Target Trial Emulation Protocol and Report Template (TARGET 2025)

Target trial emulation: causal question, data sources, target trial vs emulation table, time zero, estimands, assumptions, weighting, bias checks, TARGET map

अनुसंधान विज्ञानपूर्णTARGET 2025 (21 items; JAMA 2025;334(12):1084-1093; BMJ 2025;390:e087179)target trial emulation mathअपडेट किया गया 29 अक्टू॰ 2026

एडिटर में खोलें.md डाउनलोड करें

उपयोग कैसे करें: Use it when you estimate the causal effect of a treatment strategy from observational data. Fill the target trial protocol table and time zero first, before looking at outcomes. The main pitfall is misaligned eligibility, assignment and start of follow-up, which creates immortal time bias.

प्रीव्यू

Target Trial Emulation Protocol and Report: [Short title of the study]

Protocol and report template for an observational study that estimates the causal effect of an intervention by explicitly emulating a hypothetical randomised trial (the target trial). The report follows the 21 items of the TARGET guideline (Transparent Reporting of Observational Studies Emulating a Target Trial; Cashin et al., JAMA 2025;334(12):1084-1093 and BMJ 2025;390:e087179), organised in six sections; item numbers appear in square brackets, for example [6c]. TARGET covers emulations of a parallel-group, individually randomised target trial with adjustment for baseline confounders; general observational checklists such as STROBE and RECORD do not capture these target trial items. Replace every [bracketed] value and every row marked Example. This is a document format, not clinical, epidemiological or statistical advice; involve a statistician or epidemiologist with causal inference experience.

Document control

Field	Value
Document ID	[TTE-ACRONYM-001]
Causal question	[Effect of initiating X versus Y on outcome Z in population P]
Data source(s)	[Database names and years]
Protocol registration	[Registry ID / OSF DOI / "not registered"]
Ethics reference	[Committee and number, or exemption]
Version	[0.1]
Status	[Draft protocol / Protocol frozen / Analysis / Report drafted / Approved]
Owner	[@principal-investigator]
Approvers / reviewers	[@study-statistician], [@clinical-lead], [@data-custodian]
Last updated	YYYY-MM-DD
Classification	[Confidential until publication]

Approvals

Role	Name	Decision	Date
Principal investigator	[@name]	[Protocol frozen]	YYYY-MM-DD
Study statistician	[@name]	[Analysis plan approved]	YYYY-MM-DD
Clinical lead	[@name]	[Treatment strategies approved]	YYYY-MM-DD
Data custodian	[@name]	[Aggregate results approved for release]	YYYY-MM-DD

Revision history

Version	Date	Author	Change summary
0.1	YYYY-MM-DD	[@name]	First draft of target trial protocol
1.0	YYYY-MM-DD	[@name]	Protocol frozen before outcome analysis
1.1	YYYY-MM-DD	[@name]	Example: Permissible refill gap changed from 60 to 90 days after feasibility counts (no outcome data seen), see DEV-01
2.0	YYYY-MM-DD	[@name]	Report drafted

Title and abstract
Introduction
Data sources
Target trial protocol and emulation
Design-related biases
Results
Discussion
Other information
TARGET traceability
Glossary and references

1. Title and abstract

The title and abstract identify the design and summarise the study.

Title: [Initiating X versus Y] and [outcome] in [population]: a target trial emulation using [data source]

Heading	Content	TARGET
Design	Observational study emulating a target trial	[1a]
Data source	[Name, country, years]	[1b]
Target trial and emulation	[Eligibility, strategies, time zero, outcome, estimand]	[1a]
Key assumptions	[No unmeasured baseline confounding, positivity, consistency]	[1c]
Results	Example: Among 31,381 eligible adults (12,406 initiating class A, 18,975 initiating class B), the weighted 5-year risk of hospitalisation for heart failure was 6.2% with class A and 7.4% with class B: risk difference -1.2 percentage points (95% CI -1.8 to -0.6), risk ratio 0.84 (0.76 to 0.92)	[1c]
Conclusions	[Interpretation proportionate to the assumptions]	[1c]

2. Introduction

The introduction states what is unknown and why this emulation helps.

2.1 Background [2]

[Summarise what is known, the gap and why a causal answer matters for decisions.]

2.2 Causal question [3]

Element	Content
Population	Example: Adults aged 40 to 79 with type 2 diabetes on metformin who need a second-line drug
Strategies compared	Initiate class A versus initiate class B
Outcome	Hospitalisation for heart failure
Time horizon	5 years
Effect of interest	[Intention-to-treat effect / per-protocol effect]

2.3 Why emulate a target trial [4]

[Explain why a randomised trial is unavailable, unfeasible, too slow or unethical, why the data are suitable for the emulation, and which trials informed the design or could serve as a benchmark.]

Example: Head-to-head trials of class A versus class B are not available for this population. A placebo-controlled trial of class A reported a hazard ratio for heart failure hospitalisation that the team will use to check whether the emulation of a similar comparison reproduces the trial result (benchmarking, section 4.9).

3. Data sources

This section describes where the data come from [5].

Item	Detail
Source and purpose of collection	Example: Linked primary care records, hospital admissions and pharmacy dispensing collected for clinical care and reimbursement
Type of data	[Electronic health records / claims / registry / cohort study]
Location and setting	[Country, region, care setting]
Calendar period	[YYYY-MM-DD to YYYY-MM-DD]
Linkage	[Method, linkage rate, who performed it]
Coding systems	[Diagnosis, procedure and drug code systems with versions]
Validation of key variables	[References or local validation studies]
Data access	[Data custodian, approval reference]

4. Target trial protocol and emulation

The table summarises the protocol of the hypothetical target trial and its emulation. Sections 4.1 to 4.8 give the detail.

Component	Target trial [6a to 6h]	Emulation with the data [7a to 7h]
Eligibility criteria	Example: Adults 40 to 79, type 2 diabetes, on metformin for 6 months or more, no prior use of class A or B, no heart failure	Same criteria applied from records at each eligible date; diagnosis and dispensing codes in appendix B
Treatment strategies	Initiate class A; initiate class B; continue unless contraindicated	Initiation defined by first dispensing; adherence from refill records
Assignment	Random, unblinded	Assumed exchangeable conditional on baseline confounders in 4.7
Follow-up	From randomisation to outcome, death, loss or 5 years	From first dispensing (time zero) to outcome, death, deregistration or 5 years
Outcome	Hospitalisation for heart failure	First admission with a primary heart failure code, positive predictive value [x]% in validation
Causal contrasts	Intention-to-treat and per-protocol effects	Observational analogue of intention-to-treat; per-protocol by censoring at deviation plus weighting
Identifying assumptions	Randomisation ensures exchangeability	Conditional exchangeability, positivity, consistency, correct model specification
Analysis plan	Risk curves, 5-year risk difference and ratio	Pooled logistic regression with inverse probability weights, bootstrap CIs

4.1 Eligibility criteria [6a, 7a]

ID	Criterion	Target trial	How measured in the data	Look-back window
EL-01	Age	40 to 79 years	Date of birth	At time zero
EL-02	Diagnosis	Type 2 diabetes	[Code list]	Any time before
EL-03	Background therapy	Metformin for 6 months or more	Dispensing records	6 months
EL-04	Prior use of strategies	None	Dispensing records	Example: 12 months (new-user design)
EL-05	Exclusion	Heart failure, dialysis	[Code lists]	Any time before

All criteria must be assessable using information available on or before time zero only.

4.2 Treatment strategies [6b, 7b]

Strategy	Target trial definition	Data definition
S-A	Initiate class A at baseline, continue during follow-up unless a clinical reason to stop	First dispensing of class A; continuation defined by refills with gaps of no more than [90] days (the permissible gap)
S-B	Initiate class B at baseline, continue likewise	First dispensing of class B, same continuation rule

State what is allowed during follow-up (add-on therapy, switching for side effects) and how it is captured.

4.3 Assignment procedures [6c, 7c]

[In the target trial, individuals are randomly assigned to a strategy at baseline. Describe how the emulation assigns each person to a strategy, for example by the drug dispensed at time zero, and which baseline variables are needed to make the groups exchangeable.]

4.4 Follow-up period [6d, 7d]

Item	Target trial	Emulation
Start (time zero)	Randomisation	Example: Date of first dispensing, the date on which eligibility is confirmed and the strategy is assigned
End	Outcome, death, loss to follow-up, administrative end at 5 years	Same; loss defined as leaving the practice or end of data
Permissible gap or grace period (if used)	[Not needed / defined]	Example: A refill gap of more than 90 days counts as stopping; used only for the per-protocol analysis (EST-02). If a strategy allows initiation within a grace period after time zero, describe cloning, censoring and weighting here
Sequential trials (if used)	Not applicable	[Monthly eligible dates; a person can enter several trials; robust variance]

4.5 Outcomes [6e, 7e]

ID	Outcome	Target trial definition	Data definition	Ascertainment
OUT-01	Example: Hospitalisation for heart failure	First hospital admission for heart failure	Primary diagnosis code [list]	Hospital episode linkage
OUT-02	All-cause death	Death from any cause	Death registry	Registry linkage
OUT-03	[Negative control outcome]	[Outcome not plausibly affected by treatment]	[Codes]	[Source]

4.6 Causal contrasts and estimands [6f, 7f]

ID	Estimand	Population	Contrast	Summary measure
EST-01	Intention-to-treat effect	All eligible	Initiating S-A versus S-B, regardless of later adherence	5-year risk difference and risk ratio
EST-02	Per-protocol effect	All eligible	Initiating and adhering to S-A versus S-B	5-year risk difference and risk ratio

With $Y^{a}$ the potential outcome under strategy $a$ , the 5-year risk difference and risk ratio for EST-01 are:

\text{RD} = \Pr\left(Y^{A}=1\right) - \Pr\left(Y^{B}=1\right), \qquad \text{RR} = \frac{\Pr\left(Y^{A}=1\right)}{\Pr\left(Y^{B}=1\right)}

4.7 Identifying assumptions [6g, 7g]

Assumption	Meaning	How it is addressed
Conditional exchangeability	No unmeasured confounding given the baseline covariates	Example: Covariates chosen with a DAG: age, sex, HbA1c, eGFR, BMI, blood pressure, prior cardiovascular disease, statin use, deprivation, calendar year
Positivity	Every covariate pattern has people on both strategies	Inspect weights and propensity score overlap; trim or restrict if needed
Consistency	Strategies are well defined and the data version matches	Drug class defined by dispensing, dose ignored
No measurement error	Treatment, outcome and covariates correctly recorded	Validation references in section 3

The dashed arrows show the unmeasured confounding that the exchangeability assumption rules out; section 4.9 tests how sensitive the result is to it.

4.8 Statistical analysis plan [6h, 7h]

Item	Method
Estimator for EST-01	Example: Inverse probability of treatment weighting; weighted pooled logistic regression on months of follow-up to estimate standardised risk curves
Estimator for EST-02	Censor at deviation from the assigned strategy; inverse probability of censoring weights using time-varying covariates
Confidence intervals	Nonparametric bootstrap, 500 samples
Missing data	[Multiple imputation of baseline covariates, 20 data sets; outcome assumed complete through linkage]
Weight diagnostics	Mean, range and 99th percentile of weights; standardised mean differences below 0.1 after weighting
Software	[R / Stata / SAS with versions and packages]

The stabilised treatment weight for person $i$ with strategy $A_i$ and baseline covariates $L_i$ is:

SW_i = \frac{\Pr\left(A = A_i\right)}{\Pr\left(A = A_i \mid L = L_i\right)}

4.9 Additional and sensitivity analyses [7h]

ID	Analysis	Purpose
SA-01	E-value for the main risk ratio	Strength of unmeasured confounding needed to explain away the estimate
SA-02	Negative control outcome OUT-03	Detect residual confounding
SA-03	Permissible refill gap 60 versus 90 days	Sensitivity of EST-02 to the definition of stopping
SA-04	Benchmark against the placebo-controlled trial result	Check that the emulation can reproduce a known trial finding
SA-05	Complete-case analysis	Sensitivity to the missing data model

4.10 Deviations from the frozen protocol

ID	Date	Component	Change	Reason	Outcome data seen?
DEV-01 (Example)	YYYY-MM-DD	4.2 and 4.4 permissible refill gap	60 to 90 days	Feasibility counts (no outcomes) showed that many continuing users had a refill gap between 61 and 90 days	No

5. Design-related biases

Most serious biases in observational treatment studies come from the design, not from confounding. Check each one below.

Bias	How it arises	Prevention in this design
Immortal time bias	Follow-up starts before the strategy is assigned, so one group cannot have the outcome during a period	Time zero equals the date of assignment and eligibility
Prevalent user bias	Current users who survived early treatment are compared with new users	New-user design with a 12-month washout (EL-04)
Selection by post-baseline events	Eligibility depends on what happens after time zero, such as adherence	All criteria assessed on or before time zero
Time-varying confounding	Adherence depends on factors that also affect the outcome	Inverse probability of censoring weights for EST-02

The timeline shows correct alignment for one participant.

6. Results

This section reports who entered the emulated trial and what was found.

6.1 Participant flow [8]

All counts are Example values: 64,530 - 33,149 = 31,381 eligible and assigned; 12,406 + 18,975 = 31,381.

Reason not eligible	n
[Reason]	[n]
[Reason]	[n]
Total	[33,149]

6.2 Baseline characteristics [9]

Characteristic	Class A, unweighted	Class B, unweighted	SMD unweighted	SMD weighted
Age, years, mean (SD)	[value]	[value]	[0.21]	[0.02]
HbA1c, mmol/mol, mean (SD)	[value]	[value]	[value]	[value]
Prior cardiovascular disease, n (%)	[value]	[value]	[value]	[value]
[Confounder]	[value]	[value]	[value]	[value]

6.3 Follow-up [10]

[Median (IQR) follow-up by strategy, total person-years, and the number ending follow-up for each reason: outcome, death, loss, end of data, administrative end.]

6.4 Missing data [11]

Variable	Missing in class A, n (%)	Missing in class B, n (%)	Handling
[HbA1c]	[n (%)]	[n (%)]	[Multiple imputation]
[BMI]	[n (%)]	[n (%)]	[Multiple imputation]

6.5 Outcome frequency [12]

Strategy	Participants	Events (crude)	Crude 5-year proportion
Class A	Example: 12,406	720	5.8%
Class B	18,975	1,499	7.9%

Crude proportions ignore censoring and confounding; they describe the data and are not effect estimates.

6.6 Effect estimates [13]

Estimand	Weighted 5-year risk A	Weighted 5-year risk B	Risk difference (95% CI)	Risk ratio (95% CI)
EST-01	Example: 6.2%	7.4%	-1.2 percentage points (-1.8 to -0.6)	0.84 (0.76 to 0.92)
EST-02	[value]	[value]	[value]	[value]

Example: A risk difference of -1.2 percentage points corresponds to about 83 people initiating class A instead of class B for one fewer hospitalisation over 5 years (1 / 0.012 = 83.3).

6.7 Additional analyses [14]

ID	Result	Interpretation
SA-01	[E-value for RR and for the CI limit]	[Strength of confounding needed]
SA-02	[Estimate for negative control outcome]	[Close to null suggests limited residual confounding]
SA-03	[EST-02 estimate with a 60-day permissible gap]	[Robust or sensitive]
SA-04	[Emulated versus trial estimate]	[Agreement]
SA-05	[Complete-case estimate]	[Comparison]

7. Discussion

The discussion interprets the findings and their limits.

7.1 Interpretation [15]

[Summarise the key results for each estimand, compare them with randomised trials and other observational studies, and state what they mean for the decision the causal question was meant to inform.]

7.2 Limitations [16]

[Components where the emulation departs from the target trial, for example no blinding or a grace period not present in the trial]
[Plausibility of conditional exchangeability, positivity and consistency, with the evidence from section 6.7]
[Measurement error in treatment, outcome or confounders]
[Generalisability of the eligible population to the people who face the decision]

8. Other information

Transparency information for readers:

8.1 Ethics [17]

[Name of the approving committee and reference number, or the basis of exemption.]

8.2 Registration and protocol [18]

[Whether the protocol was registered or published, when (relative to data access and analysis) and where.]

8.3 Data, code and materials [19]

Item	Availability
Data	[Access route through the data custodian, conditions]
Code	[Repository and commit or DOI]
Code lists	[Appendix B or repository]

8.4 Funding [20]

[Funders, grant numbers and the role of each funder in design, analysis and reporting.]

8.5 Conflicts of interest [21]

[Financial and non-financial interests of each author.]

9. TARGET traceability

Item	Section of TARGET	Topic (short)	Section in this file	Page
1a to 1c	Abstract	Identification, data source, summary	1	[p]
2	Introduction	Background	2.1	[p]
3	Introduction	Causal question	2.2	[p]
4	Introduction	Rationale for emulation	2.3	[p]
5	Methods	Data sources	3	[p]
6a to 6h	Methods	Target trial protocol components	4, 4.1 to 4.8	[p]
7a to 7h	Methods	Emulation of each component	4, 4.1 to 4.8	[p]
8	Results	Participant flow	6.1	[p]
9	Results	Baseline characteristics	6.2	[p]
10	Results	Follow-up	6.3	[p]
11	Results	Missing data	6.4	[p]
12	Results	Outcome frequency	6.5	[p]
13	Results	Effect estimates	6.6	[p]
14	Results	Additional analyses	6.7	[p]
15	Discussion	Interpretation	7.1	[p]
16	Discussion	Limitations	7.2	[p]
17	Other information	Ethics	8.1	[p]
18	Other information	Registration	8.2	[p]
19	Other information	Data and code	8.3	[p]
20	Other information	Funding	8.4	[p]
21	Other information	Conflicts of interest	8.5	[p]

10. Glossary and references

10.1 Glossary

Term	Meaning in this template
Target trial	The randomised trial that would answer the causal question if it could be run
Time zero	The moment when eligibility is met, a strategy is assigned and follow-up starts
Estimand	The precise quantity the study aims to estimate
Intention-to-treat effect	The effect of being assigned to (here, initiating) a strategy, regardless of later adherence
Per-protocol effect	The effect of following the strategy as specified in the protocol
Cloning, censoring and weighting	A method for strategies that cannot be distinguished at time zero, such as grace periods
E-value	The minimum strength of association an unmeasured confounder would need with both treatment and outcome to explain away an estimate
SMD	Standardised mean difference, a balance measure

10.2 Methodological references

Cashin AG, Hansford HJ, Hernán MA, Swanson SA, et al. Transparent reporting of observational studies emulating a target trial: the TARGET Statement. BMJ. 2025;390:e087179. https://doi.org/10.1136/bmj-2025-087179 (also JAMA. 2025;334(12):1084-1093, https://doi.org/10.1001/jama.2025.13350)
Hernán MA, Robins JM. Using big data to emulate a target trial when a randomized trial is not available. Am J Epidemiol. 2016;183(8):758-764. https://doi.org/10.1093/aje/kwv254
Hernán MA, Wang W, Leaf DE. Target trial emulation: a framework for causal inference from observational data. JAMA. 2022;328(24):2446-2447. https://doi.org/10.1001/jama.2022.21383

Related templates in this library: science/observational-study (STROBE report), science/preregistration (study registration), academic/stats-analysis-plan (statistical analysis plan), academic/stats-logistic-regression (regression reporting), science/trial-protocol (protocol for a real randomised trial) and science/prediction-model-report (prediction rather than causal questions).

Standard and licence note

Structure aligned with the TARGET guideline (Transparent Reporting of Observational Studies Emulating a Target Trial), published simultaneously in JAMA 2025;334(12):1084-1093 and BMJ 2025;390:e087179, listed by the EQUATOR Network at https://www.equator-network.org/reporting-guidelines/22462/. The TARGET checklist is published by the TARGET group under CC BY-ND 4.0, so this template does not reproduce or adapt its wording: it uses only the item numbers and short topic names, with original explanatory text. Official checklist: TARGET group, https://target-guideline.org. Checked in October 2026; check the current version before use. This is a document format, not clinical, epidemiological or statistical advice. Template licence: MIT.