diagtab

Tables & Reporting

Diagnostic accuracy and cutoff analysis

 . net install diagtab, from(...)  

View on GitHub →

diagtab hex logo

Version 2.1.0 | 2026-10-13

Quick Start · Installation · Options · Worked Examples · Stata-Tools · Report a bug

diagtab computes diagnostic accuracy measures and confidence intervals from binary classifications or continuous scores. It produces publication-ready console, Excel, CSV, Markdown, and frame output for clinical diagnostic and screening studies.

Quick Start

After installing, score a logistic model for low birth weight and evaluate it at one cutoff, with the ROC area:

webuse lbw, clear
logit low age lwt smoke
predict double phat
diagtab phat low, cutoff(0.30) auc

Add xlsx(), csv(), markdown(), or frame() to send the same table to a file or a frame.

Features

The standard measures in one call. Sensitivity, specificity, PPV, NPV, accuracy, both likelihood ratios, the diagnostic odds ratio, and Youden's index, each with its confidence interval.
Binary tests or continuous scores. cutoff() evaluates one threshold, cutoffs() builds a threshold table, optimal picks the cutoff that maximizes Youden's J, and auc adds the ROC area through Stata's roctab.
Interval method on record. Wilson score intervals by default or Clopper–Pearson with exact, and r(methods) returns a methods paragraph describing the choice.
Predictive values for another population. prevalence() recalculates PPV and NPV at a fixed target prevalence, with delta-method intervals when sensitivity and specificity are both interior estimates.

Requirements

Stata 17 or later

Installation

capture ado uninstall diagtab
net install diagtab, from("https://raw.githubusercontent.com/tpcopeland/Stata-Tools/main/diagtab") replace

Commands

Command	Description
`diagtab`	Diagnostic accuracy, confidence intervals, ROC AUC, and cutoff analysis

How It Works

With a binary test variable, diagtab builds the 2 × 2 classification table directly. For a continuous score, use cutoff() for one threshold, cutoffs() for a threshold table, auc for ROC area, or optimal to select the cutoff that maximizes Youden's J.

The command reports sensitivity, specificity, PPV, NPV, accuracy, likelihood ratios, diagnostic odds ratio, and Youden's index. Wilson score intervals are the default; exact selects Clopper–Pearson intervals for directly estimated binomial proportions. prevalence() recalculates predictive values for a fixed target prevalence and uses delta-method intervals when sensitivity and specificity are both interior estimates.

Worked Examples

1. Binary classification

webuse lbw, clear
logit low age lwt smoke
predict double phat
generate byte predicted = phat >= 0.30
diagtab predicted low

2. Continuous score with AUC

webuse lbw, clear
logit low age lwt smoke
predict double phat
diagtab phat low, cutoff(0.30) auc level(95)

3. Compare thresholds and export

webuse lbw, clear
logit low age lwt smoke
predict double phat
diagtab phat low, cutoffs(0.20 0.30 0.40 0.50) ///
    xlsx("diagnostic_cutoffs.xlsx") sheet("Cutoffs") ///
    title("Low birth weight prediction")

4. Target-population predictive values

webuse lbw, clear
logit low age lwt smoke
predict double phat
generate byte predicted = phat >= 0.30
diagtab predicted low, prevalence(0.07) exact

Demo

Run demo/demo_diagtab.do from a repository checkout to regenerate demo/demo_diagtab.xlsx. The workbook is a repository documentation asset and is not part of the net install payload.

Sheet	What it holds
`Diagnostic`	`cutoff(0.30) auc` on the low-birth-weight model, with `theme(nejm)`
`Prevalence`	The same cutoff with predictive values at `prevalence(0.07)`
`Cutoffs`	A threshold table for `cutoffs(0.20 0.30 0.40 0.50)`

Options

Option	Purpose
`cutoff(#)`	Evaluate one nonmissing threshold; values at or above it are test-positive
`cutoffs(numlist)`	Evaluate sorted, distinct thresholds, return exact round-trip values in `r(cutoffs)`, and use unique row identifiers in `r(cutoff_table)`
`prevalence(#)`	Adjust PPV and NPV to a fixed external prevalence strictly between 0 and 1
`exact`	Use Clopper–Pearson intervals for directly estimated binomial proportions
`wilson`	Use Wilson score intervals, the default
`auc`	Report ROC area and its confidence interval through Stata's `roctab`
`optimal`	Choose the cutoff maximizing Youden's J
`level(#)`	Set the confidence level; default `c(level)`
`digits(#)`	Set displayed decimal places from 0 through 6; default 1
`xlsx(filename)`	Export an `.xlsx` workbook
`excel(filename)`	Alias for `xlsx()`; the two aliases may not be specified together
`sheet(string)`	Set the Excel sheet name; default `Diagnostics`
`csv(filename)`	Export the rendered output dataset as CSV
`markdown(filename)`	Export a GitHub-Flavored Markdown table
`mdappend`	Append to an existing Markdown file; requires `markdown()`
`frame(name[, replace])`	Store output in a named Stata frame, optionally replacing it
`open`	Open the workbook after export; requires `xlsx()` or `excel()`
`title(string)`	Set the table title
`footnote(string)`	Add text below the table
`theme(string)`	Apply `lancet`, `nejm`, `bmj`, `apa`, `jama`, `plos`, `nature`, `cell`, `annals`, or `custom` formatting
`borderstyle(string)`	Set `default`, `thin`, `medium`, or `academic` workbook borders
`headercolor(string)`	Set the header fill using a Stata color name or RGB triplet
`headershade`	Fill workbook header rows
`zebracolor(string)`	Set the alternating-row fill color
`zebra`	Enable alternating-row shading

Stored Results

Scalars

Result	Description
`r(TP)`, `r(FP)`, `r(FN)`, `r(TN)`	Two-by-two cell counts in single-cutoff mode
`r(ci_level)`	Resolved confidence level
`r(sensitivity)`, `r(sensitivity_lb)`, `r(sensitivity_ub)`	Sensitivity and confidence bounds
`r(specificity)`, `r(specificity_lb)`, `r(specificity_ub)`	Specificity and confidence bounds
`r(ppv)`, `r(ppv_lb)`, `r(ppv_ub)`	Positive predictive value and confidence bounds
`r(npv)`, `r(npv_lb)`, `r(npv_ub)`	Negative predictive value and confidence bounds
`r(accuracy)`, `r(accuracy_lb)`, `r(accuracy_ub)`	Accuracy and confidence bounds
`r(lr_pos)`, `r(lr_pos_lb)`, `r(lr_pos_ub)`	Positive likelihood ratio and confidence bounds
`r(lr_neg)`, `r(lr_neg_lb)`, `r(lr_neg_ub)`	Negative likelihood ratio and confidence bounds
`r(dor)`, `r(dor_lb)`, `r(dor_ub)`	Diagnostic odds ratio and confidence bounds
`r(youden)`	Youden's J index
`r(auc)`, `r(auc_lb)`, `r(auc_ub)`	ROC area and confidence bounds when `auc` is requested
`r(optimal_cutoff)`	Youden-optimal cutoff when `optimal` is requested
`r(markdown_rows)`, `r(markdown_cols)`	Markdown body dimensions when exported

Macros and matrix

Result	Description
`r(cutoffs)`	Normalized cutoff values used in multi-cutoff mode
`r(xlsx)`, `r(sheet)`	Workbook filename and sheet name when exported
`r(frame)`	Output frame name when requested
`r(markdown)`	Markdown filename when exported
`r(methods)`	Methods paragraph describing interval choices
`r(cutoff_table)`	Multi-cutoff matrix with sensitivity, specificity, PPV, NPV, accuracy, and bounds

Assumptions and Limits

The gold-standard variable must be coded 0/1 and contain both classes for AUC analysis.
Without a cutoff option, the test variable must be coded 0/1.
prevalence() is treated as fixed without uncertainty. Adjusted PPV/NPV confidence bounds are missing when sensitivity or specificity equals 0 or 1, avoiding a degenerate point interval.
Undefined measures caused by zero cells are displayed as -- and retained as missing numeric results.

References

Clopper CJ, Pearson ES. The use of confidence or fiducial limits illustrated in the case of the binomial. Biometrika. 1934;26:404–413.
DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics. 1988;44:837–845.
Glas AS, Lijmer JG, Prins MH, Bonsel GJ, Bossuyt PMM. The diagnostic odds ratio: a single indicator of test performance. Journal of Clinical Epidemiology. 2003;56:1129–1135.
Wilson EB. Probable inference, the law of succession, and statistical inference. Journal of the American Statistical Association. 1927;22:209–212.
Youden WJ. Index for rating diagnostic tests. Cancer. 1950;3:32–35.

QA

QA suites and how to run them are documented in qa/README.md.

Version History

2.1.0 (2026-10-13): With prevalence(), the accuracy row is now labelled "Accuracy (sample)" in the console, Excel, CSV, Markdown and frame output and r(methods) states that accuracy is not adjusted to the external prevalence; r(accuracy) is unchanged. The first call in a session no longer echoes Mata source to the console, and a helper that does not compile now stops with an explicit message and its return code.
2.0.3 (2026-10-09): Help References now give a DOI for every entry.
2.0.2 (2026-09-30): Filename preflight now examines the actual quoted path and reports invalid characters accurately. Existing refusal of quote characters and shell metacharacters is retained; filenames with spaces and Unicode remain accepted. Diagnostic calculations and cold helper loading preserve caller native S_1–S_6 globals and matastrict. Titles and footnotes retain literal macro-looking bytes in table and export contents. Early refusals preserve exact caller r() scalars, macros and matrices; analytical payloads still remain available after a later export failure.
2.0.1 (2026-08-30): Preserved all real cutoff values, made multi-cutoff identifiers lossless and unique, preflighted output conflicts, rejected conflicting Excel aliases and invalid font sizes before export, and avoided degenerate prevalence-adjusted intervals at boundary estimates.

Earlier versions

2.0.0 (2026-08-19): Extracted diagtab into a standalone package while preserving its command and stored-result contracts.

Author

Timothy P Copeland, Karolinska Institutet

Questions and bug reports: GitHub issues. Please include the first line of which diagtab and your Stata version.

To cite: Copeland TP. diagtab: Diagnostic accuracy and cutoff analysis. Stata-Tools, https://github.com/tpcopeland/Stata-Tools/tree/main/diagtab. Please give the version you used (which diagtab prints it).

License

MIT (see LICENSE)