Inherently explainable deep neural network-based interpretation of electrocardiograms using variational auto-encoders

Background Deep neural networks (DNNs) show excellent performance in interpreting electrocardiograms (ECGs), both for conventional ECG interpretation and for novel applications such as detection of reduced ejection fraction and prediction of one-year mortality. Despite these promising developments, clinical implementation is severely hampered by the lack of trustworthy techniques to explain the decisions of the algorithm to clinicians. Especially, currently employed heatmap-based methods have shown to be inaccurate. Methods We present a novel approach that is inherently explainable and uses an unsupervised variational auto-encoder (VAE) to learn the underlying factors of variation of the ECG (the FactorECG) in a database with 1.1 million ECG recordings. These factors are subsequently used in a pipeline with common and interpretable statistical methods. As the ECG factors are explainable by generating and visualizing ECGs on both the model- and individual patient-level, the pipeline becomes fully explainable. The performance of the pipeline is compared to a state-of-the-art black box DNN in three tasks: conventional ECG interpretation with 35 diagnostic statements, detection of reduced ejection fraction and prediction of one-year mortality. Results The VAE was able to compress the ECG into 21 generative ECG factors, which are associated with physiologically valid underlying anatomical and (patho)physiological processes. When applying the novel pipeline to the three tasks, the explainable FactorECG pipeline performed similar to state-of-the-art black box DNNs in conventional ECG interpretation (AUROC 0.94 vs 0.96), detection of reduced ejection fraction (AUROC 0.90 vs 0.91) and prediction of one-year mortality (AUROC 0.76 vs 0.75). Contrary to state-of-the-art, our pipeline provided inherent explainability on which morphological ECG features were important for prediction or diagnosis. Conclusion Future studies should employ DNNs that are inherently explainable to facilitate clinical implementation by gaining confidence in artificial intelligence, and more importantly, making it possible to identify biased or inaccurate models.

Sex-Related Differences in Outcomes After Endovascular Treatment of Patients With Late-Window Stroke

Background and Purpose: Sex-related differences exist in many aspects of acute stroke and were mainly investigated in the early time window with conflicting results. However, data regarding sex disparities in late presenters are scarce. Therefore, we sought to investigate differences in outcomes between women and men treated with endovascular treatment in the late time window. Methods: Analyses were based on the SOLSTICE Consortium (Selection of Late-Window Stroke for Thrombectomy by Imaging Collateral Extent), which was an individual-patient level analysis of seven trials and registries. Baseline characteristics, 90-day functional independence (modified Rankin Scale score ≤2), mortality, and symptomatic intracranial hemorrhage were compared between women and men. Effect of sex on the association of age and successful reperfusion (final Thrombolysis in Cerebral Infarction 2b–3) with outcomes was assessed using multivariable logistic regression adjusted for age, National Institutes of Health Stroke Scale score, Alberta Stroke Program Early CT Score, time from onset to puncture, occlusion location, intravenous thrombolysis, and successful reperfusion, with interaction terms. Results: Among 608 patients treated with endovascular treatment, 50.5% were women. Women were older than men (median age of 72 versus 68 years, P =0.02) and had a lower prevalence of tandem occlusions (14.0% versus 22.9%, P =0.005). Workflow times were similar between sexes. Adjusted outcomes did not differ between women and men. Functional independence at 90 days was achieved by 127 out of 292 women (43.5%) and 135 out of 291 men (46.4%). Mortality at 90 days (54 [18.5%] versus 48 [16.5%]) and symptomatic intracranial hemorrhage (37 [13.3%] versus 33 [11.6%]) were similar between women and men. There was no sex-by-age interaction on functional outcomes. However, men had higher likelihood of mortality ( P interaction =0.003) and symptomatic intracranial hemorrhage ( P interaction =0.017) with advancing age. Sex did not influence the relation between successful reperfusion and outcomes. Conclusions: In this multicenter analysis of late patients treated with endovascular treatment, sex was not associated with functional outcome. However, sex influenced the association between age and safety outcomes, with men experiencing worse outcomes with advancing age.

Clinical profiles of Asians with NAFLD: A systematic review and meta-analysis

Introduction: NAFLD is increasingly prevalent in Asia, where people suffer more metabolic comorbidities at a lower body mass index (BMI), suggesting potential differences in their clinical profile. Therefore, we attempted to characterize the clinical profile of Asians with NAFLD via a meta-analytic approach. Methods: We searched Pubmed, EMBASE, and Cochrane databases from January 1, 2000 to January 17, 2019. Two authors independently reviewed and selected 104 articles (2,247,754 persons) that identified NAFLD in Asians and reported relevant data, especially BMI and ALT, and excluded individuals with other liver disease and excessive alcohol consumption. Individual patient-level data were obtained from seven cohorts in Asia to complement meta-analyzed data. Results: Overall, the mean age was 52.07 (95%CI:51.28-52.85) years with those from Southeast Asia (42.66, 95%CI: 32.23-53.11) being significantly younger. The mean BMI was 26.2 kg/m2, higher in moderate-severe vs. mild hepatic steatosis (28.3 vs. 25.7) patients and NFS ≥-1.455 vs. <-1.455 (27.09 vs. 26.02), with 34% having non-obese NAFLD. The mean ALT was 31.74 U/L, higher in NFS <-1.455 vs. ≥-1.455 (33.74 vs. 27.83), though no differences were found by obesity or steatosis severity. The majority of males (85.7%) and females (60.7%) had normal to minimally elevated ALT (1-1.5x 95% ULN). Individual patient-level data analysis (N=7,668) demonstrated similar results. Conclusion: About one-third of Asians with NAFLD were non-obese and the majority did not have markedly elevated ALT. Therefore, abnormal ALT or BMI are not recommended as a criterion for NAFLD screening in this population. Additionally, there were significant differences in the clinical profiles of NAFLD among the different regions of Asia.

Limitations of Muscle Ultrasound Shear Wave Elastography for Clinical Routine—Positioning and Muscle Selection

Shear wave elastography (SWE) is a clinical ultrasound imaging modality that enables non-invasive estimation of tissue elasticity. However, various methodological factors—such as vendor-specific implementations of SWE, mechanical anisotropy of tissue, varying anatomical position of muscle and changes in elasticity due to passive muscle stretch—can confound muscle SWE measurements and increase their variability. A measurement protocol with a low variability of reference measurements in healthy subjects is desirable to facilitate diagnostic conclusions on an individual-patient level. Here, we present data from 52 healthy volunteers in the areas of: (1) Characterizing different limb and truncal muscles in terms of inter-subject variability of SWE measurements. Superficial muscles with little pennation, such as biceps brachii, exhibit the lowest variability whereas paravertebral muscles show the highest. (2) Comparing two protocols with different limb positioning in a trade-off between examination convenience and SWE measurement variability. Repositioning to achieve low passive extension of each muscle results in the lowest SWE variability. (3) Providing SWE shear wave velocity (SWV) reference values for a specific ultrasound machine/transducer setup (Canon Aplio i800, 18 MHz probe) for a number of muscles and two positioning protocols. We argue that methodological issues limit the current clinical applicability of muscle SWE.

Evaluation of Psoriasis Area and Severity Index Thresholds as Proxies for Systemic Inflammation on an Individual Patient Level

<b><i>Background:</i></b> Psoriasis is a chronic and systemic inflammatory disease with a loss of up to 5 life years, which is thought to be reduced by biologic treatment. Disease severity and eligibility for systemic treatment are often based on the cutaneous psoriasis area and severity index (PASI) with a cut-off of 10 in several European countries. However, it is unclear how well this cut-off reflects systemic inflammation and, consequently, the risk for the development of comorbidity. <b><i>Objectives:</i></b> (1) To assess whether specific PASI thresholds, in particular PASI 10, predict elevated biomarkers of systemic inflammation and cardiovascular risk on an individual patient level. (2) To assess the association of PASI and psoriatic arthritis with biomarkers of systemic inflammation and cardiovascular risk. <b><i>Methods:</i></b> Retrospective cross-sectional study of 72 psoriasis patients without systemic treatment. <b><i>Results:</i></b> Overall, 68, 42, and 50% of patients had cardiovascular risk level neutrophil-to-lymphocyte ratio (NLR), C-reactive protein, and elevated platelet-to-lymphocyte ratio (PLR) values, respectively. The respective positive predictive values of PASI 10 were 70, 45, and 70. The performance of the optimal PASI cut-offs according to the Youden index was similarly weak. Subgrouping of patients with a PASI below 10 did not result in a considerably improved reflection of systemic inflammation. PLR was significantly higher in patients with moderate-to-severe compared to mild psoriasis and significantly correlated with PASI in patients with a PASI above 2 (<i>r</i>_s = 0.266, <i>n</i> = 64). NLR was significantly higher in patients with psoriatic arthritis. <b><i>Conclusion:</i></b> Specific PASI thresholds were not well suited to predict elevated biomarkers of systemic inflammation and cardiovascular risk on an individual patient level. Therefore, PASI, and possibly other purely cutaneous measures, may not be ideal as stand-alone parameters to define disease severity and eligibility for systemic treatment. Our results are relevant for the ongoing discussion on the definition of psoriasis severity and eligibility for systemic treatment. Further research addressing the added value of a set of biomarkers of systemic inflammation in the assessment of psoriasis severity would be desirable.

The Comparison of Kte-X19 to Current Standards of Care: A Pre-Specified Synthetic Control Study Utilizing Individual Patient Level Data from Historic Clinical Trials (SCHOLAR-3)

Background: The ZUMA-3 trial of KTE-X19 in relapsed/refractory (r/r) adult B-cell precursor acute lymphoblastic leukaemia (B-ALL) utilizes a single arm design. To contextualize these results, a synthetic control (SC) study derived from individual patient-level data (IPD) sampled from historical clinical trials (CTs) was constructed. Aims: To compare the overall survival (OS) and objective complete response rate at week 24 (OCR24) results of the ZUMA-3 pivitol study using a matched (SC) derived from IPD from CTs. Methods: This study had two distinct phases: firstly, CTs from which patients were sampled were identified using the medidata enterprise data store (MEDS). The second phase of this study constructed SCs to the ZUMA-3 population, one SC for patients naïve to blinatumomab (blin) or inotuzuamab (ino) therapy (SCA-1) and one in blin or ino pre-treated patients (SCA-2). All analysis were pre-specified and conducted by an external third party to Kite Pharma. For the trial identification phase, the MEDS database was searched to identify CTs that had inclusion and exlusion criteria congruent with ZUMA-3 and treatment assignments representative of current standards of care inclusive of blin, ino or chemotherapy regimens. Once appropriate CTs had been identified the endpoints were re-engineerd to match the same defintions as the ZUMA-3 trial. For the SC construction phase, propensity score (PS) matching was used. The PS was derived as a function of the number of previous lines of therapy, prior allo-SCT, age, sex, ECOG, Philadelphia chromosome status, percentage bone marrow blasts and extramedullary disease. For outcomes analysis, time-to-event endpoints of interest were analyzed using the Kaplan-Meier method and compared using a Cox proportional hazard regression model. OCR rate was described through crude incidence rates and corresponding 95% CI. ln addition, an odds ratio together with associated 95% CI and 2-sided p-value were estimated from a logistic regression model. Results A total of 20 SCA-1 patients were matched to 20 patients from ZUMA-3 and a total 20 SCA-2 patients were matched to 29 patients from ZUMA-3. Analysis of the SCA-1 cohort shows an OCR24 of 85% (95% CI 62.1%, 96.8%) in the Zuma-3 patients and 35.0% (95% CI 15.4, 59.2) among propensity matched controls. This corresponds to an OR of 10.5 (95% CI 2.3, 48.7; p-value 0.0031). No OCR24 data was available for SCA-2. A post hoc analysis was conducted in order to further contextualize the OCR24 rate. ZUMA-3 patients irrespective if they had been pre-treated with blin or ino were matched to patients from HCTs who were previously naïve to blin or ino therapy (SCA-3). The OCR24 rate in the ZUMA-3 arm was 69.8% (95% CI 55.7%, 81.7%) while for SCA-3 patients was 35.8% (95% CI 23.1%, 50.2%) meaning that ZUMA-3 patients had 4.1 times higher odds of achieving OCR in comparison to SCA-3 patients (95% CI 1.8, 9.3) p-value 0.0009. The comparison of overall survival (OS) between all matched ZUMA-3 and all SCA patients demonstrated a significantly higher median OS of 18.20 months (95% CI 12.22, NE months) for patients in ZUMA-3 versus 5.49 months (95% CI 3.32, 9.23 months) in SCA-3. A cox regression model showed that ZUMA-3 patients had a 64% lower risk of death with a hazard ratio 0.36 (95% CI 0.20, 0.66) p-value 0.0005. Conclusion: This comparative study demonstrated a clinically relevant improvement of OCR24 and OS following KTE-X19 vs available therapies and provides strong evidence for its use in adult patients with R/R B-ALL Disclosures Shah: Jazz Pharmaceuticals: Research Funding; Servier Genetics: Other; BeiGene: Consultancy, Honoraria; Incyte: Research Funding; Kite, a Gilead Company: Consultancy, Honoraria, Other: Expenses, Research Funding; Pharmacyclics/Janssen: Honoraria, Other: Expenses; Acrotech/Spectrum: Honoraria; Pfizer: Consultancy, Other: Expenses; Amgen: Consultancy; Novartis: Consultancy, Other: Expenses; Bristol-Myers Squibb/Celgene: Consultancy, Other: Expenses; Precision Biosciences: Consultancy; Adaptive Biotechnologies: Consultancy. Faghmous: Kite A Gilead Company: Current Employment, Current equity holder in publicly-traded company. Whitmore: Kite, A Gilead Company: Current Employment, Current equity holder in publicly-traded company. Masouleh: GSK: Current equity holder in publicly-traded company; Kite, A Gilead Company: Current Employment, Current equity holder in publicly-traded company; Immatics: Current equity holder in publicly-traded company; Roche: Current equity holder in publicly-traded company; Bristol-Myers Squibb: Current equity holder in publicly-traded company; Novartis: Current equity holder in publicly-traded company. Xu: Gilead Sciences: Other: stock or other ownership ; Kite, A Gilead Company: Current Employment.

Comparison of Clinical Outcomes Among Patients with Relapsed/Refractory Follicular Lymphoma Treated with Tisagenlecleucel in the Elara Trial Versus a Real-World External Control Arm of Patients Treated with Standard of Care

Introduction: In the single-arm phase 2 ELARA trial (NCT03568461), tisagenlecleucel demonstrated efficacy and favorable safety profile in patients with relapsed/refractory (r/r) follicular lymphoma (FL) after ≥ 2 lines of prior therapy, including in high-risk sub-populations (Schuster, et al. ASCO 2021). To contextualize these results, we performed a retrospective non-interventional study to compare the efficacy of tisagenlecleucel seen in the single-arm ELARA trial with the standard-of-care (SoC) using individual patient-level data from the US Flatiron Health Research Database (FHRD). FHRD is a database derived from electronic health records from over 280 cancer clinics. The objective was to assess the effect of prescribing tisagenlecleucel vs SoC in patients who participated in ELARA. Methods: Individual patient-level data from FHRD were used to create an external control arm to carry out an indirect comparison with the ELARA trial. Eligible inclusion and exclusion criteria from ELARA were applied to the external control arm. A single eligible line was selected using propensity scores when patients were qualified at multiple lines. Key prognostic factors including age, race, gender, number of prior treatment lines, group stage at initial FL diagnosis, number of months between initial FL diagnosis and indication of index treatment, double refractoriness, and disease progression within 24 months were included in a propensity score model to reduce confounding due to systematic differences in ELARA patients from FHRD patients at baseline for the selected line. Weighting by odds of receiving tisagenlecleucel was used to estimate the average treatment effect on progression-free survival (PFS), overall survival (OS), time to next treatment (TTNT), overall response rate (ORR), and complete response rate (CRR). The rates and difference in rates were calculated for CRR and ORR. Kaplan-Meier (KM) analysis and Cox proportional hazards model were used to analyze all time-to-event endpoints. 95% confidence interval (CI) was calculated using bootstrapping. Data from the first 24 months after enrollment in ELARA or after treatment in FHRD were used for the follow-up period in the time to event endpoints, as few patients in ELARA trial had &gt; 24 month data (Figure). Results were reported based on the post-weighting sample by incorporating a weight factor in the above analyses. A series of sensitivity analyses were conducted to assess the robustness of the primary analysis. Results: As of Mar 29, 2021, 98 patients were enrolled in the ELARA trial, of which 97 were included in this indirect comparison (median follow-up, 15 months). In the FHRD cohort (data cut-off, Jun 30, 2020), 98 patients with ≥ 3 treatment lines who met the ELARA eligibility were included (median follow-up, 14 months in the post-weighted sample) (Table 1). In the ELARA vs FHRD cohorts, after applying weighting by odds, the ORR was 85.6% vs 58.1%, and the CRR was 69.1% vs 17.7%. The difference in CRR (51.4%; 95% CI: 21.2, 68.8) was clinically meaningful (Table 2). The median TTNT or death was not reached in the ELARA cohort and was 19.0 months in the FHRD cohort after weighting (HR = 0.34 [95% CI: 0.15, 0.78]) (Table 2). The median OS was not reached for both ELARA and FHRD cohorts in the first 24-month period. KM estimate of OS at 12 months was 96.6% in the ELARA cohort and 84.5% in the FHRD cohort, post weighting. The estimated 59% risk reduction was in favor of tisagenlecleucel over SoC (hazard ratio [HR] = 0.41 [95% CI: 0.11, 1.47]) (Table 2). The median PFS was not reached in the ELARA cohort and was 9.9 months in the FHRD cohort, after weighting. In the ELARA vs FHRD cohorts, the 12-month PFS was 73.2% vs 41.8%, with a HR of 0.45 indicating a 55% reduction in the risk of progression and death with tisagenlecleucel vs SoC. The median PFS considering new anti-cancer therapy as an event was not reached for ELARA and was 9.9 months for FHRD; the estimated probability of being progression-free at 12 months was 70.5% in the ELARA cohort and 39.4% in the FHRD cohort (Table 2). Sensitivity analyses results were consistent with that of primary analysis. Conclusion: In the weighted analyses with adjustment for baseline prognostic factors, there was a consistent trend towards greater CRR, TTNT, OS and PFS in favor of tisagenlecleucel vs SoC in patients with r/r FL. These results support the clinically meaningful treatment benefit of tisagenlecleucel observed in the ELARA trial. Figure 1 Figure 1. Disclosures Hao: Novartis: Current Employment. Hsu: Novartis: Consultancy. Parzynski: Novartis: Consultancy. Lobetti Bodoni: Spouse: Amgen: Honoraria, Membership on an entity's Board of Directors or advisory committees, Speakers Bureau; Spouse: Janssen: Honoraria, Membership on an entity's Board of Directors or advisory committees, Speakers Bureau; Spouse: Celgene: Honoraria; Spouse: Harlcok Healthcare: Current holder of individual stocks in a privately-held company; Spouse: Takeda: Consultancy, Honoraria, Speakers Bureau; Spouse: NHS: Ended employment in the past 24 months; Spouse: Sanofi: Consultancy, Membership on an entity's Board of Directors or advisory committees; Spouse: F. Hoffmann-La Roche: Current Employment, Current equity holder in publicly-traded company; Gilead: Other: Travel sponsorship in June 2019; Novartis: Current Employment, Current equity holder in publicly-traded company. Degtyarev: Novartis: Current Employment, Current equity holder in publicly-traded company. Hampson: Novartis: Current Employment. Masood: Novartis: Current Employment, Current holder of stock options in a privately-held company. Wu: Novartis: Consultancy. OffLabel Disclosure: Tisagenlecleucel (Kymriah) an autologous CD19-directed CAR-T-cell therapy, has been approved for children and young adults with relapsed/refractory (r/r) acute lymphoblastic leukemia and, adults with r/r diffuse large B-cell lymphoma.

A Novel Systems Model to Simulate the Dynamics for the Analytical Study of the Proliferation of COVID-19

In this paper, we present a systems model for COVID-19 called ‘Multilevel Integrated Model with a Novel Systems Approach’ or MIMANSA for short. This model goes to the individual patient level and mimics small steps in the process of virus spread. Despite many models simulating the growth of COVID-19 cases, it is rare to see a comprehensive model that takes into account the transmissibility of variants, the prevalence of multiple variants, growth of the variants, percentage of vaccinated population, exposure rate, infection rate, silent carrier rate, secondary infection rate, mask usage, and people mobility all in one model.We begin by categorizing the in-person social interactions of an individual into three areas: household, workplace, and public places. With each interaction, the virus spreads from an infectiousperson to a healthy individual. We build the model, one level at a time, covering daily person-to-person interactions. Further, MIMANSA forms a new layer of network for every new day. These layers form a part of the virus proliferation network.MIMANSA models the virus incubation period using a Weibull distribution. Once the network starts building, the virus growth can be curtailed only through increased vaccinations or through non-pharmaceutical interventions such as mask usage, reduced mobility, and/or quarantine.MIMANSA takes the mobility data, mask usage data, variant prevalence data, and vaccination data as inputs. Despite the model being intricate, it uses only 5 parameters for training. Once the model is trained, it can be slightly adjusted by the daily environmental variable. This variable corrects for day to events as well as varying environmental conditions in a given location.We present the results of the training and validation for the USA, California, and the UK. It is seen that during the validation stage, the model accuracy is within 2%. Further, projections are made for about 3 weeks. In the end, we have presented the results of a study on the effect of vaccination on the number of COVID cases in the USA. It shows how MIMANSA can be used for studying the impact of multiple scenarios. Additionally, the model is not only useful for making predictions in the number of cases but also useful as an educational aid for explaining the proliferation network of the virus in real life. Although MIMANSA is originally developed for the SARS-CoV-2, it can be modified to study the spread of any other virus, and in any region.

Interobserver Agreement of PD-L1/SP142 Immunohistochemistry and Tumor-Infiltrating Lymphocytes (TILs) in Distant Metastases of Triple-Negative Breast Cancer: A Proof-of-Concept Study. A Report on Behalf of the International Immuno-Oncology Biomarker Working Group

Patients with advanced triple-negative breast cancer (TNBC) benefit from treatment with atezolizumab, provided that the tumor contains ≥1% of PD-L1/SP142-positive immune cells. Numbers of tumor-infiltrating lymphocytes (TILs) vary strongly according to the anatomic localization of TNBC metastases. We investigated inter-pathologist agreement in the assessment of PD-L1/SP142 immunohistochemistry and TILs. Ten pathologists evaluated PD-L1/SP142 expression in a proficiency test comprising 28 primary TNBCs, as well as PD-L1/SP142 expression and levels of TILs in 49 distant TNBC metastases with various localizations. Interobserver agreement for PD-L1 status (positive vs. negative) was high in the proficiency test: the corresponding scores as percentages showed good agreement with the consensus diagnosis. In TNBC metastases, there was substantial variability in PD-L1 status at the individual patient level. For one in five patients, the chance of treatment was essentially random, with half of the pathologists designating them as positive and half negative. Assessment of PD-L1/SP142 and TILs as percentages in TNBC metastases showed poor and moderate agreement, respectively. Additional training for metastatic TNBC is required to enhance interobserver agreement. Such training, focusing on metastatic specimens, seems worthwhile, since the same pathologists obtained high percentages of concordance (ranging from 93% to 100%) on the PD-L1 status of primary TNBCs.

Impact of COVID-19 on telepsychiatry at the service and individual patient level across two UK NHS mental health Trusts

BackgroundThe effects of COVID-19 on the shift to remote consultations remain to be properly investigated.ObjectiveTo quantify the extent, nature and clinical impact of the use of telepsychiatry during the COVID-19 pandemic and compare it with the data in the same period of the 2 years before the outbreak.MethodsWe used deidentified electronic health records routinely collected from two UK mental health Foundation Trusts (Oxford Health (OHFT) and Southern Health (SHFT)) between January and September in 2018, 2019 and 2020. We considered three outcomes: (1) service activity, (2) in-person versus remote modalities of consultation and (3) clinical outcomes using Health of the Nation Outcome Scales (HoNOS) data. HoNOS data were collected from two cohorts of patients (cohort 1: patients with ≥1 HoNOS assessment each year in 2018, 2019 and 2020; cohort 2: patients with ≥1 HoNOS assessment each year in 2019 and 2020), and analysed in clusters using superclasses (namely, psychotic, non-psychotic and organic), which are used to assess overall healthcare complexity in the National Health Service. All statistical analyses were done in Python.FindingsMental health service activity in 2020 increased in all scheduled community appointments (by 15.4% and 5.6% in OHFT and SHFT, respectively). Remote consultations registered a 3.5-fold to 6-fold increase from February to June 2020 (from 4685 to a peak of 26 245 appointments in OHFT and from 7117 to 24 987 appointments in SHFT), with post-lockdown monthly averages of 23 030 and 22 977 remote appointments/month in OHFT and SHFT, respectively. Video consultations comprised up to one-third of total telepsychiatric services per month from April to September 2020. For patients with dementia, non-attendance rates at in-person appointments were higher than remote appointments (17.2% vs 3.9%). The overall HoNOS cluster value increased only in the organic superclass (clusters 18–21, n=174; p<0.001) from 2019 to 2020, suggesting a specific impact of the COVID-19 pandemic on this population of patients.Conclusions and clinical implicationsThe rapid shift to remote service delivery has not reached some groups of patients who may require more tailored management with telepsychiatry.