Health Care Costs After Genome-Wide Sequencing for Children With Rare Diseases in England and Canada

This cohort study investigates the association of diagnosis of rare diseases using genome-wide sequencing with health care casts among children in England and Canada.


100KGP Cohort
In England, patients were referred by their healthcare professionals to 100KGP and recruited from nine English hospitals. 1Patients were eligible if they were suspected to have rare diseases with a likely single gene or oligogenic cause and had not yet received a genetic diagnosis.Patients often had a history of genetic and non-genetic investigations, including single gene tests, karyotyping, single-nucleotide polymorphism arrays, multi-gene panels, or exome sequencing.Where feasible, parents or other family members were enrolled in 100KGP to support duo or trio testing.
The 100KGP classified EoE patients into epilepsy categories and ID patients into ID categories based on a combination of assessment criteria and clinical expertise.EoE patients were identified as belonging to epilepsy categories in the Genomics England dataset and having an age of onset up to and including 48 months old. 2 ID patients were identified as belonging to the ID categories in the Genomics England dataset.Patients not classified as having EoE or ID were excluded from the analysis.

CAUSES Cohort
In British Columbia, Canada, patients were referred to CAUSES by their physicians.Referring physicians included general practitioners, pediatric specialists and subspecialists, and medical genetics physicians from across the province, although medical genetics, metabolic disease/biochemical genetics, and neurology were the most common referring disciplines. 3,4atients were selected for research-based GWS through CAUSES if they were strongly suspected to have a single gene disorder, had both biological parents available for trio testing, and met at least one of the following additional criteria: previous sequential genetic testing, including chromosomal microarrays, single-gene or multi-gene tests, and first tier biochemical testing did not identify a genetic cause; condition exhibited extensive genetic heterogeneity; and/or family history was suggestive of a Mendelian single-gene disorder.Anticipated low diagnostic yield was an exclusion criterion.

BC Publicly Reimbursed GWS Cohort
Reimbursement of GWS occurs on a case-by-case basis in BC and requires submission of an application justifying the need for testing and expected implications for patient care.Patients were eligible to apply for publicly reimbursed GWS if they had a suspected genetic disorder, had completed prior consultations and in-province tests, and testing offered potential for clear patient benefit (e.g., through treatment change). 5

Canadian Data Sources
To assess cohort inclusion, information on genomic testing, patient demographics and phenotypic characteristics captured in departmental referrals and the BC Children's Hospital (BCCH) institutional EMR system (Cerner) was manually reviewed by four coders (JD, HC, ME, FM).Initially, patient records were reviewed in duplicate to assess concordance.Any coding discrepancies that emerged were resolved by the coders and when necessary, through consultation with the larger project team.Ineligible phenotypes were then excluded.Owing to the infeasibility of conducting extensive manual record reviews for a large cohort, we randomly selected 20% of CAUSES Clinic participants for consideration.Following initial review, we excluded 39 (34%) of CAUSES Clinic participants and 192 (62%) of clinical patients owing to ineligible phenotypes, location of residence (outside of BC), or denial of GWS requisition.
Coders accessed both the BCCH institutional EMR system and the BC province-wide EMR system (CareConnect) to comprehensively capture diagnostic services rendered during the study period.Data was gathered from first interaction with the public healthcare system (excluding normal birth-related interaction), until death or end of the study period (July 2019 in BC).In BC, data spanned up to 8 years prior and 3 years after GWS, although sample sizes were greatly reduced in the years furthest removed from GWS. Eligible diagnostic services included diagnostic imaging, physiological tests, specimen collection, genetic testing and laboratory testing.Coders abstracted GWS information pertaining to service dates, turnaround times, test type (e.g., whole exome or whole genome sequencing, singleton or trio) and GWS results (e.g., pathogenic variant, variant of uncertain significant (VUS), secondary findings) recorded by clinicians within patient EMRs.Patient characteristics, including sex, age, and geographic region of residence were determined based on clinician notes.Rural and urban locations were determined using Statistics Canada definitions for rural postal codes and small population centres. 6,7

Canadian Costing
Costing was performed using the Canadian Agency for Drugs and Technologies in Health's (CADTH) guidelines for the costing of healthcare resources in a Canadian setting. 8

NHS Reference Costs data
NHS Improvement (the administrative arm of the NHS performing a range of functions) regularly collects information from service providers who quantify their true costs.This information is published annually (the National Schedule of Reference Costs).This reference cost collection provides information on Healthcare Resource Group (HRG)-based unit costs that represent the unit costs to the NHS for providing defined services in a given financial year.These reference costs account for the direct costs of producing a healthcare service, indirect costs, and overheads.

Attaching costs to HES data
Reference costs were matched to HES data on an episode-by-episode basis.HES data were cleaned, then costs were allocated to each episode using specialised "Grouper" software version HRG4+ 2017/18.The software uses information on the episode to assign a HRG code.This includes condition and diagnostic information, treatment information, provider information and individual-specific information.National average unit costs from the National Schedule of Reference Costs 2017/18 were then linked to each episode-specific HRG.The same year was applied to all costs to avoid cost variation being attributed to year-to-year differences in reference cost procedures.

Inverse probability of censoring weighting
In Canadian analyses, we estimated inverse probability weights for each one-year time interval using Kaplan-Meier product limit estimates of probability of censoring. 12In the English analyses, one-year interval inverse probability weights were generated with logistic regression.Inverse probability weighting reduces estimation bias by recreating the sample population expected in the absence of censoring.

Regression model specifications
The following equation represents the regression model used for our pre-post analysis: (  ) = exp (  +   +   +   +   +   ) Where cit are costs for participant i in year t; αi are individual random effects; θ estimates the trend in costs over time, postt is a dummy variable denoting the post-GWS period; δ is the coefficient of interest and estimates the level change in costs post GWS, Xit is a vector of baseline covariates and β is vector of corresponding coefficients; γt are calendar year fixed effects; and εit is a standard error term capturing unexplained variation in the dependent variable.
In pre-post analysis, this equation was fit using mixed effects generalized linear models estimated by maximum likelihood, assuming a log link and Gamma distribution.An independent correlation structure was assumed for random effects.Pre-post estimates reflect the estimated marginal effects of the level change on predicted mean costs for the fixed portion of the model leaving values of all other covariates as observed.Associated standard errors and confidence intervals were estimated using the Delta method.
The following equation represents the regression model used for our difference-in-differences analysis: Where cit are costs for participant i in year t; α is an intercept term, (postt x diagnosedi) is the interaction between a dummy variable denoting the post period and a dummy variable denoting the diagnosed group (so is a binary indicator equal to 1 for a diagnosed patient after GWS and 0 otherwise), δ is the coefficient of interest and estimates the effect of diagnosis on costs, Xit is a vector of baseline covariates and β is vector of corresponding coefficients; θi capture group differences at baseline, γt are calendar year fixed effects; and εit is a standard error term capturing unexplained variation in the dependent variable.In difference-in-differences analysis, this equation was fit using linear models to ensure interpretability of the coefficient of interest. 13n England, the final pre-post and difference-in-differences model adjusted for: age (continuously specified using natural splines), length of diagnostic odyssey (linearly specified); and categorical covariates of gender, ethnicity, deprivation decile, and region.In Canada, final models were stratified according to GWS setting (only-in-research vs. publicly reimbursed).In a BC research setting, the final pre-post model adjusted for: outcome trends (linearly specified), number of comorbidities, age at GWS (continuously specified with squared term), and random effects.The difference-in-differences model adjusted: calendar year and group fixed effects, number of comorbidities, sex, age at GWS (continuously specified with squared term), sex, and year of diagnostic odyssey in which GWS was accessed.In a BC publicly reimbursed setting, the final pre-post model adjusted for: outcome trends, number of comorbidities, age at GWS (continuously specified with squared term), phenotype, sex, area of residence, and random effects.The difference-in-differences model adjusted for: calendar year and group fixed effects, number of comorbidities, sex, age at GWS (continuously specified with squared term), sex, phenotype, and area of residence. ©

eMethods Table: Canadian unit costs for diagnostic services
Unit costs of diagnostic services were identified based on: the Ministry of Health Medical Services Commission Payment Schedule 9 ; the Ministry of Health Schedule of Fees for Laboratory Services 10 ; the B.C. Ministry of Health MSP Fee-For-Service Payment Analysis 2016/2017 -2020/2021; 11 departmental tracking from BCCH, literature sources and publicly listed information from the following laboratories in Canada: Lifelabs Genetics, Pacific Fertility, InitioMedical, Prevention Genetics, and Canada Diagnostic.When public, published, or commercial prices for laboratory tests within Canada were not readily available, U.S. prices were researched on www.findlabtest.com.U.S. prices were converted into Canadian Dollars using a 2019 Bank of Canada conversion rate of CAD$1.32 per USD$1.00.Unit costs and corresponding sources are provided in the eMethods Table.All unit costs were converted to 2019 Canadian Dollars using Bank of Canada CPI inflation rates.
Digital, which is collated and known collectively as Hospital Episode Statistics (HES), wherein each unit of activity is referred to as an episode.HES offer between 12 to 22 years of secondary care history across different data sets.Data on admitted patient care (APC), outpatient appointments (OP), and emergency attendances (A&E) are available from 1997, 2003 and 2007, respectively, until the end of 2019.HES provides information on the age of individuals, the date of the episode, the length of stay, the treatment speciality under which healthcare was provided, inter alia.Each episode in outpatient care and admitted patient care is defined by ICD-10 codes (International Statistical Classification of Diseases and Related Health Problems) and OPCS-4 codes (Classification of Interventions and Procedures).

Difference-in-difference regression estimates for 2 years pre and 2 years post in English cohorts
2024 Weymann D et al.JAMA Network Open.

Logistic regression of probability of diagnosis from GWS, Canadian Cohorts
Few baseline differences across diagnosed and not diagnosed patients were statistically significant in either a research or publicly reimbursed Canadian setting.Research participants with more concomitant disorders were significantly more likely to be diagnosed from GWS (aOR: 1.38, p=0.02).

eTable 4: Logistic regression of probability of diagnosis from GWS, United Kingdom Cohorts
© 2024 Weymann D et al.JAMA Network Open.eFigure 2: Unadjusted annual total cost trajectory across cohorts © 2024 Weymann D et al.JAMA Network Open.Unadjusted for covariates or censoring, mean annual costs reported in 2019 $CAD or £GBP for complete cases only.

eTable 5: Unadjusted annual costs over 2 years pre and 2 years post period, Canada
Clinical Assessment of the Utility of Sequencing and Evaluation as a Service; GWS: genome-wide sequencing; SE: standard error Unadjusted for covariates or censoring, mean annual costs reported in 2019 $CAD or £GBP for complete cases only.