Category: Statistics

JMIR Form Res. 2024 May 29;8:e51408. doi: 10.2196/51408.

ABSTRACT

BACKGROUND: The potential of health apps for health promotion and disease prevention is widely recognized. However, uptake is limited due to barriers individuals face in finding suitable and trustworthy apps, such as the overwhelming amount of available health apps. Therefore, the health app platform “FitKnip” was developed, enabling individuals to purchase preselected, trustworthy health apps with a budget of 100 euros (a currency exchange rate of EUR €1=US $1.0831 is applicable). The platform aimed to empower individuals to improve their health and vitality, ultimately supporting a more healthy society.

OBJECTIVE: The primary aim of this study was to evaluate the health app platform in terms of feasibility and acceptability. Potential effects on health empowerment and health outcomes were secondarily explored.

METHODS: This quantitative study was part of a mixed methods study with a prospective pre-post interventional design. We collected web-based user data, and self-reported web-based questionnaires were collected over 5 measurements over an 8-month period. Use statistics were tracked on the platform, including the number of purchased apps and euros spent per user registered within the health app platform. We measured the user-friendliness of the health app platform using the System Usability Scale (SUS) and satisfaction using the Client Satisfaction Questionnaire-8 (CSQ-8) and several 10-point Likert items. We asked participants to indicate, on a scale from 1 (not at all) to 10 (completely), how much the health app platform contributed to various areas related to health empowerment. We assessed health-related quality of life by the 12-item Short-Form Health Survey (SF-12) and one’s perceived level of stress by the 10-item Perceived Stress Scale (PSS-10).

RESULTS: A total of 1650 participants were included, of whom 42% (685/1650) bought at least 1 app. The majority of those purchased one app (244/685, 35.6%). The health app platform was rated as user-friendly (SUS mean 66.5, SD 20.7; range 66.5-70.0), and the acceptability of the health app platform was moderate (CSQ-8 mean 20.0, SD 1.5; range 19.6-20.0). Results furthermore showed that participants were generally satisfied to highly satisfied with the ease of the payment system to purchase apps on the platform (median 8, IQR 7-10), the look and feel of the platform (median 7, IQR 6-8), as well as the provided budget of 100 euros (median 9, IQR 7-10). Participants were less satisfied with the amount (median 6, IQR 4-7) and diversity (median 6, IQR 4-7) of apps offered on the platform.

CONCLUSIONS: A health app platform is a promising initiative to enhance public health. Feasibility and acceptability are critical for success, as they ensure that such a platform is accessible, user-friendly, and meets end users’ needs and preferences. This can help to increase uptake, engagement, and ultimately the platform’s adoption and effectiveness.

PMID:38809585 | DOI:10.2196/51408

JAMA Psychiatry. 2024 May 29. doi: 10.1001/jamapsychiatry.2024.0994. Online ahead of print.

ABSTRACT

IMPORTANCE: Placebo is the only substance systematically evaluated across common psychiatric diagnoses, but comprehensive cross-diagnostic comparisons are lacking.

OBJECTIVE: To compare changes in placebo groups in recent high-quality randomized clinical trials (RCTs) across a broad spectrum of psychiatric disorders in adult patients.

DATA SOURCES: MEDLINE and the Cochrane Database of Systematic Reviews were systematically searched in March 2022 for the latest systematic reviews meeting predetermined high-quality criteria for 9 major psychiatric diagnoses.

STUDY SELECTION: Using these reviews, the top 10 highest-quality (ie, lowest risk of bias, according to the Cochrane Risk of Bias tool) and most recent placebo-controlled RCTs per diagnosis (totaling 90 RCTs) were selected, adhering to predetermined inclusion and exclusion criteria.

DATA EXTRACTION AND SYNTHESIS: Following the Cochrane Handbook, 2 authors independently carried out the study search, selection, and data extraction. Cross-diagnosis comparisons were based on standardized pre-post effect sizes (mean change divided by its SD) for each placebo group. This study is reported following the Meta-analysis of Observational Studies in Epidemiology (MOOSE) reporting guideline.

MAIN OUTCOME AND MEASURE: The primary outcome, pooled pre-post placebo effect sizes (dav) with 95% CIs per diagnosis, was determined using random-effects meta-analyses. A Q test assessed statistical significance of differences across diagnoses. Heterogeneity and small-study effects were evaluated as appropriate.

RESULTS: A total of 90 RCTs with 9985 placebo-treated participants were included. Symptom severity improved with placebo in all diagnoses. Pooled pre-post placebo effect sizes differed across diagnoses (Q = 88.5; df = 8; P < .001), with major depressive disorder (dav = 1.40; 95% CI, 1.24-1.56) and generalized anxiety disorder (dav = 1.23; 95% CI, 1.06-1.41) exhibiting the largest dav. Panic disorder, attention-deficit/hyperactivity disorder, posttraumatic stress disorder, social phobia, and mania showed dav between 0.68 and 0.92, followed by OCD (dav = 0.65; 95% CI, 0.51-0.78) and schizophrenia (dav = 0.59; 95% CI, 0.41-0.76).

CONCLUSION AND RELEVANCE: This systematic review and meta-analysis found that symptom improvement with placebo treatment was substantial in all conditions but varied across the 9 included diagnoses. These findings may help in assessing the necessity and ethical justification of placebo controls, in evaluating treatment effects in uncontrolled studies, and in guiding patients in treatment decisions. These findings likely encompass the true placebo effect, natural disease course, and nonspecific effects.

PMID:38809560 | DOI:10.1001/jamapsychiatry.2024.0994

JAMA Netw Open. 2024 May 1;7(5):e2413453. doi: 10.1001/jamanetworkopen.2024.13453.

ABSTRACT

IMPORTANCE: Associations of domain-specific physical activity with stroke incidence and poststroke outcomes have not been extensively studied using long-term, population-based data.

OBJECTIVE: To investigate associations of leisure time, work time, transport, and household physical activity with stroke incidence and death or dependency in activities of daily living (ADL) 3 months after stroke.

DESIGN, SETTING, AND PARTICIPANTS: The prospective, population-based Interplay Between Genetic Susceptibility and External Factors (INTERGENE) cohort study was conducted among a random sample of individuals from an urban-rural area covering western Sweden; 3614 individuals aged 24 to 77 years were examined in 2001 to 2004, and 1394 individuals were reexamined in 2014 to 2016. The median (range) follow-up was 20.0 years (56 days to 21.9 years). Data were analyzed from September through October 2023.

EXPOSURE: Physical activity levels were self-reported for leisure time, work time, transportation, and household domains. The mean number of steps taken over a 6-day period was collected in a subgroup of participants using a sealed pedometer.

MAIN OUTCOMES AND MEASURES: Follow-up for stroke incidence and mortality rates continued until December 31, 2022. The composite outcome of death or ADL dependency was assessed at 3 months after stroke.

RESULTS: Among 3614 individuals (mean [SD] age, 51.4 [13.1] years; 1910 female [52.9%]); 269 individuals (7.4%) developed stroke, of whom 120 individuals (44.6%) were dead or ADL dependent at 3 months. Intermediate (adjusted hazard ratio [aHR], 0.54; 95% CI, 0.38-0.77) and high (aHR, 0.47; 95% CI, 0.31-0.73) levels of leisure time physical activity were associated with a reduced incidence of stroke compared with low levels, as was an intermediate level of physical activity in transportation (aHR, 0.69; 95% CI, 0.52-0.93). High levels of leisure time physical activity were also associated with a reduced risk of poststroke death or ADL dependency (adjusted odds ratio, 0.34; 95% CI, 0.16-0.71) compared with low levels. Work time and household physical activity were not associated with stroke incidence or stroke outcomes. In exploratory subgroup analyses, there were interactions between physical activity and smoking (current smoking or smoking in the past year associated with stroke risk only in participants with low or intermediate physical activity: aHR, 2.33; 95% CI, 1.72-3.15) and family history of stroke (first-degree relative with a history of stroke associated with stroke risk only in participants with low or intermediate physical activity: aHR, 1.73; 95% CI, 1.27-2.38).

CONCLUSIONS AND RELEVANCE: In this study, leisure time and transport-related physical activities were associated with a reduced risk of stroke. A high level of leisure time physical activity was also associated with a lower risk of death or ADL dependency 3 months after stroke.

PMID:38809556 | DOI:10.1001/jamanetworkopen.2024.13453

JAMA Netw Open. 2024 May 1;7(5):e2413644. doi: 10.1001/jamanetworkopen.2024.13644.

ABSTRACT

IMPORTANCE: Sweetened beverage taxes have been associated with reduced purchasing of taxed beverages. However, few studies have assessed the association between sweetened beverage taxes and health outcomes.

OBJECTIVE: To evaluate the association between the Seattle sweetened beverage tax and change in body mass index (BMI) among children.

DESIGN, SETTING, AND PARTICIPANTS: In this longitudinal cohort study, anthropometric data were obtained from electronic medical records of 2 health care systems (Kaiser Permanente Washington [KP] and Seattle Children’s Hospital Odessa Brown Children’s Clinic [OBCC]). Children were included in the study if they were aged 2 to 18 years (between January 1, 2014, and December 31, 2019); had at least 1 weight measurement every year between 2015 and 2019; lived in Seattle or in urban areas of 3 surrounding counties (King, Pierce, and Snohomish); had not moved between taxed (Seattle) and nontaxed areas; received primary health care from KP or OBCC; did not have a recent history of cancer, bariatric surgery, or pregnancy; and had biologically plausible height and BMI (calculated as weight in kilograms divided by height in meters squared). Data analysis was conducted between August 5, 2022, and March 4, 2024.

EXPOSURE: Seattle sweetened beverage tax (1.75 cents per ounce on sweetened beverages), implemented on January 1, 2018.

MAIN OUTCOMES AND MEASURES: The primary outcome was BMIp95 (BMI expressed as a percentage of the 95th percentile; a newly recommended metric for assessing BMI change) of the reference population for age and sex, using the Centers for Disease Control and Prevention growth charts. In the primary (synthetic difference-in-differences [SDID]) model used, a comparison sample was created by reweighting the comparison sample to optimize on matching to pretax trends in outcome among 6313 children in Seattle. Secondary models were within-person change models using 1 pretax measurement and 1 posttax measurement in 22 779 children and fine stratification weights to balance baseline individual and neighborhood-level confounders.

RESULTS: The primary SDID analysis included 6313 children (3041 female [48%] and 3272 male [52%]). More than a third of children (2383 [38%]) were aged 2 to 5 years); their mean (SE) age was 7.7 (0.6) years. With regard to race and ethnicity, 789 children (13%) were Asian, 631 (10%) were Black, 649 (10%) were Hispanic, and 3158 (50%) were White. The primary model results suggested that the Seattle tax was associated with a larger decrease in BMIp95 for children living in Seattle compared with those living in the comparison area (SDID: -0.90 percentage points [95% CI, -1.20 to -0.60]; P < .001). Results from secondary models were similar.

CONCLUSIONS AND RELEVANCE: The findings of this cohort study suggest that the Seattle sweetened beverage tax was associated with a modest decrease in BMIp95 among children living in Seattle compared with children living in nearby nontaxed areas who were receiving care within the same health care systems. Taken together with existing studies in the US, these results suggest that sweetened beverage taxes may be an effective policy for improving children’s BMI. Future research should test this association using longitudinal data in other US cities with sweetened beverage taxes.

PMID:38809555 | DOI:10.1001/jamanetworkopen.2024.13644

JAMA Netw Open. 2024 May 1;7(5):e2413754. doi: 10.1001/jamanetworkopen.2024.13754.

ABSTRACT

IMPORTANCE: People with kidney failure receiving maintenance dialysis visit the emergency department (ED) 3 times per year on average, which is 3- to 8-fold more often than the general population. Little is known about the factors that contribute to potentially preventable ED use in this population.

OBJECTIVE: To identify the clinical and sociodemographic factors associated with potentially preventable ED use among patients receiving maintenance dialysis.

DESIGN, SETTING, AND PARTICIPANTS: This cohort study used linked administrative health data within the Alberta Kidney Disease Network to identify adults aged 18 years or older receiving maintenance dialysis (ie, hemodialysis or peritoneal dialysis) between April 1, 2010, and March 31, 2019. Patients who had been receiving dialysis for more than 90 days were followed up from cohort entry (defined as dialysis start date plus 90 days) until death, outmigration from the province, receipt of a kidney transplant, or end of study follow-up. The Andersen behavioral model of health services was used as a conceptual framework to identify variables related to health care need, predisposing factors, and enabling factors. Data were analyzed in March 2024.

MAIN OUTCOMES AND MEASURES: Rates of all-cause ED encounters and potentially preventable ED use associated with 4 kidney disease-specific ambulatory care-sensitive conditions (hyperkalemia, heart failure, volume overload, and malignant hypertension) were calculated. Multivariable negative binomial regression models were used to examine the association between clinical and sociodemographic factors and rates of potentially preventable ED use.

RESULTS: The cohort included 4925 adults (mean [SD] age, 60.8 [15.5] years; 3071 males [62.4%]) with kidney failure receiving maintenance hemodialysis (3183 patients) or peritoneal dialysis (1742 patients) who were followed up for a mean (SD) of 2.5 (2.0) years. In all, 3877 patients had 34 029 all-cause ED encounters (3100 [95% CI, 2996-3206] encounters per 1000 person-years). Of these, 755 patients (19.5%) had 1351 potentially preventable ED encounters (114 [95% CI, 105-124] encounters per 1000 person-years). Compared with patients with a nonpreventable ED encounter, patients with a potentially preventable ED encounter were more likely to be in the lowest income quintile (38.8% vs 30.9%; P < .001); to experience heart failure (46.8% vs 39.9%; P = .001), depression (36.6% vs 32.5%; P = .03), and chronic pain (60.1% vs 54.9%; P = .01); and to have a longer duration of dialysis (3.6 vs 2.6 years; P < .001). In multivariable regression analyses, potentially preventable ED use was higher for younger adults (incidence rate ratio [IRR], 1.69 [95% CI, 1.33-2.15] for those aged 18 to 44 years) and patients with chronic pain (IRR, 1.35 [95% CI, 1.14-1.61]), greater material deprivation (IRR, 1.57 [95% CI, 1.16-2.12]), a history of hyperkalemia (IRR, 1.31 [95% CI, 1.09-1.58]), and historically high ED use (ie, ≥3 ED encounters in the prior year; IRR, 1.46 [95% CI, 1.23-1.73).

CONCLUSIONS AND RELEVANCE: In this study of adults receiving maintenance dialysis in Alberta, Canada, among those with ED use, 1 in 5 had a potentially preventable ED encounter; reasons for such encounters were associated with both psychosocial and medical factors. The findings underscore the need for strategies that address social determinants of health to avert potentially preventable ED use in this population.

PMID:38809552 | DOI:10.1001/jamanetworkopen.2024.13754

JAMA Netw Open. 2024 May 1;7(5):e2413847. doi: 10.1001/jamanetworkopen.2024.13847.

ABSTRACT

IMPORTANCE: The Supreme Court decision Dobbs v Jackson Women’s Health Organization (Dobbs) overturned federal protections to abortion care and altered the reproductive health care landscape. Thus far, aggregated state-level data reveal increases in the number of abortions in states where abortion is still legal, but there is limited information on delays to care and changes in the characteristics of people accessing abortion in these states after Dobbs.

OBJECTIVE: To examine changes in abortion provision and delays to care after Dobbs.

DESIGN, SETTING, AND PARTICIPANTS: Retrospective cohort study of all abortions performed at an independent, high-volume reproductive health care clinic network in Washington state from January 1, 2017, to July 31, 2023. Using an interrupted time series, the study assessed changes in abortion care after Dobbs.

EXPOSURE: Abortion care obtained after (June 24, 2022, to July 31, 2023) vs before (January 1, 2017, to June 23, 2022) Dobbs.

MAIN OUTCOME AND MEASURE: Primary outcomes included weekly number of abortions and out-of-state patients and weekly average of gestational duration (days) and time to appointment (days).

RESULTS: Among the 18 379 abortions during the study period, most were procedural (13 192 abortions [72%]) and funded by public insurance (11 412 abortions [62%]). The mean (SD) age of individuals receiving abortion care was 28.5 (6.44) years. Following Dobbs, the number of procedural abortions per week increased by 6.35 (95% CI, 2.83-9.86), but then trended back toward pre-Dobbs levels. The number of out-of-state patients per week increased by 2 (95% CI, 1.1-3.6) and trends remained stable. The average gestational duration per week increased by 6.9 (95% CI, 3.6-10.2) days following Dobbs, primarily due to increased gestations of procedural abortions. The average gestational duration among out-of-state patients did not change following Dobbs, but it did increase by 6 days for in-state patients (5.9; 95% CI, 3.2-8.6 days). There were no significant changes in time to appointment.

CONCLUSIONS AND RELEVANCE: These findings provide a detailed picture of changes in abortion provision and delays to care after Dobbs in a state bordering a total ban state. In this study, more people traveled from out of state to receive care and in-state patients sought care a week later in gestation. These findings can inform interventions and policies to improve access for all seeking abortion care.

PMID:38809551 | DOI:10.1001/jamanetworkopen.2024.13847

JAMA Netw Open. 2024 May 1;7(5):e2413855. doi: 10.1001/jamanetworkopen.2024.13855.

ABSTRACT

IMPORTANCE: Free video-sharing platforms (VSPs) make up a high proportion of children’s daily screen use. Many VSPs make algorithmic recommendations, appearing as thumbnail images from the video, which content creators use to advertise their video content.

OBJECTIVE: To explore how VSP thumbnails use attention-capture designs to encourage engagement with content and to test whether VSP algorithmic recommendations offer more problematic thumbnail features over time.

DESIGN, SETTING, AND PARTICIPANTS: In this cross-sectional study conducted in January 2022, researchers mimicked the search behavior of children on a popular VSP by randomly clicking on recommended videos in order to test whether thumbnail designs changed over 20 sequential video engagements. A digital, footprint-free data collection setting was created by using a new computer and wireless internet router. Data were collected from YouTube via an internet browser not logged into a user account. Data analysis occurred from April to December 2022.

EXPOSURES: Manual searches using 12 top-searched terms popular with school-aged children were conducted. Researchers captured the video thumbnails recommended at the end of each video and randomly clicked subsequent videos for 20 sequential engagements.

MAIN OUTCOMES AND MEASURES: Thumbnail content codes were developed through iterative review of screenshots by a multidisciplinary research team and applied by trained coders (reliability, κ >.70). The prevalence of problematic thumbnail content and change in prevalence over 20 engagements was calculated using the Cochran-Armitage trend test.

RESULTS: A total of 2880 video thumbnails were analyzed and 6 features were coded, including visual loudness; drama and intrigue; lavish excess and wish fulfillment; creepy, bizarre, and disturbing; violence, peril, and pranks; and gender stereotypes. A high proportion contained problematic features including the creepy, bizarre, and disturbing feature (1283 thumbnails [44.6%]), violence, peril, and pranks feature (1170 thumbnails [40.6%]), and gender stereotypes feature (525 thumbnails [18.2%]). Other features included attention-capture designs such as the visual loudness feature (2278 thumbnails [79.1%]), drama and intrigue feature (2636 thumbnails [91.5%]) and lavish excess and wish fulfillment feature (1286 thumbnails [44.7%]). Contrary to the hypotheses, problematic feature prevalence did not increase over time, but the gender stereotypes feature increased with more engagement in the recommendations feed (P for trend < .001).

CONCLUSIONS AND RELEVANCE: In this study of video recommendations for search terms popular with children, thumbnails contained problematic and attention-capturing designs including violent, stereotyped, and frightening themes. Research is needed to understand how children respond to thumbnail designs and whether such designs influence the quality of content children consume.

PMID:38809550 | DOI:10.1001/jamanetworkopen.2024.13855

JAMA Dermatol. 2024 May 29. doi: 10.1001/jamadermatol.2024.1315. Online ahead of print.

ABSTRACT

IMPORTANCE: Frontal fibrosing alopecia (FFA) is an increasingly prevalent form of follicular lichen planus, causing irreversible hair loss predominantly in postmenopausal individuals. An earlier genome-wide meta-analysis of female FFA identified risk loci in genes implicated in self-antigen presentation and T-cell homeostasis, including HLA-B*07:02, ST3GAL1, and SEMA4B. However, CYP1B1, which is important for hormone metabolism, was also implicated with the substitution of serine for asparagine at position 453 (c.1358A>G, p.Asn453Ser) exhibiting a protective effect against FFA. Increasing understanding of genetic and environmental variables and their interactions will improve understanding of disease pathogenesis and has the potential to inform risk mitigation strategies.

OBJECTIVE: To investigate whether oral contraceptive pill (OCP) use modulates the protective effect of the common missense variant in CYP1B1 (c.1358A>G, p.Asn453Ser) on FFA risk.

DESIGN, SETTING, AND PARTICIPANTS: This gene-environment interaction study using a case-control design enrolled female patients with FFA from UK-based dermatology clinics. The patients were matched with unrelated age- and ancestry-matched female control individuals derived from UK Biobank in a 1:66 ratio, determined by the first 4 principal components from genome-wide genotypes. Data were collected from July 2015 to September 2017, and analyzed from October 2022 to December 2023.

MAIN OUTCOME AND MEASURE: The main outcomes were the modulatory effect of OCP use on the contribution of the CYP1B1 missense variant to female FFA risk and a formal gene-environment interaction test evaluated by a logistic regression model with a multiplicative interaction term, under the assumptions of an additive genetic model interaction term, under the assumptions of an additive genetic model.

RESULTS: Of the 489 female patients with FFA, the mean (SD) age was 65.8 (9.7) years, and 370 (75.7%) had a history of OCP use. Of the 34 254 age- and ancestry-matched control individuals, the mean (SD) age was 65.0 (8.4) years, and previous OCP use was reported in 31 177 (91.0%). An association between female FFA and the CYP1B1 risk allele was observed in individuals who reported OCP use (odds ratio, 1.90 [95% CI, 1.50-2.40]; P = 8.41 × 10-8) but not in those with no documented exposure to OCPs (odds ratio, 1.16 [95% CI, 0.82-1.64]; P = .39). A full gene-environment interaction model demonstrated a significant additive statistical interaction between c.1358A, p.453Asn, and history of OCP use on FFA risk (OR for interaction, 1.63 [95% CI, 1.07-2.46]; P = .02).

CONCLUSIONS AND RELEVANCE: This gene-environment interaction analysis suggests that the protective effect of the CYP1B1 missense variant on FFA risk might be mediated by exposure to OCPs. The allele that encodes an asparagine at position 453 of CYP1B1 was associated with increased odds of FFA only in participants with OCP history.

PMID:38809548 | DOI:10.1001/jamadermatol.2024.1315

Cancer. 2024 May 29. doi: 10.1002/cncr.35383. Online ahead of print.

ABSTRACT

BACKGROUND: Pancreatic ductal adenocarcinoma (PDAC) surveillance is recommended for some individuals with a pathogenic or likely pathogenic variant (PV/LPV) in a PDAC susceptibility gene; the recommendation is often dependent on family history of PDAC. This study aimed to describe PDAC family history in individuals with PDAC who underwent genetic testing to determine the appropriateness of including a family history requirement in these recommendations.

METHODS: Individuals with PDAC with a germline heterozygous PV/LPV in ATM, BRCA1, BRCA2, EPCAM, MLH1, MSH2, MSH6, PALB2, or PMS2 (PV/LPV carriers) were assessed for family history of PDAC in first-degree relatives (FDRs) or second-degree relatives (SDRs) from nine institutions. A control group of individuals with PDAC without a germline PV/LPV was also assessed.

RESULTS: The study included 196 PV/LPV carriers and 1184 controls. In the PV/LPV carriers, 25.5% had an affected FDR and/or SDR compared to 16.9% in the control group (p = .004). PV/LPV carriers were more likely to have an affected FDR compared to the controls (p = .003) but there was no statistical difference when assessing only affected SDRs (p = .344).

CONCLUSIONS: Most PV/LPV carriers who developed PDAC did not have a close family history of PDAC and would not have met most current professional societies’ recommendations for consideration of PDAC surveillance before diagnosis. However, PV/LPV carriers were significantly more likely to have a family history of PDAC, particularly an affected FDR. These findings support family history as a risk modifier in PV/LPV carriers, and highlight the need to identify other risk factors.

PMID:38809542 | DOI:10.1002/cncr.35383

Transl Vis Sci Technol. 2024 May 1;13(5):23. doi: 10.1167/tvst.13.5.23.

ABSTRACT

PURPOSE: To develop convolutional neural network (CNN)-based models for predicting the axial length (AL) using color fundus photography (CFP) and explore associated clinical and structural characteristics.

METHODS: This study enrolled 1105 fundus images from 467 participants with ALs ranging from 19.91 to 32.59 mm, obtained at National Taiwan University Hospital between 2020 and 2021. The AL measurements obtained from a scanning laser interferometer served as the gold standard. The accuracy of prediction was compared among CNN-based models with different inputs, including CFP, age, and/or sex. Heatmaps were interpreted by integrated gradients.

RESULTS: Using age, sex, and CFP as input, the mean ± standard deviation absolute error (MAE) for AL prediction by the model was 0.771 ± 0.128 mm, outperforming models that used age and sex alone (1.263 ± 0.115 mm; P < 0.001) and CFP alone (0.831 ± 0.216 mm; P = 0.016) by 39.0% and 7.31%, respectively. The removal of relatively poor-quality CFPs resulted in a slight MAE reduction to 0.759 ± 0.120 mm without statistical significance (P = 0.24). The inclusion of age and CFP improved prediction accuracy by 5.59% (P = 0.043), while adding sex had no significant improvement (P = 0.41). The optic disc and temporal peripapillary area were highlighted as the focused areas on the heatmaps.

CONCLUSIONS: Deep learning-based prediction of AL using CFP was fairly accurate and enhanced by age inclusion. The optic disc and temporal peripapillary area may contain crucial structural information for AL prediction in CFP.

TRANSLATIONAL RELEVANCE: This study might aid AL assessments and the understanding of the morphologic characteristics of the fundus related to AL.

PMID:38809531 | DOI:10.1167/tvst.13.5.23