS. We then simulated risk-stratified strategies in a population with the same risk distribution as the trial and evaluated the cost-effectiveness of risk-stratified screening versus universal (risk-agnostic) screening. Prompt PGS risk-adapted screening was more cost-effective when universal screening was conservative. Risk-stratified strategies improved outcomes at a cost of less than $100,000 per quality-adjusted life year compared with biennial screening starting at age 55 years, but risk stratification was not cost-effective compared with biennial screening starting at age 45. Heterogeneity of risk and fraction of the population within each stratum were also important determinants of cost-effectiveness.

Differences in the Selection of Health State Utility Values by Sponsorship in Published Cost-Effectiveness Analyses.

BACKGROUND: Health state utility values (HSUVs) are among the most influential attributes of cost-effectiveness analyses (CEAs). Our objective was to evaluate whether industry-funded studies select systematically different HSUVs as compared with studies without industry funding. METHODS: Among 10 diseases with high disease burden in the United States, we further identified 31 progressive health states. We then searched the Tufts Medical Center's CEA Registry to identify studies that included HSUVs and were submitted to the registry between 2002 and 2019. Two reviewers mapped the free-text descriptions of health states onto the 31 predefined health states. We analyzed the effect of industry funding on the point estimates of these HSUVs with a beta regression. We also analyzed the difference between related health states within studies by funding source with a linear regression. RESULTS: After identifying 26,222 HSUVs from 4198 CEAs, we matched 2573 HSUVs to the 31 predefined health states. We observed large variations within each health state: 12 of 31 health states included a range of HSUVs greater than 0.5. The point estimate model showed 1 statistically significant difference of 31 comparisons between studies with any industry funding and those without. The utility difference model found 3 significant differences out of 39 comparisons between CEAs with any industry funding and those without. LIMITATIONS: Inclusion of unpublished CEAs may have affected our conclusions about the effect of industry funding on selection of HSUVs. We also relied on free-text descriptions of health states available in the CEA Registry and did not include adjustment for multiple comparisons. CONCLUSION: Limited evidence exists that industry-funded studies select different HSUVs compared to non-industry-funded studies for the health states we considered.

Artificial intelligence in breast cancer screening: primary care provider preferences.

BACKGROUND: Artificial intelligence (AI) is increasingly being proposed for use in medicine, including breast cancer screening (BCS). Little is known, however, about referring primary care providers' (PCPs') preferences for this technology. METHODS: We identified the most important attributes of AI BCS for ordering PCPs using qualitative interviews: sensitivity, specificity, radiologist involvement, understandability of AI decision-making, supporting evidence, and diversity of training data. We invited US-based PCPs to participate in an internet-based experiment designed to force participants to trade off among the attributes of hypothetical AI BCS products. Responses were analyzed with random parameters logit and latent class models to assess how different attributes affect the choice to recommend AI-enhanced screening. RESULTS: Ninety-one PCPs participated. Sensitivity was most important, and most PCPs viewed radiologist participation in mammography interpretation as important. Other important attributes were specificity, understandability of AI decision-making, and diversity of data. We identified 3 classes of respondents: "Sensitivity First" (41%) found sensitivity to be more than twice as important as other attributes; "Against AI Autonomy" (24%) wanted radiologists to confirm every image; "Uncertain Trade-Offs" (35%) viewed most attributes as having similar importance. A majority (76%) accepted the use of AI in a "triage" role that would allow it to filter out likely negatives without radiologist confirmation. CONCLUSIONS AND RELEVANCE: Sensitivity was the most important attribute overall, but other key attributes should be addressed to produce clinically acceptable products. We also found that most PCPs accept the use of AI to make determinations about likely negative mammograms without radiologist confirmation.

Provider preferences for resolving uncertainty and avoiding harms in precision medicine: a discrete choice experiment.

Background: Substantial uncertainty exists about how providers assess the value of genomic testing. Materials & methods: We developed and administered a discrete choice experiment to a national sample of providers. We analyzed responses using an error components mixed logit model. Results: We received responses from 356 providers. The attributes important to providers were patient health and function, life expectancy, cost, expert agreement, and biomarker prevalence. Providers significantly valued reducing uncertainty only when it eliminated the possibility of decreased life expectancy. Providers valued improving certainty about life expectancy gains from 12 ± 18 to 12 ± 6 months at US$400 (US$200-600) versus US$200 (-US$60-500) for 4 ± 4 to 4 ± 2 years. Conclusion: Providers value resolving uncertainty most when it eliminates the possibility of substantial harm.

Inferring antenatal care visit timing in low- and middle-income countries: Methods to inform potential maternal vaccine coverage.

BACKGROUND: The timing of antenatal care (ANC) visits directly affect health intervention coverage and impact, especially for those interventions requiring strict gestational age windows for administration, such as maternal respiratory syncytial virus (RSV) vaccine. Existing nationally representative population-based surveys do not record the timing of ANC visits beyond the first, limiting the availability of reliable data around timing of subsequent ANC visits in most low- and middle-income countries (LMICs). Here, we describe a model that estimates the timing of ANC visits by gestational age using publicly available multi-country survey data. METHODS AND FINDINGS: We used the Demographic and Health Surveys (DHS) data from 69 LMICs. We used several factors to estimate the timing of subsequent ANC visits by gestation age: the timing of the first ANC visit (ANC1) in a given pregnancy, derived from the DHS; the country's reported average ANC coverage at each ANC visit (ANC1 through the fourth ANC visit [ANC4]); and the World Health Organization's guidance on recommended ANC visit. We then used the timing of ANC visit by gestation age to predict the coverage of a potential maternal RSV vaccine administered at 24-36 weeks of gestation. We calculated the maternal immunization coverage by summing the number of eligible women vaccinated at any ANC visit divided by the total number of pregnant women. We find, in general, countries with higher ANC1 coverage were predicted to have higher vaccination coverage. In 82% of countries, the modeled vaccine coverage is less than ANC4 coverage. CONCLUSIONS: The methods illustrated in this paper have implications on the precision of estimating impact and programmatic feasibility of time-critical interventions, especially for pregnant women. The methods can be easily adapted to vaccine demand forecasts models, vaccine impact assessments, and cost-effectiveness analyses and can be adapted to other maternal interventions that have administration timing restrictions.