Protocol Generator · DA

Multi-national descriptive analysis protocol

How common a disease is, how a drug is used, who dies — counted the same way in every database so the numbers can be put side by side. Descriptive work is where a causal study gets its denominator, and where most cross-country comparisons quietly go wrong.

← All designs

1 · Study identification

Study title Short ID Version · Date Lead investigator(s)

2 · Research question

Population What we measure Time window

3 · Sites

Tick every database (not just country). Feature chips show what each carries; for a descriptive protocol the denominator + the metric features need to match (e.g. cause-specific mortality needs Mortality linkage; DDD needs Rx; cancer-survivor cohorts need Cancer).

Legend Rx Prescriptions Dx Diagnoses Lab Lab results Img Imaging Mor Mortality linkage Can Cancer-registry linkage Preg Pregnancy / birth linkage Gen Genomics / biobank linkage

Taiwan

CGRD RxDxLabImg NHIRD RxDxLabImgMorCanPreg TWB LabGen

South Korea

HIRA RxDx NHIS-NHID RxDxLabMorCan

Japan

JMDC Claims Database RxDxLab Medical Data Vision (MDV) Database RxDxLab NDB RxDx

Hong Kong

CDARS RxDxLabImgMorCan

Australia

PBS Rx VLHD RxDxMor

Finland

FinReg RxDxMorCanPreg

United Kingdom

CPRD RxDxMorCanPreg ISD RxDxLabMorCanPreg THIN RxDx UKB LabImgMorCanGen

Netherlands

PHARMO Database Network RxDxLabMorCan

United States

20% sample RxDxMorCan Medicaid RxDxMor

Others (one per line)

⚠ Feasibility heads-up; not a blocker

Your chosen metrics or text imply data features that some selected sites don't carry. Cross-country descriptive studies often accommodate this by reporting which sites contribute to which metric; make that explicit in the per-site analytic plan.

4 · Population (denominator)

Population; name Operational definition Code set / criteria Inclusion criteria (one per line) Exclusion criteria (one per line) Index date / observation-start rule Study calendar period

5 · Numerator phenotypes

What you are counting in the numerator. Pick from each library as relevant; drug exposures for utilisation metrics, outcomes for incidence / prevalence, mortality causes for cause-specific mortality. Free-text fields below for codes not in the libraries.

Drug exposure; name Drug exposure; code set Outcome; name Outcome; code set Outcome; operational definition Cause of death; name Cause of death; code set

6 · Output metrics; what to report

Tick every metric to report. Each ticked metric gets its own section in the protocol output (Markdown + Word), with operational definition, formula, and unit. Mix-and-match freely; most descriptive multi-database papers report 3-6 metrics.

Occurrence

Incidence rate New cases over person-time at risk. cases / 1000 person-years Period prevalence Cases ÷ cohort, over a calendar window. % of cohort in window Point prevalence Cases ÷ cohort on a single calendar date (annual / mid-year typical). % at index date

Mortality

All-cause mortality rate Requires mortality linkage at the site. deaths / 1000 person-years Cause-specific mortality Uses cause-of-death codes; see library. deaths / 1000 person-years

Utilisation

DDD / 1000 inhabitants / day WHO ATC/DDD index dose; classic drug-utilisation metric. DDD/1000/d Initiation rate Incident drug exposure rate. new users / 1000 person-years Persistence / discontinuation Time-to-discontinuation (Kaplan-Meier description). median days; % at 1 y Switching pattern (Sankey) First-line → second-line → … transitions. flows between drug classes Concomitant use Proportion using multiple drug classes within a window. % on ≥2 classes

Adjustment

Age-standardised rates Direct or indirect standardisation to a reference population. per 1000 (standardised)

Trends

Joinpoint / segmented trend Annual percent change, with the joinpoints located by grid search and their number chosen by a separate selection rule (Kim 2000). APC / AAPC Trend-in-trend (Ji 2017) Advanced causal design; strata are quantiles of each person's predicted probability of exposure. Better done in its own template. odds ratio, exp(β₁)

7 · Stratification

Report each ticked metric stratified by these variables (in addition to overall).

Age (5-year bands) Age (10-year bands) Age (≥65 vs <65) Sex (male vs female) Calendar year Country / site Index year stratum (5-year blocks) Urbanicity (urban / rural) Socioeconomic status / education Insurance plan / payer

Custom strata (one per line)

8 · Age-standardisation (active when "Age-standardised rates" is ticked above)

Reference population WHO 2000 is the most common for cross-country comparisons; ESP 2013 is preferred in European contexts. Method Age bands for standardisation

9 · Joinpoint / trend analysis (active when "Joinpoint trend" is ticked)

Trend year range Max joinpoints Joinpoint fitting and selection Report

These are two stages, not two alternatives. A grid search locates candidate joinpoints; a separate rule then decides how many there are, and the two are used together — Kim HJ et al (Stat Med 2000) do exactly that. This control previously offered "Grid search" labelled as the NCI Joinpoint Regression Program's default, which it has never been: that program's default rule for choosing the number of joinpoints was the permutation test through v4.9 and is a weighted BIC from v5.0. Whichever you pick, name the software and version in the protocol — "the software default" is not a fixed specification.

Trend-in-trend (Ji X, Small DS, Leonard CE, Hennessy S. Epidemiology 2017;28(4):529-536) is not a descriptive metric — it is a causal design. It stratifies the population by each person's predicted probability of exposure (the cumulative probability of exposure, CPE) from a stage-1 logistic model, then compares outcome trends across those strata. If you tick it in Section 6 the protocol carries a methods stub only; use the dedicated trend-in-trend builder if it is your central analytic question.

10 · Statistical analysis

Rate / proportion confidence intervals Cross-site comparison Site contribution Suppression / small-cell rule

11 · Sensitivity analyses

Alternative denominator definition (mid-year vs annual average vs Lexis triangle) Alternative time-window length (calendar-year vs fiscal-year vs rolling 12-month) Restrict to incident users only (vs all users) Restrict to ≥1 year of baseline observation Negative-control outcome; should show null / flat trend Exclude index-year transition (washout 30 / 60 / 90 d) Alternative case-finding rule (≥1 inpatient vs ≥2 outpatient vs algorithm) Sub-divide by data-source quality (full vs partial linkage) Limit to sites with ≥X years of follow-up to reduce small-sample noise

Custom sensitivity analyses (one per line)

12 · Execution & timeline

Architecture

Timeline preset

Milestones (one per line)

13 · Ethics, dissemination, references (optional)

Ethics Dissemination Key references (one per line)

Lai ECC, Stang P, Yang YHK, Kubota K, Wong ICK, Setoguchi S. International multi-database pharmacoepidemiology: potentials and pitfalls. Curr Epidemiol Rep 2015;2(4):229-238. doi:10.1007/s40471-015-0059-z
Lai EC, Man KK, Chaiyakunapruk N, et al. Brief report: databases in the Asia-Pacific region: the potential for a distributed network approach. Epidemiology 2015;26(6):815-820. doi:10.1097/EDE.0000000000000325 (PMID 26133022)
Kim HJ, Fay MP, Feuer EJ, Midthune DN. Permutation tests for joinpoint regression with applications to cancer rates. Stat Med 2000;19(3):335-351. doi:10.1002/(SICI)1097-0258(20000215)19:3<335::AID-SIM336>3.0.CO;2-Z
Ji X, Small DS, Leonard CE, Hennessy S. The Trend-in-trend research design for causal inference. Epidemiology 2017;28(4):529-536. doi:10.1097/EDE.0000000000000579
Wagner AK, Soumerai SB, Zhang F, Ross-Degnan D. Segmented regression analysis of interrupted time-series studies in medication use research. J Clin Pharm Ther 2002;27(4):299-309. doi:10.1046/j.1365-2710.2002.00430.x
WHO. Age-standardisation of rates: a new WHO standard. GPE Discussion Paper 31, 2001 (Ahmad OB et al).
Lee H, Baek YH, Kim JH, Liao TC, Lau WCY, Man KKC, et al. Trends of polypharmacy among older people in Asia, Australia and the United Kingdom: a multinational population-based study. Age Ageing 2023;52(2):afad014. doi:10.1093/ageing/afad014 (PMID 36821644)
Hsieh CY, Su CC, Shao SC, Sung SF, Lin SJ, Kao Yang YH, Lai EC. Taiwan's NHIRD: past and future. Clin Epidemiol 2019;11:349-58.
Shao SC, Chan YY, Kao Yang YH, Lin SJ, Hung MJ, Chien RN, Lai CC, Lai EC. The Chang Gung Research Database. Pharmacoepidemiol Drug Saf 2019;28(5):593-600.

Live preview

Auto-saved to this browser

Consistency checks · descriptive 防呆

A descriptive study fails quietly: the numerator and the denominator get counted over different people, and the rate that comes out looks perfectly reasonable. These checks look for that.

Study design diagram

Descriptive designs aggregate over the calendar study period; this per-person schematic shows the baseline look-back and the person-time / measurement window relative to cohort entry.

Phenotype library

Pick a phenotype

Each entry says whether a study has measured how well it identifies what it names. Only a validation study is added to your protocol's reference list, and it stays there for as long as this entry's name is the one in the form — so if you edit the code set or the definition underneath it, check that the citation still covers what you ended up with. Everything else here is a note you cite yourself.