Application of Deep Learning Models to Improve Ulcerative Colitis Endoscopic Disease Activity Scoring Under Multiple Scoring Systems.

BACKGROUND AND AIMS: Lack of clinical validation and inter-observer variability are two limitations of endoscopic assessment and scoring of disease severity in patients with ulcerative colitis [UC]. We developed a deep learning [DL] model to improve, accelerate and automate UC detection, and predict the Mayo Endoscopic Subscore [MES] and the Ulcerative Colitis Endoscopic Index of Severity [UCEIS]. METHODS: A total of 134 prospective videos [1550 030 frames] were collected and those with poor quality were excluded. The frames were labelled by experts based on MES and UCEIS scores. The scored frames were used to create a preprocessing pipeline and train multiple convolutional neural networks [CNNs] with proprietary algorithms in order to filter, detect and assess all frames. These frames served as the input for the DL model, with the output being continuous scores for MES and UCEIS [and its components]. A graphical user interface was developed to support both labelling video sections and displaying the predicted disease severity assessment by the artificial intelligence from endoscopic recordings. RESULTS: Mean absolute error [MAE] and mean bias were used to evaluate the distance of the continuous model's predictions from ground truth, and its possible tendency to over/under-predict were excellent for MES and UCEIS. The quadratic weighted kappa used to compare the inter-rater agreement between experts' labels and the model's predictions showed strong agreement [0.87, 0.88 at frame-level, 0.88, 0.90 at section-level and 0.90, 0.78 at video-level, for MES and UCEIS, respectively]. CONCLUSIONS: We present the first fully automated tool that improves the accuracy of the MES and UCEIS, reduces the time between video collection and review, and improves subsequent quality assurance and scoring.

Lung cancer risk associated with occupational exposure to nickel, chromium VI, and cadmium in two population-based case-control studies in Montreal.

BACKGROUND: Nickel, chromium VI, and cadmium have been identified as lung carcinogens in highly exposed cohorts. The purpose of this study was to examine the etiological link between lung cancer and these metals in occupations, that usually entail lower levels of exposure than those seen in historical cohorts. METHODS: Two population-based case-control studies were conducted in Montreal, from 1979 to 1986 and from 1996 to 2001, comprising 1,598 cases and 1,965 controls. A detailed job history was obtained to evaluate lifetime occupational exposure to many agents, including nickel, chromium VI, and cadmium compounds. RESULTS: Lung cancer odds ratios were increased only among former or non-smokers: 2.5 (95% CI: 1.3-4.7) for nickel exposure, 2.4 (95% CI: 1.2-4.8) for chromium VI, and 4.7 (95% CI: 1.5-14.3) for cadmium. The metals did not increase risk among smokers. CONCLUSIONS: While excess risks due to these metal compounds were barely discernable among smokers, carcinogenic effects were seen among non-smokers.

Sexual violence among two populations of men at high risk of HIV infection.

This study sought to compare the prevalence of, and relationship between, age at first experience of sexual violence and HIV and other health risk behaviors in two populations of men at high risk of HIV infection. Data were drawn from two cohorts: Vanguard, a prospective study of young men who have sex with men (MSM), and VIDUS, the Vancouver Injection Drug Users Study. Controlling for fixed sociodemographics, multivariate logistic regression was used to assess the relationship between age at first sexual violence (vs. never experiencing it) and several health risk behaviors. There were 140/498 (28%) MSM from Vanguard and 173/932 (19%) injection drug users (IDU) from VIDUS who reported having experienced sexual violence. Among VIDUS men, 130/852 (15%) IDU-only and 43/80 (54%) who were both IDU and MSM reported a history of sexual violence. The prevalence of child sexual abuse was 13% in Vanguard MSM, and 11% among VIDUS IDU-only, but 26% among VIDUS MSM/IDU. The median age of onset was significantly lower among VIDUS IDU-only compared to the two other groups. Experiencing sexual violence first in childhood was strongly related to ever being in the sex trade in both IDU and MSM. MSM in Vanguard who experienced sexual violence in childhood were more likely to have attempted suicide, and have a diagnosed mood disorder. Non-MSM IDU in VIDUS who experienced sexual violence in childhood were more likely to have a diagnosed mental illness, to binge on alcohol, and to have ever accidentally overdosed. In conclusion, men who have ever had sex with men appear to have a higher lifetime prevalence of sexual violence, compared to non-MSM injection drug users. Sexual violence is differentially associated with different health risk behaviors, depending on the age at first occurrence and the primary HIV risk factor (i.e. MSM vs. IDU).

Molecular and clinical epidemiology of CXCR4-using HIV-1 in a large population of antiretroviral-naive individuals.

OBJECTIVE: We wished to characterize the epidemiological and clinical correlates of CXCR4-using human immunodeficiency virus type 1 (HIV-1) ("X4 variants") in a cross-sectional analysis of a large population of antiretroviral-naive individuals. METHODS: HIV-1 coreceptor use was determined in the last pretherapy plasma sample for 1191 individuals initiating triple-combination therapy in British Columbia, Canada. Baseline variables investigated included sociodemographic characteristics, plasma viral load (pVL), CD4 cell count, AIDS diagnosis, HIV-1 V3 loop sequence, and human CCR5 Delta 32 genotype. RESULTS: Individuals harboring X4 variants (n = 178 of 979 phenotyped samples; 18.2%) displayed a poorer baseline clinical profile than individuals harboring exclusively CCR5-using HIV-1 ("R5 variants") (median pVL, 175,000 vs. 120,000 copies of HIV-1 RNA/mL [P = .0006]; median CD4 cell count, 110 vs. 290 cells/mm(3) [P < .0001]). Individuals heterozygous for the CCR5 Delta 32 deletion (n = 128 of 967; 13.2%) were at 2.5 times higher risk of harboring X4 variants, compared with those without the deletion (multivariate P = .0005). The presence of basic amino acids at codon 11 and/or codon 25 of HIV-1 V3 (n = 109 of 955; 11.4%) was associated with a 9.1 times higher risk of harboring X4 variants (multivariate P < .0001), regardless of CCR5 Delta 32 genotype. In multivariate analyses adjusting for baseline parameters, HIV-1 coreceptor use was not found to be a significant predictor of survival or treatment response. CONCLUSION: Baseline CD4 cell count, pVL, HIV-1 V3 sequence, and CCR5 Delta 32 genotype were the strongest determinants of CXCR4-using HIV-1 in this population. After adjustment for baseline parameters, the presence of X4 variants before initiation of highly active antiretroviral therapy was not independently associated with a poorer outcome of therapy.

Simplification of therapeutic drug monitoring for twice-daily regimens of lopinavir/ritonavir for HIV infection.

Cost and inconvenience limit the application of full 12-hour pharmacokinetic (PK) analysis for routine therapeutic drug monitoring of antiretroviral medications. We explore whether lopinavir (LPV) and ritonavir (RTV) exposures can be estimated with limited sampling for patients taking twice-daily LPV/RTV. One hundred and one PK profiles from 81 patients, most receiving salvage therapies including twice-daily LPV/RTV, were obtained for the analysis. After a minimum of 2 weeks on a stable regimen, blood was drawn immediately before and at 1, 2, 4, 6, 8, 10, and 12 hours after a timed medication dose. Plasma drug concentrations were determined by a validated HPLC-MS-MS assay. Peak concentrations, evening troughs, and AUC0-12 h were entered into linear and log10-log10 linear regression models to determine the best correlation with LPV and RTV plasma concentrations using a maximum of 2 time points. The accuracy and precision of PK parameter estimates of the resultant models were tested on data collected for an additional 25 patients. Twelve models using various combinations of 2 timed LPV concentrations afforded accurate (maximum % bias = -6.45) and precise (relative standard deviation < 15%) estimates for the LPV peak concentration or AUC0-12h. Four sets of 2 concentrations provided simultaneous estimates of both PK parameters, with the best estimates derived from data collected at 2 and 6 hours postdose. Evening trough concentrations were the best estimators of the daily nadir; however, no adequate substitute for collecting blood 12 hours postdose emerged from this analysis.

Antiretroviral concentrations in untimed plasma samples predict therapy outcome in a population with advanced disease.

This study was designed to examine the relationship between untimed antiretroviral concentrations measured in plasma samples collected for virus-load testing and response to highly active antiretroviral therapy. Plasma nonnucleoside reverse-transcriptase-inhibitor and protease-inhibitor concentrations were retrospectively measured in all virus-load plasma samples collected during the first year of therapy, for 122 patients in British Columbia, Canada, who initiated therapy between August 1996 and September 1999 and who had CD4 counts <50 cells/micro L. Drug levels were designated a priori as "low" if the concentrations were below the published Ctrough-SD. A single low drug level measured shortly after initiation of therapy (median, 6 weeks) is common (30%) and is predictive of both more-rapid immunological failure (P=.06) and failure to achieve virologic success during the first year of therapy (P=.01). These results may reflect incomplete adherence, since a strong association (P<.001) was found between low drug levels and an imperfect prescription-refill record (<95%).

Mitochondrial:nuclear DNA ratios in peripheral blood cells from human immunodeficiency virus (HIV)-infected patients who received selected HIV antiretroviral drug regimens.

Mitochondrial:nuclear DNA (mtDNA:nDNA) ratios in blood cells were investigated in relation to selected human immunodeficiency virus antiretroviral drug regimens. Patients (n = 214) continually received a regimen consisting of either (1) saquinavir (SAQ) + ritonavir (RTV) + either nevirapine (NVP) or lamivudine (3TC) (n=32) or (2) SAQ+RTV (+/- either NVP or 3TC) + either stavudine (d4T) (n=127), didanosine (ddI) (n=19), d4T+ddI (n=21), or zidovudine (ZDV) (n=15), for >/=4 months. NVP- or 3TC-only regimens were associated with median mtDNA:nDNA ratios that were significantly higher than those for ddI- and d4T+ddI-containing regimens (P<.01) but that were not significantly higher than those for d4T- or ZDV-containing regimens. Patients received thymidine analogue- and/or ddI-containing regimens for a shorter time (median, 14 vs. 24 months; P<.01). Because of survivor-bias effect, these results may represent a conservative estimate of these nucleosides' effect on mtDNA.