Generated by scripts/longitudinal_examples.py. Every example is a real PTB-XL pair whose
current report contains the reading cardiologist's own comparison against the prior
tracing, so each APEX output is graded against a physician's account of the same two ECGs
rather than against my own reading of them. 110 PTB-XL reports contain such a statement;
these 12 are the ones whose prior tracing is also in the dataset and recoverable.
These are not a benchmark — 12 pairs, and the ones where a comparison happened to be dictated and the prior happened to be retained, which is not a random sample. They are worked examples with an unusually good reference standard.
| Verdict | n |
|---|---|
| Concordant | 3 |
| Concordant but incomplete | 3 |
| Mixed | 2 |
| Discordant | 4 |
6 of 12 outputs agreed with the cardiologist on the principal change, 2 were mixed, and 4 disagreed. The consistency gate passed on all 12: no output asserted a change absent from its structured delta.
What the failures have in common. Three of the four discordant cases are false
positives on stable studies — 11737 → 11740 is the cleanest, where the cardiologist wrote
"it is not significantly changed" and APEX reported three changes, all spurious. The fourth
(7814 → 7824) is the opposite failure: a real anterior ST elevation that fell under the
family-wise ST bar. Sensitivity and specificity trade against each other here exactly as
the threshold analysis in report.md predicts, and this small sample lands on
the specific side of that trade.
What the successes have in common. The strongest outputs (16404 → 16408, 8636 → 8641, 4097 → 4103) are all cases where the change is large relative to the noise floor — a rhythm conversion, a 0.12 mV ST shift, a disappearing ectopic. None required fine judgement. The module is reliable exactly where the signal is unambiguous, which is worth knowing before trusting it where the signal is not.
| Elapsed | 4 hours (1987-07-02 → 1987-07-02) |
| PTB-XL fold | 3 |
| Prior codes | NST_, RAO/RAE, STACH, STD_ |
| Current codes | NORM, NT_, STACH |
| Prior recovered by | earlier-today |
Cardiologist's comparison (PTB-XL report, verbatim):
compared with tracing of earlier today. the previously noted st segment depression is now negligible. there is t wave flattening in ii, iii, avf. suggest repeat in one week or so. Edit: NORM 100, (f,29) , ansteigende ST-Senkungen in II,III, flache T V5,6, V.a.RAO
APEX change report:
Compared with the prior study of 1987-07-02 (4 hours earlier): New since the prior study: non-diagnostic T-wave abnormality and normal ECG. Previously noted ST deviation in lead II has resolved. Measurable but sub-threshold ST shift in leads V3-V5, remaining within normal limits.
Impression: interval change since the prior study — new non-diagnostic T-wave abnormality. Persisting: sinus tachycardia.
Not compared — PR interval not compared: no P wave detected in the prior study — PR interval is undefined; QT interval not compared: not measurable in one or both studies; QTc (Bazett) not compared: not measurable in one or both studies; QTc (Fridericia) not compared: not measurable in one or both studies.
Measured: HR 130 → 111 bpm · PR — → 122 ms · QRS 80 → 80 ms · QT — → 333 ms · QTcF — → 408 ms
Review — Concordant but incomplete. Both changes the cardiologist named were caught: the resolving ST depression ('now negligible') and the new T-wave abnormality ('t wave flattening'). APEX localised the resolved ST deviation to lead II, which matches. It did not reproduce the II/III/aVF localisation of the T-wave change, because the diagnostic channel carries no lead information — only the ST channel does. Listing 'normal ECG' as a new finding alongside a T-wave abnormality is faithful to the annotation but reads oddly.
Consistency gate: passed
| Elapsed | 2 hours (1988-04-10 → 1988-04-10) |
| PTB-XL fold | 3 |
| Prior codes | INVT, NDT, SR |
| Current codes | LPFB, NDT, NORM, SR |
| Prior recovered by | earlier-today |
Cardiologist's comparison (PTB-XL report, verbatim):
compared with tracing of earlier today. axis is more to the right. t wave changes in chest leads are not significantly changed.
APEX change report:
Compared with the prior study of 1988-04-10 (2 hours earlier): New since the prior study: incomplete right bundle branch block. No longer present: anteroseptal ischemia. The heart rate has increased from 69 bpm to 100 bpm (+31 bpm).
Impression: interval change since the prior study — new incomplete right bundle branch block. Persisting: sinus rhythm.
Not compared — PR interval not compared: no P wave detected in the current study — PR interval is undefined; QT interval not compared: heart rate changed by more than the detectable threshold and the QT interval is rate-dependent, so QTc (Fridericia) is reported instead; QTc (Bazett) not compared: heart rate changed by more than the detectable threshold and Bazett's correction is rate-dependent, so QTc (Fridericia) is reported instead.
Measured: HR 69 → 100 bpm · PR 184 → — ms · QRS 79 → 78 ms · QT 378 → 318 ms · QTcF 396 → 377 ms
Review — Discordant. The cardiologist's headline was 'axis is more to the right', which the annotation records as new LPFB — left posterior fascicular block is the classic cause of right-axis shift. APEX flagged a conduction change but named the wrong one (incomplete RBBB): a detector error, not a comparison error. It then asserted resolved anteroseptal ischemia against an explicit 't wave changes in chest leads are not significantly changed' — a false positive from probability flicker. The +31 bpm rate rise is real but went unremarked by the cardiologist.
Consistency gate: passed
| Elapsed | 4 hours (1989-07-24 → 1989-07-24) |
| PTB-XL fold | 1 |
| Prior codes | INVT, ISC_, LVH, SR, STD_, VCLVH |
| Current codes | ISCAL, SR |
| Prior recovered by | earlier-today |
Cardiologist's comparison (PTB-XL report, verbatim):
compared with tracing of earlier today. st segments remain depressed in i, ii, v4,5,6. t waves are now low rather than inverted in i, v5,6. the slightly changed pattern suggests recent ischaemia.
APEX change report:
Compared with the prior study of 1989-07-24 (4 hours earlier): The PR interval has increased from 98 ms to 176 ms (+78 ms). The QRS duration has increased from 80 ms to 98 ms (+18 ms).
Impression: PR interval and QRS duration changed beyond the measurable threshold; no new diagnostic finding. Persisting: non-specific ischemia, left ventricular hypertrophy, and sinus rhythm.
Measured: HR 69 → 67 bpm · PR 98 → 176 ms · QRS 80 → 98 ms · QT 355 → 380 ms · QTcF 372 → 394 ms
Review — Discordant. The worst measurement failure in the set. APEX reports PR increasing 98 -> 176 ms, but 98 ms is below the physiologic floor for a conducted P wave and is almost certainly a mis-placed P onset in the prior study; the 80 ms plausibility guard was too permissive to catch it. Meanwhile the change the cardiologist actually described — 't waves are now low rather than inverted in i, v5,6' — was missed, because T-wave morphology change is not something this module measures. It correctly did not call the persistently depressed ST segments new.
Consistency gate: passed
| Elapsed | 6 hours (1990-04-30 → 1990-04-30) |
| PTB-XL fold | 10 (held-out test fold) |
| Prior codes | NORM, PAC, SR |
| Current codes | NORM, SR |
| Prior recovered by | earlier-today |
Cardiologist's comparison (PTB-XL report, verbatim):
compared with tracing of earlier today. it is not significantly changed except that atrial premature beats are not recorded on this occasion.
APEX change report:
Compared with the prior study of 1990-04-30 (6 hours earlier): Sinus arrhythmia has reverted to sinus rhythm. No longer present: atrial premature complexes.
Impression: interval change since the prior study — new sinus rhythm. Persisting: normal ECG.
Measured: HR 86 → 83 bpm · PR 156 → 151 ms · QRS 84 → 85 ms · QT 325 → 330 ms · QTcF 367 → 368 ms
Review — Concordant. Near-exact. The cardiologist wrote 'not significantly changed except that atrial premature beats are not recorded on this occasion'; APEX reported the resolved atrial premature complexes and nothing else of substance. The SARRH -> SR transition is an annotation-level distinction the cardiologist did not draw. Before the Bonferroni correction on the ST channel this pair also carried a spurious 'new ST elevation in V3', which the correction removed.
Consistency gate: passed
| Elapsed | 4 days (1992-01-24 → 1992-01-28) |
| PTB-XL fold | 3 |
| Prior codes | 1AVB, DIG, ISC_, LOWT, LPR, LVH, PVC, SR |
| Current codes | 3AVB, ISC_, LVH, NST_, SR |
| Prior recovered by | exact-date |
Cardiologist's comparison (PTB-XL report, verbatim):
compared with tracing of 24:1,92, it is otherwise generally similar with st-t wave changes.
APEX change report:
Compared with the prior study of 1992-01-24 (4 days earlier): No longer present: ventricular premature complexes and sinus rhythm. The heart rate has decreased from 67 bpm to 41 bpm (-26 bpm). The PR interval has decreased from 278 ms to 180 ms (-98 ms).
Impression: heart rate and PR interval changed beyond the measurable threshold; no new diagnostic finding.
Not compared — QT interval not compared: heart rate changed by more than the detectable threshold and the QT interval is rate-dependent, so QTc (Fridericia) is reported instead; QTc (Bazett) not compared: heart rate changed by more than the detectable threshold and Bazett's correction is rate-dependent, so QTc (Fridericia) is reported instead.
Data quality — current median beat built from only 4 beats.
Measured: HR 67 → 41 bpm · PR 278 → 180 ms · QRS 91 → 92 ms · QT 344 → 418 ms · QTcF 358 → 368 ms
Review — Mixed. The patient developed complete heart block (1AVB -> 3AVB). APEX captured its haemodynamic signature exactly — rate down 67 -> 41 bpm, sinus rhythm no longer present — but never named third-degree AV block, and worse, reported 'PR decreased from 278 ms to 180 ms'. In complete AV block the P waves are dissociated from the QRS and PR is undefined. The suppression rule added for this case fires on a surfaced 2AVB/3AVB label; here the detector did not surface one, so the guard could not help. The guard is only as good as the rhythm call behind it, which is a real limit worth stating plainly.
Consistency gate: passed
| Elapsed | 23 hours (1992-02-27 → 1992-02-28) |
| PTB-XL fold | 3 |
| Prior codes | IMI, INJAS, PVC, STACH |
| Current codes | ASMI, IMI, STACH |
| Prior recovered by | exact-date |
Cardiologist's comparison (PTB-XL report, verbatim):
compared with tracing of 27:2:92, st segment elevation in anterior chest is more marked. a recent anterior infarct is likely.
APEX change report:
Compared with the prior study of 1992-02-27 (23 hours earlier): No longer present: ventricular premature complexes. The QRS duration has decreased from 101 ms to 79 ms (-21 ms).
Impression: QRS duration changed beyond the measurable threshold; no new diagnostic finding. Persisting: sinus tachycardia.
Not compared — PR interval not compared: no P wave detected in the prior and current study — PR interval is undefined.
Measured: HR 103 → 111 bpm · PR — → — ms · QRS 101 → 79 ms · QT 291 → 263 ms · QTcF 349 → 323 ms
Review — Discordant. A clean miss on the thing that mattered. The cardiologist: 'st segment elevation in anterior chest is more marked. a recent anterior infarct is likely', and the annotation adds ASMI. APEX reported only resolved PVCs and a QRS narrowing, and said nothing about anterior ST. Both studies are sinus tachycardia at 103-111 bpm with no measurable P wave, and the ST elevation increment fell under the 0.07 mV family-wise bar — the cost of that correction, paid here.
Consistency gate: passed
| Elapsed | 2 days (1992-07-14 → 1992-07-17) |
| PTB-XL fold | 10 (held-out test fold) |
| Prior codes | ISCAL, LOWT, SR, STD_ |
| Current codes | ISC_, LVH, NST_, STACH |
| Prior recovered by | exact-date |
Cardiologist's comparison (PTB-XL report, verbatim):
compared with tracing of 14:7,92, it is very similar but rate is faster. st-t wave changes are a little more marked.
APEX change report:
Compared with the prior study of 1992-07-14 (2 days earlier): Sinus rhythm has changed to sinus tachycardia. New since the prior study: non-specific ischemia. New ST-segment depression in lead I (-0.02 mV to -0.16 mV). New ST-segment elevation in lead V1 (-0.00 mV to +0.17 mV). The heart rate has increased from 72 bpm to 128 bpm (+55 bpm). The PR interval has increased from 174 ms to 224 ms (+50 ms).
Impression: interval change since the prior study — new non-specific ischemia.
Not compared — QT interval not compared: heart rate changed by more than the detectable threshold and the QT interval is rate-dependent, so QTc (Fridericia) is reported instead; QTc (Bazett) not compared: heart rate changed by more than the detectable threshold and Bazett's correction is rate-dependent, so QTc (Fridericia) is reported instead.
Measured: HR 72 → 128 bpm · PR 174 → 224 ms · QRS 94 → 100 ms · QT 348 → 279 ms · QTcF 371 → 359 ms
Review — Concordant but incomplete. Matches both of the cardiologist's points: 'rate is faster' (72 -> 128 bpm, and the rhythm transition SR -> sinus tachycardia) and 'st-t wave changes are a little more marked' (new ST depression in I, new elevation in V1). The reported PR lengthening 174 -> 224 ms is doubtful — PR shortens rather than lengthens as rate rises — and is more likely P-onset error at 128 bpm than a real finding.
Consistency gate: passed
| Elapsed | 2 days (1992-07-26 → 1992-07-29) |
| PTB-XL fold | 8 |
| Prior codes | ASMI, ISCAL, LOWT, SR, STD_ |
| Current codes | ANEUR, ASMI, SR |
| Prior recovered by | exact-date |
Cardiologist's comparison (PTB-XL report, verbatim):
compared with tracing of 26:7:92, it is not significantly changed. evidence of anterior myocardial infarction, probably old, is present. persistent st segment elevation may indicate a ventricular aneurysm.
APEX change report:
Compared with the prior study of 1992-07-26 (2 days earlier): Increased ST-segment elevation in leads V3-V5 (V3 +0.12 mV to +0.27 mV). The QRS duration has decreased from 83 ms to 70 ms (-12 ms).
Impression: new or evolving repolarization change since the prior study. Persisting: anteroseptal myocardial infarction and sinus rhythm.
Measured: HR 95 → 90 bpm · PR 146 → 134 ms · QRS 83 → 70 ms · QT 297 → 305 ms · QTcF 347 → 348 ms
Review — Mixed. The cardiologist called this 'not significantly changed' while noting persistent ST elevation suggesting aneurysm (the annotation adds ANEUR). APEX agreed the ST elevation is there but called it increased in V3-V5 (V3 +0.12 -> +0.27 mV). Both read the same segment; they disagree on whether the change is real. A 0.15 mV shift is twice the fitted threshold, so this is not a threshold failure — it is the harder question of whether a measured change is a clinical one.
Consistency gate: passed
| Elapsed | 1 day (1992-07-29 → 1992-07-30) |
| PTB-XL fold | 8 |
| Prior codes | ANEUR, ASMI, SR |
| Current codes | ALMI, ASMI, IMI, SR |
| Prior recovered by | exact-date |
Cardiologist's comparison (PTB-XL report, verbatim):
compared with tracing of 29:7:92. st segment elevation in anterior chest leads is a little more pronounced, but there has not been significant change in the t waves. t waves are now flat rather than slightly inverted in i. some recent ante
APEX change report:
Compared with the prior study of 1992-07-29 (1 day earlier): Increased ST-segment elevation in lead V2 (+0.15 mV to +0.27 mV). Measurable but sub-threshold ST shift in lead V3, remaining within normal limits.
Impression: new or evolving repolarization change since the prior study. Persisting: anteroseptal myocardial infarction and sinus rhythm.
Measured: HR 90 → 87 bpm · PR 134 → 143 ms · QRS 70 → 66 ms · QT 305 → 306 ms · QTcF 348 → 346 ms
Review — Concordant. 'ST segment elevation in anterior chest leads is a little more pronounced' — APEX: 'Increased ST-segment elevation in lead V2 (+0.15 mV to +0.27 mV)'. Same segment, same direction, same territory, with a number attached. It also correctly declined to call the sub-threshold V3 shift a finding, matching 'there has not been significant change in the t waves'.
Consistency gate: passed
| Elapsed | 1 day (1992-07-30 → 1992-07-31) |
| PTB-XL fold | 8 |
| Prior codes | ALMI, ASMI, IMI, SR |
| Current codes | ILMI, SR |
| Prior recovered by | exact-date |
Cardiologist's comparison (PTB-XL report, verbatim):
compared with tracing of 30:7:92. there is now st segment elevation in v5,6 and to a minimal degree in ii, iii, avf. recent ischaemia or early infarction is possible.
APEX change report:
Compared with the prior study of 1992-07-30 (1 day earlier): New ST-segment elevation in lead V6 (+0.01 mV to +0.10 mV). Measurable but sub-threshold ST shift in lead V2, remaining within normal limits.
Impression: new or evolving repolarization change since the prior study. Persisting: anteroseptal myocardial infarction and sinus rhythm.
Measured: HR 87 → 71 bpm · PR 143 → 144 ms · QRS 66 → 67 ms · QT 306 → 345 ms · QTcF 346 → 364 ms
Review — Concordant but incomplete. The cardiologist: 'there is now st segment elevation in v5,6 and to a minimal degree in ii, iii, avf'. APEX found V6 (+0.01 -> +0.10 mV) and reported it as new, which is right, but missed V5 and the inferior leads the cardiologist himself called minimal. Sensitivity at the margin is exactly what the family-wise ST correction trades away.
Consistency gate: passed
| Elapsed | 4 hours (1994-01-27 → 1994-01-27) |
| PTB-XL fold | 7 |
| Prior codes | HVOLT, LVH, NT_, PAC, SR |
| Current codes | LAFB, NDT, PAC, SR |
| Prior recovered by | earlier-today |
Cardiologist's comparison (PTB-XL report, verbatim):
compared with tracing of earlier today. it is not significantly changed.
APEX change report:
Compared with the prior study of 1994-01-27 (4 hours earlier): No longer present: left ventricular hypertrophy. The PR interval has decreased from 198 ms to 155 ms (-43 ms). The QRS duration has increased from 73 ms to 84 ms (+11 ms).
Impression: PR interval and QRS duration changed beyond the measurable threshold; no new diagnostic finding. Persisting: sinus rhythm.
Measured: HR 57 → 54 bpm · PR 198 → 155 ms · QRS 73 → 84 ms · QT 396 → 406 ms · QTcF 390 → 394 ms
Review — Discordant. The cardiologist: 'it is not significantly changed'. APEX produced three changes — resolved LVH, PR 198 -> 155 ms, QRS 73 -> 84 ms — and all three are almost certainly spurious. The LVH call is detector flicker across the 0.5 boundary; the PR and QRS shifts clear their thresholds (30 ms, 10 ms) but only just. This pair is the clearest illustration of the phase's central difficulty: on a genuinely unchanged study, a system that reports anything at all is wrong.
Consistency gate: passed
| Elapsed | 53 minutes (1996-07-16 → 1996-07-16) |
| PTB-XL fold | 4 |
| Prior codes | AFIB, NDT, STD_ |
| Current codes | 1AVB, DIG, ISC_, LPR, SR, STD_ |
| Prior recovered by | earlier-today |
Cardiologist's comparison (PTB-XL report, verbatim):
compared with tracing of earlier today. atrial fibrillation has reverted to sinus rhythm. st segments are depressed in i, ii, v4,5,6. non-specific but consistent with myocardial ischaemia and digitalis effect.
APEX change report:
Compared with the prior study of 1996-07-16 (53 minutes earlier): Atrial fibrillation has reverted to sinus rhythm. New since the prior study: first-degree AV block. Measurable but sub-threshold ST shift in lead V2, remaining within normal limits.
Impression: interval change since the prior study — new first-degree AV block.
Not compared — PR interval not compared: no P wave detected in the prior study (atrial fibrillation) — PR interval is undefined.
Measured: HR 87 → 67 bpm · PR — → 242 ms · QRS 72 → 74 ms · QT 343 → 368 ms · QTcF 388 → 383 ms
Review — Concordant. The best output in the set. The cardiologist: 'atrial fibrillation has reverted to sinus rhythm'; APEX: the same sentence, unprompted, from the rhythm-transition rule. It added new first-degree AV block, corroborated both by the annotation (1AVB, LPR) and by its own measured PR of 242 ms in the current study — and it correctly refused to report a PR change, because the prior study was in AF and had no P wave to measure from. The persistent ST depression was rightly not called new, since it was present in both.
Consistency gate: passed