brycewang-stanford/Auto-Empirical-Research-Skills / skills/14-luischanci-claude-code-research-starter/dot-claude/skills/stata
stata
VerifiedComprehensive Stata reference for writing correct .do files, data management, econometrics, causal inference, graphics, Mata programming, and 17+ community packages (reghdfe, estout, did, rdrobust, etc.). Covers syntax, options, gotchas, and idiomatic patterns. Use this skill whenever the user asks you to write, debug, or explain Stata code.
references/survival-analysis.md
View on GitHub# Survival Analysis in Stata
## Declaring Survival Data (stset)
All survival commands require `stset` first. This creates special variables: `_t0` (interval start), `_t` (interval end), `_d` (failure indicator), `_st` (sample indicator).
```stata
// Basic setup
stset timevar, failure(failvar)
// Specific failure value (e.g., status==1 is death, status==2 is censored)
stset time, failure(status==1)
// Scale time units (days to years)
stset time_days, failure(event) scale(365.25)
// Panel data with subject ID (required for multiple records per subject)
stset time, failure(event) id(patient_id)
// Delayed entry (left truncation)
stset exit_time, failure(died) enter(entry_time) origin(time birth_date)
// Suppress stset banner in subsequent commands
stset time, failure(died) noshow
// Inspect the setup
stdes // Summary of survival data structure
stsum // Summary statistics (median survival, etc.)
```
---
## Kaplan-Meier Estimator
### Survival Curves (sts graph)
```stata
webuse drug2, clear
stset studytime, failure(died)
// Basic survival curve
sts graph
// By group
sts graph, by(drug)
// Customized with risk table
sts graph, by(drug) ///
title("Survival by Treatment Group") ///
xtitle("Time (days)") ytitle("Survival Probability") ///
legend(label(1 "Placebo") label(2 "Drug A")) ///
risktable
// Cumulative hazard plot
sts graph, cumhaz by(drug)
// Hazard function (smoothed)
sts graph, hazard by(drug) kernel(epan2)
```
### Survival Tables (sts list)
```stata
sts list // Full table
sts list, at(0 100 200 300 400 500) // At specific times
sts list, at(0 to 500 by 100) failure // Failure probability
sts list, by(drug) ci // By group with CIs
```
### Median Survival Time
```stata
stsum // Overall median survival
stsum, by(drug) // By group
stci // Median with CI
stci, by(drug) // By group
stci, p(25) // 25th percentile
stci, p(75) // 75th percentile
```
If median is not reached, more than 50% survived to end of study.
---
## Log-Rank Test (sts test)
Compares survival curves between groups. H0: survival functions are equal.
```stata
sts test drug // Log-rank (default)
sts test drug, wilcoxon // Wilcoxon/Breslow (weights early events)
sts test drug, tware // Tarone-Ware (intermediate weighting)
sts test drug, peto // Peto-Peto
sts test drug, fh(1 0) // Fleming-Harrington (weights early)
sts test drug, fh(0 1) // Fleming-Harrington (weights late)
sts test drug, strata(age_group) // Stratified log-rank
sts test dose_level, trend // Trend test for ordered groups
```
**Which test to use:**
- **Log-rank**: Most powerful when hazards are proportional
- **Wilcoxon**: More sensitive to early differences
- **Fleming-Harrington**: Flexible weighting via fh(p q) parameters
---
## Cox Proportional Hazards Model (stcox)
Semi-parametric model: h(t|X) = h0(t) * exp(b1*X1 + b2*X2 + ...). Baseline hazard h0(t) is unspecified.
```stata
stcox drug // Simple model
stcox drug age // Multiple covariates
stcox i.drug age // Factor notation for categorical
stcox i.drug##c.age // Interaction
// Display options
stcox drug age // Default: hazard ratios
stcox drug age, nohr // Coefficients instead
// Standard errors
stcox drug age, vce(robust) // Robust SE
stcox drug age, vce(cluster hospital_id) // Clustered SE
```
### Interpreting Hazard Ratios
- **HR = 1**: No effect
- **HR > 1**: Increased hazard (higher risk, shorter survival)
- **HR < 1**: Decreased hazard (lower risk, longer survival)
- **Formula**: % change = (HR - 1) x 100%
- **Continuous vars**: For k-unit increase, HR_k = HR^k (e.g., 10-year age effect: 1.12^10 = 3.11)
- **CI excluding 1**: Statistically significant
```stata
stcox drug age
// drug HR=0.104 → 89.6% reduction in hazard (1 - 0.104)
// age HR=1.12 → 12% increase in hazard per year
```
### Stratified Cox Models
When PH assumption is violated for a variable, stratification allows different baseline hazards per stratum. The stratifying variable's effect is not estimated.
```stata
stcox drug age, strata(female)
stcox drug age, strata(female hospital)
```
---
## Testing Proportional Hazards Assumption
### Schoenfeld Residuals
```stata
stcox drug age
estat phtest // Global test
estat phtest, detail // Per-variable test
// p > 0.05 → PH assumption reasonable
// rho = correlation between residuals and time
// Plot residuals against time
estat phtest, plot(drug)
estat phtest, plot(age)
```
**If PH is violated:**
- Stratify: `stcox x1, strata(x2)`
- Include time interaction: `stcox x1, tvc(x1) texp(_t)`
- Use parametric model with time-varying effects
### Graphical Assessment
```stata
// Log-log plots (should be parallel if PH holds)
sts graph, by(drug) gwood
// Observed KM vs Cox-predicted survival
stcox drug age
stcurve, survival at1(drug=0) at2(drug=1)
```
---
## Parametric Survival Models (streg)
Unlike Cox, these specify the baseline hazard distribution. Two metrics:
- **Proportional hazards (PH)**: Reports hazard ratios (default)
- **Accelerated failure time (AFT)**: Reports time ratios (option `time`)
### Available Distributions
```stata
streg x1 x2, distribution(exponential) // Constant hazard
streg x1 x2, distribution(weibull) // Monotone hazard (increasing or decreasing)
streg x1 x2, distribution(gompertz) // Exponentially changing hazard
streg x1 x2, distribution(lognormal) // Non-monotone hazard (AFT only)
streg x1 x2, distribution(loglogistic) // Non-monotone hazard (AFT only)
streg x1 x2, distribution(ggamma) // Generalized gamma (nests exp, Weibull, lognormal)
```
### PH vs AFT Metric
```stata
// PH metric (hazard ratios)
streg drug age, distribution(weibull)
// HR=0.10 → drug reduces hazard by 90%
// AFT metric (time ratios)
streg drug age, distribution(weibull) time
// TR=10.1 → drug multiplies survival time by 10.1
// Coefficients (log scale)
streg drug age, distribution(weibull) nohr
```
### Weibull Shape Parameter
```stata
streg drug age, distribution(weibull)
di "p (shape) = " exp(_b[/ln_p])
// p < 1: hazard decreases over time
// p = 1: constant hazard (reduces to exponential)
// p > 1: hazard increases over time
```
### Frailty (Random Effects)
```stata
streg drug age, distribution(weibull) frailty(gamma) shared(hospital_id)
```
### Model Comparison
```stata
streg drug age, distribution(exponential)
estimates store exp_model
streg drug age, distribution(weibull)
estimates store weib_model
streg drug age, distribution(lognormal)
estimates store lnorm_model
streg drug age, distribution(loglogistic)
estimates store llogistic_model
// AIC/BIC comparison (lower is better)
estimates stats exp_model weib_model lnorm_model llogistic_model
// LR test: exponential nested in Weibull (test p=1)
streg drug age, distribution(weibull)
test [/]ln_p = 0
```
---
## Time-Varying Covariates
### Using stsplit
Splits each subject's record into multiple time intervals for time-varying analysis.
```stata
webuse drugtr, clear
stset studytime, failure(died)
// Split at specific times
stsplit period, at(0 100 200 300)
// Split into equal intervals
stsplit interval, every(50)
// Inspect split data
list studytime died _t0 _t period in 1/20, sepby(studytime)
// Create time-varying covariate
gen age_current = age + _t/365.25
```
### Cox Model with Time-Varying Covariates
```stata
// Manual approach: split then create TVC
stset studytime, failure(died) id(id)
stsplit month, every(30)
gen time_on_drug = _t
gen drug_time = drug * time_on_drug
stcox drug drug_time age
// drug: HR at time 0
// drug_time: change in log(HR) per unit time
// Built-in tvc() option (equivalent, no splitting needed)
stcox drug age, tvc(drug) texp(_t)
// Negative tvc coefficient → effect becomes more protective over time
// Non-significant → PH assumption holds for that variable
```
### Treatment Switching Example
```stata
stset failure_time, failure(died) id(patient_id)
stsplit switch, at(switch_time)
gen current_treatment = (switch_time != . & _t >= switch_time)
stcox current_treatment age baseline_severity
```
---
## Competing Risks
When subjects can experience multiple event types (e.g., death from cancer vs. death from other causes).
### Cumulative Incidence Function (CIF)
```stata
// failure_type: 1=relapse, 2=death in remission, 0=censored
stset time_months, failure(failure_type==1) id(id)
// CIF for cause 1, with cause 2 as competing event
stcompet CIF_relapse = ci, compet1(2)
// CIF by group
stcompet CIF = ci, compet1(2) by(treatment)
```
### Fine-Gray Subdistribution Hazards (stcrreg)
```stata
// Fine-Gray model for cause 1
stcrreg i.treatment age, compete(failure_type==2)
// Subdistribution HR: interpretation similar to Cox HR but accounts for competing events
// Model for each cause
stset time_months, failure(failure_type==1)
stcrreg i.treatment age, compete(failure_type==2)
estimates store cause1
stset time_months, failure(failure_type==2)
stcrreg i.treatment age, compete(failure_type==1)
estimates store cause2
estimates table cause1 cause2, eform
```
### CIF Predictions After Fine-Gray
```stata
stcrreg i.treatment age, compete(failure_type==2)
stcurve, cif at1(treatment=1 age=40) at2(treatment=2 age=40)
```
---
## Survival Predictions
### After Cox Model
```stata
stcox drug age
// Predicted curves at specific covariate values
stcurve, survival at1(drug=0 age=50) at2(drug=1 age=50)
stcurve, cumhaz at1(drug=0 age=50) at2(drug=1 age=50)
stcurve, hazard at1(drug=0 age=50) at2(drug=1 age=50)
// Individual-level predictions
predict xb, xb // Linear predictor
predict hr, hr // Relative hazard (exp(xb))
predict basehaz, basehazard // Baseline cumulative hazard
predict basesurv, basesurv // Baseline survival
```
### After Parametric Model
```stata
streg drug age, distribution(weibull) time
predict median_t, median time // Median survival time
predict mean_t, mean time // Mean survival time
predict S_at_100, surv time at(100) // S(t=100)
predict h_at_100, hazard time at(100) // h(t=100)
```
### Marginal Effects
```stata
stcox drug i.stage age female
// Marginal survival at specific times
margins drug, predict(surv time(100))
margins drug, predict(surv time(200))
// Plot marginal effects
margins, at(drug=(0 1) age=(40 50 60)) predict(surv time(200))
marginsplot
// Pairwise comparisons of categorical groups
stcox i.treatment_group age
contrast treatment_group, effects
```
---
## Model Diagnostics
```stata
estat concordance // Harrell's C statistic
predict cs, csnell // Cox-Snell residuals
predict mg, mgale // Martingale residuals
predict dev, deviance // Deviance residuals
predict sch*, scaledsch // Scaled Schoenfeld residuals
predict dfb*, dfbeta // DFbeta influence
predict lmax, lmax // Likelihood displacement
```
---
## Quick Reference
### stset Options
| Option | Purpose |
|---|---|
| `failure(failvar)` | Failure indicator variable |
| `failure(status==1)` | Specific failure value |
| `id(subject_id)` | Subject identifier (multi-record data) |
| `enter(time t0)` | Delayed entry / left truncation |
| `origin(time birthdate)` | Origin for time scale |
| `scale(365.25)` | Scale time (e.g., days to years) |
| `noshow` | Suppress stset banner |
### Hazard Ratio Quick Interpretation
| HR | Interpretation |
|---|---|
| 1.0 | No effect |
| 1.5 | 50% increase in hazard |
| 2.0 | Doubled hazard |
| 0.5 | 50% decrease in hazard |
| 0.25 | 75% decrease in hazard |
### streg Distributions
| Distribution | Hazard Shape | Metrics Available |
|---|---|---|
| `exponential` | Constant | PH, AFT |
| `weibull` | Monotone | PH, AFT |
| `gompertz` | Exponential change | PH |
| `lognormal` | Non-monotone (hump) | AFT only |
| `loglogistic` | Non-monotone (hump) | AFT only |
| `ggamma` | Flexible (nests above) | AFT only |
### sts test Variants
| Test | When to Use |
|---|---|
| `sts test x` (log-rank) | Proportional hazards |
| `sts test x, wilcoxon` | Early differences |
| `sts test x, tware` | Intermediate |
| `sts test x, fh(p q)` | Custom weighting |
| `sts test x, trend` | Ordered groups |
| `sts test x, strata(z)` | Adjusted comparison |
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the work | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixes | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation code | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |