Description
Hong Kong CDARS / HADCL Registry Study
A risk-adjusted outcomes study built on Hong Kong Hospital Authority data — either the Clinical Data Analysis and Reporting System (CDARS), the territory-wide electronic health record covering more than 90% of the population since 1993, or the newer Data Collaboration Laboratory (HADCL) big-data platform. Both are administrative and clinical databases, not research-designed cohorts: diagnosis coding drives case definition, key covariates are missing in clinically meaningful patterns, and the two systems differ in what they let you extract and how. This service is built around those facts, not around the assumption that a large database makes the statistics simple.
Read This First: Data Access
You must hold the data access yourself. We do not request CDARS or HADCL data on your behalf, and we cannot. CDARS access generally requires a Hospital Authority-affiliated clinician or an approved research collaboration with ethics approval from the relevant Institutional Review Board; HADCL access is limited to researchers at Hong Kong’s eight UGC-funded universities, requires its own application and ethics approval, and is worked on SAS Viya from an authorized terminal — the platform does not let you export a raw dataset, only derived analytic output. Our role starts once you have an approved protocol and either an extracted dataset (CDARS) or authorized access to run the analysis (HADCL). If you are not yet sure which system fits your question, or whether your access route is settled, ask before you buy — see below.
When This Is the Right Study
CDARS and HADCL are the right instrument for territory-wide questions about disease incidence, prevalence, drug safety, procedure outcomes, and healthcare utilisation in a population where the Hospital Authority provides the overwhelming majority of care — the design behind the large published literature on conditions from gout to COPD-associated lung cancer to vancomycin-associated kidney injury. They are the wrong instrument for anything that depends on care delivered outside the public system, on patient-reported outcomes no clinical encounter captured, or on a subgroup too small for the coding-accuracy noise in ICD-based case definition to average out.
How the Analysis Is Actually Done
- Case definition built and defended, not assumed — ICD-9 or ICD-10 diagnosis and procedure codes are mapped to your case definition explicitly, with the algorithm stated and, where prior validation studies exist for that condition in CDARS, cited against them. Positive predictive value for CDARS diagnostic coding has been reported above 90% for several conditions in prior validation work; it is not uniform across diagnoses, and we say so rather than assume it.
- Coding-driven missingness treated as informative — administrative databases do not miss data at random. Where a covariate’s absence correlates with disease severity or care pathway, complete-case analysis biases the estimate; we examine the pattern before choosing how to handle it.
- Territory-wide clustering and secular trend checked — care patterns and coding practice both drift over a multi-year window. Where the study spans several years, we test whether an apparent trend is a real effect or a composition/coding-practice artefact before reporting one.
- Modelling matched to the question — Cox proportional hazards and competing-risk regression for time-to-event outcomes, inverse probability of treatment weighting or propensity matching for comparative drug/procedure questions, logistic regression with the covariate set the literature in your condition actually uses.
- Linkage handled explicitly where relevant — studies using the Hong Kong Death Registry or other linked data have the linkage method and its limitations stated in the methods, not left implicit.
What You Receive
- Complete manuscript — written for your target journal, structured the way HK-authored CDARS/HADCL studies are expected to read
- Publication-quality tables and figures — cohort flow diagram, adjusted-effect forest plot or Kaplan-Meier/cumulative-incidence curves as the outcome requires
- Full, runnable analysis code — R, Stata, or SAS, matching whichever platform your data extraction used, commented, including the case-definition and cleaning steps
- Completed STROBE checklist
- Analysis log — case definition, every exclusion, every derived variable, and the missingness/sensitivity analysis specification
- One round of author-requested revisions
What Is Not Included
We do not obtain or hold CDARS or HADCL data, and we cannot handle identifiable patient information — CDARS extracts and HADCL outputs are anonymised, and that is what we work with. We do not ghost-write authorship; you set the research question and own the clinical interpretation and claims. We do not submit on your behalf, do not guarantee acceptance, and peer-review response is a separate service.
When We Will Tell You Not to Buy This
If your question needs data from private-sector care — CDARS and HADCL only see the public system, which is the large majority but not all of Hong Kong healthcare. If your outcome needs patient-reported measures no clinical encounter captured. If your subgroup of interest has too few coded cases for the modelling above to converge sensibly. And if your data access is not yet settled, we say so and point you back to the application process rather than take the order.
Timeline
Standard delivery is 15 business days from receipt of the extracted dataset (CDARS route) or your first authorized HADCL analysis session output, and an agreed analysis plan.
Common Questions
Can you help me choose between CDARS and HADCL for my question? Yes — often the first useful conversation to have, before either application goes in.
Can you work from a dataset a Hospital Authority collaborator already extracted for me? Yes, provided you can confirm your data use agreement or ethics approval covers external analytic support.
I already have the statistical output and just need the paper written. That’s a narrower, less expensive service — see Medical Manuscript Writing — Data to Publication.
Multi-database or Greater Bay Area linkage studies? Quoted individually.
