Acute coronary syndrome without documented standard modifiable risk factors and the stress hyperglycaemia ratio in an Indonesian percutaneous coronary intervention registry

1. Study design, setting and population

We retrospectively analysed an institutional PCI registry maintained prospectively at a tertiary referral centre in East Java, Indonesia. The registry includes every patient undergoing invasive coronary management; data are entered by trained clinical staff at the time of the procedure. The study period was January 2021–July 2026. Records were eligible when the presentation was STEMI, non-STEMI (NSTEMI) or unstable angina, as assigned by the attending cardiologist using the European Society of Cardiology criteria in force at the time [18]. Unstable angina was defined as rest angina, new-onset or crescendo angina with or without transient ischaemic ECG changes, without a rise of high-sensitivity cardiac troponin above the 99th percentile. Records with a chronic coronary syndrome indication were excluded. When a patient had more than one ACS admission, only the first was kept. The analysis file comprised 1702 first ACS admissions from 4372 registry records (Figure 1). The report follows the STROBE checklist [19].

Figure 1-

All registry fields used in this analysis were complete in all 1702 records. The exception was the three discharge-medication fields, which were missing only for the 53 patients who died in hospital (1649 of 1702 records, 96.9%).

2. Definition and ascertainment of SMuRF status

We used the SWEDEHEART definitions [1]: hypertension, diabetes mellitus, hypercholesterolaemia (recorded in the registry as dyslipidaemia) and current smoking. Each was taken from the history fields documented at the index admission. Patients in whom all four were recorded as absent are described as having no documented SMuRF. We use ‘SMuRF-less’ only as a shorthand for this documented-history category; it does not establish that the risk factors were truly absent, particularly in patients who had had little previous preventive assessment. Patients with at least one recorded SMuRF are described as SMuRF-positive.

The history fields were completed by patient/family interview, referral letter, previous medical records, and current medication. The registry does not record whether patients had undergone blood-pressure, glucose or lipid screening before the index event, nor pre-admission medication.

All 204 patients without a documented SMuRF were admitted between 1 April 2026 and 31 July 2026. None of the 1473 patients admitted between January 2021 and March 2026 met the definition, whereas 204 of the 229 patients (89.1%) admitted from April 2026 did. This occurs due to a change in how risk-factor history fields were completed, a period of systematic inclusion of all ACS patients and retrospective back-entry of earlier records. Because SMuRF status and calendar period are completely aligned, the prevalence estimate and every comparison between groups may reflect differences in recording practice between periods as well as true clinical differences. Period could not be adjusted for statistically.

Because a documented-history definition depends on the intensity of pre-admission preventive care, five pre-specified alternative definitions were examined: hypercholesterolaemia broadened to history or measured low-density lipoprotein (LDL) cholesterol ≥130 mg/dL; diabetes broadened to history or HbA1c ≥6.5%; hypertension broadened to history or admission systolic blood pressure ≥140 mmHg; ever-smoking in place of current smoking; and all four broadenings together. Admission blood pressure during ACS is not evidence of chronic hypertension; this definition was used only to test how sensitive the estimate is to under-ascertainment, not to diagnose hypertension.

3. Glucose, HbA1c and the SHR

SHR was calculated as the registry glucose value divided by estimated average glucose, where estimated average glucose (mg/dL) = 28.7 × HbA1c (%) −46.7 [13]. An SHR above 1.0 indicates that the measured glucose exceeds the patient's chronic glycaemic level. SHR was analysed per one standard deviation (SD) and in tertiles of the analytic cohort.

The registry glucose field is labelled ‘fasting’. We could not verify these details for individual records. We therefore refer to this variable as ‘registry glucose’ rather than admission glucose. Among patients without recorded diabetes (n = 689), the median registry glucose was 153 (125–186) mg/dL and 61.4% had values ≥140 mg/dL, which suggests that many samples were not taken in a steady overnight-fasting state; however, the timing of individual samples relative to admission and PCI cannot be established.

4. Endpoints

The in-hospital outcome was all-cause death during the index admission. Killip class III–IV, Killip class II–IV and LVEF <40% were analysed as markers of index-event severity, not as downstream outcomes, because they are measured at or shortly after presentation, close in time to the glucose sample. Killip class was recorded at first assessment in the emergency department before PCI. LVEF was measured by transthoracic echocardiography using the biplane Simpson method, performed after PCI, at a median 6 hr after admission. For patients with more than one echocardiogram, the first study was used. No endpoint was missing.

5. Data-quality audit and handling of recording-bound values

Before any modelling we audited every continuous variable for heaping at its extreme recorded values. For each variable, we compared the number of records at exactly the minimum and at exactly the maximum value with the mean number at the five distinct values immediately inside each bound. In a distribution without clipping, the extreme value should be less common than its neighbours. A marked excess suggests that values below or above the reporting range, or unavailable values, were recorded at the bound [17]. Values at a bound may still be genuine measurements at an assay-reporting limit; we therefore treated them as uncertain rather than as errors.

HbA1c. The primary glycaemic analyses excluded records with HbA1c at either bound (≤4.0% or ≥15.0%; n = 208), giving an analytic cohort of 1494 patients. Because exclusion may cause selection bias, the main SHR models were repeated (1) retaining the recorded values, (2) treating boundary values as missing, with multiple imputation (20 imputations; predictors in Supplementary Table S2) and (3) treating them as censored, imputing from the same model restricted to 3.0%–4.0% or 15.0%–20.0%. Imputed estimates were combined with Rubin's rules.

Other variables. Heaping was also present in total cholesterol, LDL cholesterol, creatinine (and therefore estimated glomerular filtration rate [eGFR]), GRACE score, heart rate and body-mass index (BMI) (Table 1). For each of these variables we repeated the adjusted SMuRF-group comparisons after excluding records at that variable's bounds. For the SHR models, we repeated Model 2 after excluding records with a boundary value in any of its continuous covariates (LDL cholesterol, eGFR/creatinine or BMI; n = 180 excluded from the analytic cohort). GRACE score was no longer used as a covariate (Section Statistical analysis). Non-glycaemic analyses, including the prevalence and phenotype analyses, used all 1702 records.

For each variable, the number of records at exactly the minimum and maximum observed value is compared with the mean number at each of the five distinct values immediately inside that bound. HbA1c, total cholesterol, LDL cholesterol, creatinine (and derived eGFR), GRACE score, heart rate and BMI show marked excess at the lower bound; registry glucose, systolic blood pressure and LVEF do not. Records with HbA1c at either bound (n = 208) were excluded from the primary glycaemic analyses and handled by three alternative approaches in sensitivity analyses (Supplementary Table S2); records at the bounds of the other affected variables were examined in Supplementary Table S1 and Table 6. Of the boundary values, 94/98 (LDL), 143/156 (total cholesterol), 64/66 (creatinine) and 73/76 (GRACE) occurred in SMuRF-positive patients.

6. Statistical analysis

Proportions are given with Wilson 95% confidence intervals (CIs). Continuous variables are presented as median (interquartile range [IQR]) and compared with the Mann–Whitney U test (two groups) or Kruskal–Wallis test (three groups). Categorical variables are presented as n (%) and compared with the chi-squared test or, when expected counts were small, Fisher's exact test. Equality of variance was tested with the Levene test. eGFR was calculated with the 2021 CKD-EPI creatinine equation [20].

We estimated differences between patients with and without a documented SMuRF using linear or logistic regression, adjusted for age and sex. In the tables, we show variables that form part of the SMuRF definition, or are directly linked to it (blood pressure, lipids, HbA1c, and glucose), separately from variables not part of the classification, because differences in the former are expected by construction. To reconcile the median and mean comparisons of SHR, we also estimated adjusted differences in percentiles by quantile regression.

SHR was entered into logistic regression for each binary endpoint, per one SD and by tertile, in three nested models: unadjusted; Model 1, adjusted for age, sex, STEMI presentation, diabetes and eGFR; and Model 2, additionally adjusted for hypertension, dyslipidaemia, current smoking, BMI, LDL cholesterol and previous heart failure. In the original submission, Model 2 also included the GRACE score. We removed it because Killip class is a component of the GRACE score, so its inclusion in models with a Killip endpoint is circular, and because age, heart rate, systolic pressure and creatinine—also GRACE components—were either entered separately or affected by recording-bound heaping. The Model 2 covariates were chosen as pre-admission patient characteristics that could confound the association between SHR and severity, and none is measured from the index-event haemodynamics. Collinearity was checked with variance inflation factors.

Because there were only 46 deaths in the analytic cohort, all mortality models used Firth penalised likelihood with profile penalised-likelihood CIs and penalised likelihood-ratio tests [21]. Model 1 had about 7.7 events per estimated parameter; Model 2 had about 3.8, which is below conventional thresholds. We therefore regard Model 1 as the primary adjusted mortality model and report Model 2 only as a sensitivity analysis. As a second, parsimonious sensitivity analysis, we adjusted SHR for the GRACE score alone, which is appropriate for mortality because death is not a GRACE component. The trend across tertiles was tested by entering tertile as a single ordinal term (coded 1, 2, 3) in the same model as the tertile odds ratios (ORs); for mortality, the p value is from the penalised likelihood-ratio test. As a post hoc check of non-linearity for mortality, we added a quadratic SHR term and examined SHR <1.0 as a binary exposure.

Interaction between SHR and diabetes, and between SHR and SMuRF status, was tested with a multiplicative term. The added value of SHR over absolute glucose was examined by mutual adjustment and by the area under the receiver operating characteristic curve (AUC), with 95% CIs from 1000 bootstrap samples. The GRACE score was used as a benchmark only for in-hospital death; it was not used as a comparator for the Killip endpoints, because Killip class contributes directly to the score.

Tests were two-sided with a significance level of 0.05. Secondary analyses are exploratory; no adjustment for multiplicity was made, and those p values should be read descriptively. All analyses were performed in Python 3.12 (pandas, statsmodels 0.15, scipy, scikit-learn); the Firth routine was implemented directly and checked against unpenalised maximum likelihood in non-sparse models. All tables and figures were produced by a single script from a single dataset.

7. Relationship to a companion analysis of the same registry

A companion analysis of the same registry [22] examined two clinical phenotypes of ACS (smoking-driven and metabolic) in 1582 patients. The two reports address different questions. The companion analysis does not examine SMuRF-less status, SHR, the recording-bound audit or the association of glycaemic measures with index-event severity and the present report does not examine the smoking-driven and metabolic phenotypes. The cohorts differ because the present analysis keeps only the first ACS admission per patient and requires complete documentation of all four SMuRFs. No outcome, exposure or effect estimate is reported in both papers. The companion manuscript has been disclosed to the journal.

8. Ethics

The study used de-identified registry data collected during routine care. It was approved by the Health Research Ethics Committee, Dr. Saiful Anwar General Hospital, East Java, Indonesia (approval number 400/246/K.3/102.7/2026, issued in August 2026), which waived the requirement for individual informed consent. The study was conducted in accordance with the Declaration of Helsinki.

1. Cohort

The cohort comprised 1702 first ACS admissions (Figure 1). Median age was 60 (52–66) years and 1266 patients (74.4%) were men. The presentation was STEMI in 941 (55.3%), NSTEMI in 441 (25.9%) and unstable angina in 320 (18.8%). Femoral access was used in 85.3%. In-hospital death occurred in 53 patients (3.1%); Killip class III–IV was recorded in 21.5% and LVEF <40% in 21.8%. The median number of vessels treated was 2 (1–3): 1 vessel in 600 patients (35.3%), 2 in 477 (28.0%), 3 in 369 (21.7%) and 4 or 5 in 256 (15.0%). Multivessel PCI (2 or more vessels treated) was performed in 1102 patients (64.7%). The registry's separate ‘multivessel’ field (70.4%) disagreed with the vessel count in 356 records and probably captures angiographic multivessel disease in some; we therefore use the derived definition.

2. Data-quality audit

Eight variables showed clear heaping at the lower recording bound (Table 1, Figure 2). The most severe was HbA1c: 182 records (10.7%) had exactly 4.0%, compared with a mean of 14.0 records at each of the 5 values immediately above it, and a further 26 had exactly 15.0%. Total cholesterol was recorded as exactly 60 mg/dL in 150 records (adjacent-value mean 3.2), LDL cholesterol as 20 mg/dL in 94 (3.2), GRACE score as 40 in 73 (4.2), heart rate as 30 beats/min in 54 (7.6), creatinine as 0.3 mg/dL in 45 (7.6) and BMI as 14 kg/m² in 32 (2.2); eGFR, derived from creatinine, showed the corresponding excess at both bounds. Registry glucose, systolic blood pressure and LVEF showed no such excess.

Figure 2-

An HbA1c of 4.0% corresponds to an estimated average glucose of 68.1 mg/dL, so any glucose value above that gives an SHR above 1.0. The median SHR was 1.90 (1.53–2.35) in the 208 boundary records and 1.29 (0.98–1.69) in the others. Only 16.5% of patients with HbA1c of 4.0% had documented diabetes, whereas all 26 with 15.0% did. Apart from diabetes (26.9% vs 64.1%, p < 0.001), the boundary records were similar to the retained records in age (mean 58.1 years vs 59.5 years, p = 0.054), sex (75.0% vs 74.3% men), STEMI presentation (54.3% vs 55.4%), Killip class III–IV (22.6% vs 21.4%), LVEF <40% (24.0% vs 21.5%) and in-hospital death (3.4% vs 3.1%). None of the boundary records were from a patient without a documented SMuRF; 206 of the 208 were from the period before April 2026, when no such patient was recorded (Section Definition and ascertainment of SMuRF status).

3. Prevalence of ACS without a documented SMuRF

Overall, 204 of 1702 patients (12.0%; 95% CI 10.5–13.6) had no documented SMuRF. One SMuRF was recorded in 253 patients (14.9%), 2 in 464 (27.3%), 3 in 543 (31.9%) and 4 in 238 (14.0%). This overall figure should be read together with the capture period: the proportion was 0% (0/1473) before April 2026 and 89.1% (204/229) from April 2026 to July 2026.

The proportion increased with age, from 4.2% (6/143) below 45 years to 23.2% (23/99) at 75 years or older (p for trend < 0.001; Figure 3A). After adjustment for sex and presentation type, the odds of having no documented SMuRF increased by 39% per decade of age (OR 1.39, 95% CI 1.20–1.62; p < 0.001). The proportion was numerically higher in women (14.0% vs 11.3%), but not significantly after adjustment (OR for male sex 0.78, 0.57–1.08; p = 0.141). It was 15.9% in unstable angina, 11.8% in NSTEMI and 10.7% in STEMI (p = 0.046).

Figure 3-

Estimates under the alternative definitions are shown in Table 2. Adding measured LDL cholesterol ≥130 mg/dL reduced the estimate to 10.1%. Adding HbA1c ≥6.5% did not change it, because all patients without documented diabetes had HbA1c below 6.5%. Adding admission systolic pressure ≥140 mmHg reduced it to 9.4%. Replacing current smoking with ever-smoking reduced it to 3.8%, reflecting the high lifetime smoking exposure in this mainly male cohort (73.9% ever-smokers). With all four broadenings, the estimate was 2.6%.

4. Clinical profile of patients without a documented SMuRF

Baseline characteristics are shown in Table 3 and adjusted differences in Table 4 and Figure 3B. As expected from the definition, patients without a documented SMuRF had lower systolic blood pressure (adjusted difference −20.2 mmHg, −23.6 to −16.7), LDL cholesterol (−38.2 mg/dL, −47.6 to −28.9), total cholesterol (−35.4 mg/dL, −46.9 to −24.0) and HbA1c (−2.23%, −2.57 to −1.90). These differences largely follow from the classification and do not by themselves show a distinct biological phenotype.

Among variables that are not part of the classification, patients without a documented SMuRF were older (+3.3 years, 1.8–4.8), had a lower BMI (−1.5 kg/m², −2.4 to −0.7), a lower heart rate (−4.9 beats/min, −9.1 to −0.7) and a lower GRACE score (−18.7 points, −25.5 to −11.9). Fewer vessels were treated (−0.21, −0.38 to −0.04), and the left anterior descending artery was treated less often (adjusted OR [aOR] 0.64, 0.48–0.87). Multivessel PCI was less frequent (0.46, 0.34–0.62). Presentation was more often with unstable angina than with myocardial infarction (OR for myocardial infarction 0.65, 0.46–0.91), and radial access was used more often (1.48, 1.01–2.17).

Excluding records at the recording bounds of each affected variable did not materially change these comparisons (for example, LDL cholesterol −43.1 mg/dL, −51.8 to −34.3; GRACE score −20.0, −26.6 to −13.3; Supplementary Table S1). Boundary values occurred almost only in SMuRF-positive patients (e.g., 94 of 98 boundary LDL values).

Markers of index-event severity did not differ significantly. Killip class III–IV was recorded in 23.5% of patients without and 21.2% of patients with a documented SMuRF (aOR 1.13, 0.80–1.60; p = 0.489), Killip class II–IV in 40.2% and 40.0% (1.01, 0.75–1.36; p = 0.946), and LVEF <40% in 24.0% and 21.5% (1.14, 0.81–1.62; p = 0.450). Mean LVEF was similar (adjusted difference −0.3% points, −1.9 to 1.3). In-hospital death occurred in 6 of 204 (2.9%) and 47 of 1498 (3.1%) patients (Firth-aOR 0.82, 0.32–1.78; p = 0.637). With 53 deaths, this comparison cannot exclude a moderate difference in either direction, and the absence of a detected difference should not be read as equivalence.

Distribution of the SHR

In the analytic cohort (n = 1494), median SHR was 1.29 (0.98–1.69) (mean 1.45, SD 0.76). SHR was above 1.0 in 1089 patients (72.9%), above 1.5 in 34.3% and above 2.0 in 16.0% (Figure 4A). The tertile boundaries were 1.09 and 1.52 (tertile 1 ≤ 1.09, n = 509; tertile 2 1.10–1.52, n = 492; tertile 3 ≥ 1.53, n = 493). SHR correlated with registry glucose (Spearman ρ 0.61) and inversely with HbA1c (ρ − 0.31). Mean SHR was similar in patients with and without diabetes (1.45 ± 0.77 vs 1.46 ± 0.73; p = 0.081). Characteristics by tertile are shown in Table 5. Patients in tertile 1 more often had diabetes (71.7% vs 60.1% in tertiles 2–3) and had a higher median HbA1c (8.2% vs 6.3%).

Figure 4-

Table 5-

Characteristics of the analytic cohort by tertile of the SHR (n = 1494).

Variable	Tertile 1 (≤1.09) n = 509	Tertile 2 (1.10–1.52) n = 492	Tertile 3 (≥1.53) n = 493	p
Age (years)	60 (53–66)	60 (52–67)	61 (53–67)	0.542
Male sex	375 (73.7)	366 (74.4)	369 (74.8)	0.912
BMI (kg/m2)	24.5 (21.2–28.3)	24.5 (21.3–28.3)	24.5 (21.2–28.3)	0.981
Systolic blood pressure (mmHg)	140 (125–156)	140 (124–155)	141 (126–159)	0.169
Registry glucose (mg/dL)	154 (125–191)	173 (145–245)	310 (216–454)	<0.001
HbA1c (%)	8.2 (7.0–10.1)	6.2 (5.6–8.2)	6.6 (5.2–8.6)	<0.001
SHR	0.88 (0.73–0.99)	1.30 (1.19–1.39)	1.99 (1.71–2.45)	<0.001
eGFR (mL/min/1.73 m2)	78 (54–98)	75 (54–98)	78 (51–98)	0.958
LDL cholesterol (mg/dL)	120 (68–167)	107 (71–153)	117 (75–170)	0.069
Hypertension	356 (69.9)	242 (49.2)	347 (70.4)	<0.001
Diabetes mellitus	365 (71.7)	262 (53.3)	330 (66.9)	<0.001
Dyslipidaemia	290 (57.0)	228 (46.3)	297 (60.2)	<0.001
Current smoking	216 (42.4)	173 (35.2)	218 (44.2)	0.009
No documented SMuRF	36 (7.1)	139 (28.3)	29 (5.9)	<0.001
Previous heart failure	103 (20.2)	69 (14.0)	85 (17.2)	0.034
STEMI presentation	279 (54.8)	277 (56.3)	272 (55.2)	0.886
Myocardial infarction (STEMI or NSTEMI)	410 (80.6)	401 (81.5)	404 (81.9)	0.845
GRACE risk score	138 (101–178)	127 (90–168)	143 (98–175)	0.005
Killip class III–IV	112 (22.0)	100 (20.3)	107 (21.7)	0.789
Killip class II–IV	219 (43.0)	187 (38.0)	196 (39.8)	0.258
LV ejection fraction (%)	52 (48–56)	52 (48–56)	52 (48–56)	0.410
LV ejection fraction <40%	108 (21.2)	105 (21.3)	108 (21.9)	0.961
Vessels treated, n	2 (1–3)	2 (1–3)	2 (1–3)	0.956
LAD treated	342 (67.2)	327 (66.5)	340 (69.0)	0.689
Multivessel PCI (≥2 vessels treated)	330 (64.8)	314 (63.8)	325 (65.9)	0.788
In-hospital all-cause death	23 (4.5)	11 (2.2)	12 (2.4)	0.067

[i] BMI, body mass index; eFGR, estimated glomerular filtration rate; IQR, interquartile range; LAD, left anterior descending artery; LDL, low-density lipoprotein; LV, left ventricular; NSTEMI, non-ST-elevation myocardial infarction; PCI, percutaneous coronary intervention; SHR, stress hyperglycaemia ratio; SMuRFs, standard modifiable cardiovascular risk factors; STEMI, ST-elevation myocardial infarction.

Median (IQR) or n (%). p values from the Kruskal–Wallis test (continuous) or chi-squared test (categorical). Tertiles are sample tertiles of the registry SHR (recorded to two decimals). Abbreviations as in Table 3.

5. SHR, index-event severity and in-hospital death

Event rates across ascending SHR tertiles were 22.0%, 20.3% and 21.7% for Killip class III–IV; 43.0%, 38.0% and 39.8% for Killip class II–IV; 21.2%, 21.3% and 21.9% for LVEF <40%; and 4.5%, 2.2% and 2.4% for in-hospital death (Figure 5A).

Figure 5-

SHR was not associated with any marker of index-event severity (Table 6, Figure 5B). Per SD increase in SHR, the unadjusted OR was 1.01 (0.89–1.14) for Killip class III–IV, 0.95 (0.85–1.05) for Killip class II–IV and 1.06 (0.94–1.19) for LVEF <40%. The revised Model 2, without the GRACE score, gave 1.02 (0.90–1.16), 0.95 (0.85–1.06) and 1.08 (0.96–1.23), respectively (all p > 0.20). All VIFs in Model 2 were below 1.5. Excluding records with boundary covariate values (n = 1314) gave the same results. In tertile analyses no tertile OR differed from 1 and no trend was significant for Killip class III–IV (p = 0.863), Killip class II–IV (p = 0.264) or LVEF <40% (p = 0.835).

ORs versus tertile 1. p for trend: tertile entered as a single ordinal term (1, 2, 3) in the same model (Wald test for logistic models; penalised likelihood-ratio test for Firth models). In-hospital deaths: 23/509 (4.5%), 11/492 (2.2%) and 12/493 (2.4%). The mortality tertile estimates in the original submission were reported incorrectly; the values shown are corrected. These estimates rest on 46 deaths and are exploratory. Models as in Table 6.

For in-hospital death, the continuous estimates were inverse and not significant: OR per SD 0.73 (0.48–1.05) unadjusted, 0.71 (0.45–1.05) in Model 1, 0.69 (0.44–1.02) in Model 2 (sensitivity analysis; about 3.8 events per parameter) and 0.74 (0.49–1.07) after adjustment for the GRACE score. In the tertile analysis, however, mortality was highest in tertile 1. Compared with tertile 1, the Model 1 OR was 0.41 (0.19–0.83) for tertile 2 and 0.50 (0.24–0.99) for tertile 3 (p for trend 0.032). Unadjusted, the trend was not significant (p = 0.055). In post hoc analyses, SHR <1.0 was associated with higher mortality (18 of 394 [4.6%] vs 28 of 1100 [2.5%]; Model 1 OR 2.25, 1.20–4.14), whereas a quadratic term was not significant (p = 0.17). These findings rest on 46 deaths and are exploratory. The mortality tertile ORs in the original submission were reported incorrectly and have been corrected.

Mutual adjustment did not favour either glycaemic measure: for Killip class III–IV, the OR was 0.96 (0.81–1.15) per SD of SHR and 1.01 (0.99–1.02) per 10 mg/dL of registry glucose; for LVEF <40%, 1.11 (0.94–1.32) and 1.00 (0.98–1.01). Discrimination was close to chance for both measures: AUC 0.48–0.51 for SHR and 0.48–0.50 for glucose for the three severity markers (Figure 5C). For in-hospital death the AUC was 0.42 (0.33–0.52) for SHR and 0.44 (0.35–0.53) for glucose, compared with 0.78 (0.72–0.84) for the GRACE score. Interaction between SHR and diabetes was not significant for any severity marker (p = 0.278, 0.475 and 0.467), nor was interaction with SMuRF status (p = 0.875, 0.549 and 0.686).

The four approaches to boundary HbA1c values gave similar results (Supplementary Table S2): per SD of SHR in Model 1, the OR for Killip class III–IV ranged from 1.01 to 1.04, for LVEF <40% from 1.06 to 1.09 and for in-hospital death from 0.69 to 0.85, all with CIs including 1.

6. SHR in patients with and without a documented SMuRF

In the analytic cohort, the 204 patients without a documented SMuRF had lower HbA1c (median 5.9% [5.5–6.0] vs 7.7% [6.1–9.5]; p < 0.001) and lower registry glucose (150 [126–170] vs 204 [151–310] mg/dL; p < 0.001) (Table 7).

The two groups had similar median SHR (1.27 [1.13–1.43] vs 1.30 [0.94–1.79]; Mann–Whitney p = 0.731), but different distributions (Figure 4B, Supplementary Table S3). The SD was 0.25 in patients without and 0.81 in patients with a documented SMuRF (Levene p < 0.001), and the mean was lower (1.30 vs 1.48; age- and sex-adjusted difference −0.17, −0.28 to −0.06; p = 0.003). The adjusted difference in medians was −0.03 (−0.13 to 0.06; p = 0.473). SHR was higher at the 25th percentile (+0.18, 0.11–0.26) and lower at the 90th percentile (−0.80, −1.01 to −0.59) in patients without a documented SMuRF, so the lower mean reflects the long right tail in SMuRF-positive patients rather than a shift of the whole distribution. An SHR above 1.0 was more common in patients without a documented SMuRF (92.6% vs 69.8%; p < 0.001), but values above 1.5 (18.1% vs 36.8%) and above 2.0 (2.0% vs 18.2%) were less common.

Among the 405 SMuRF-positive patients whose HbA1c was below 6.5%, median SHR was 1.55 (1.24–2.11), higher than in patients without a documented SMuRF (1.27; p < 0.001). Within the group without a documented SMuRF, SHR was not associated with Killip class III–IV (aOR per SD 0.98, 0.35–2.74) or LVEF <40% (1.26, 0.47–3.35), but these estimates are imprecise. With the 208 boundary records retained, the adjusted mean difference in SHR between groups would have been −0.25 (−0.36 to −0.13) rather than −0.17 (−0.28 to −0.06), a difference caused by boundary HbA1c values in the SMuRF-positive group.

7. Prevalence and its interpretation

Our estimate lies within the published range: 14.9% in SWEDEHEART [1], 14.4% in an Australian PCI registry [8], 8.6% in two Singaporean cohorts [4,5] and 26.2% in a United States series [23]. However, two features limit its interpretation. First, the estimate depends strongly on the definition: it fell to 3.8% whenever-smoking replaced current smoking and to 2.6% when all four broadenings were applied (Table 2). In a population with high lifetime smoking exposure, much of what the SMuRF-less category captures is former rather than current smoking. We suggest that SMuRF-less prevalence be reported together with at least one broadened definition. Second, and more importantly, all patients without a documented SMuRF were admitted in the last 4 months of the study period, when they formed the large majority of admissions. This pattern is not biologically plausible and indicates that SMuRF status in this registry is closely linked to how, and when, records were captured. The overall 12.0% is therefore not a reliable estimate of the population prevalence, and the group comparisons may partly reflect differences in recording practice between periods.

The absence of a documented risk factor is also not the same as its confirmed absence. Patients with little previous contact with preventive care may have undiagnosed hypertension, diabetes or dyslipidaemia. In our data, adding measured LDL cholesterol or admission blood pressure reduced the estimate only modestly, and no patient without documented diabetes had an HbA1c ≥6.5%, but these measurements were taken during the acute event and cannot confirm chronic status. The age gradient we observed resembles series in which SMuRF-less patients were of similar or greater age [8] rather than younger [2]; in our setting it may reflect the high prevalence of current smoking among younger patients.

8. Clinical profile and index-event severity

Lower blood pressure, lipid and glycaemic values in a group defined by the absence of hypertension, dyslipidaemia and diabetes are expected and do not show an independent phenotype. The more informative findings concern variables outside the definition: older age, lower BMI and GRACE score, fewer vessels treated and more unstable angina. These findings are consistent with a lower atherosclerotic burden, but they could also reflect the different capture period. Killip class, LVEF and in-hospital death did not differ significantly. This differs from SWEDEHEART [1] and the global meta-analysis [7] and is closer to the Australian PCI registry [8]. Three explanations are possible. First, with 53 deaths our CI (0.32–1.78) excludes the roughly twofold excess seen in SWEDEHEART but not a smaller one. Second, our cohort includes only patients who reached an invasive centre and underwent PCI, which excludes patients with out-of-hospital arrest or very severe presentations. Third, the SWEDEHEART excess was most evident at 30 days, whereas our registry has no follow-up after discharge. In an earlier analysis of this registry [22], patients without SMuRFs appeared more likely to have LVEF <40%; in the present, complete dataset with age and sex adjustment, this association was not found (OR 1.14, 0.81–1.62).

9. The SHR

We did not detect an association between SHR and the markers of index-event severity, and this result was consistent across models, tertiles, discrimination analyses and the four approaches to boundary HbA1c values. The absence of a detected association does not show that SHR has no prognostic value. Several features of our data could explain the difference from the positive literature [14,15,16].

First, the timing of the glucose sample is uncertain. The SHR concept assumes a glucose value that reflects the acute response to the index event. Our glucose field is labelled fasting, and some samples may have been drawn after PCI, fluids, fasting, insulin or haemodynamic stabilisation, all of which would move the ratio towards 1 and bias associations towards the null. This uncertainty concerns the main exposure and could by itself explain our findings. Second, our endpoints are mainly markers of the severity of the index event, recorded close in time to the glucose sample, whereas the positive literature is based largely on mortality and adverse events over 30 days to several years. Third, with 46 deaths in the analytic cohort, the power for mortality was limited.

The inverse pattern for mortality should be interpreted cautiously. Patients in the lowest tertile more often had diabetes and had higher HbA1c, so a low SHR here mainly identifies patients whose glucose value was below their chronic glycaemic level. This may reflect poorly controlled diabetes, residual confounding or glucose sampled after treatment, rather than a protective effect of higher SHR. U- or J-shaped associations between SHR and mortality have been described, but our data, with few events and uncertain sampling time, cannot establish one.

The results also show that SHR is sensitive to how it is constructed. Because the denominator comes from a single laboratory value, recording 10.7% of HbA1c values at the reporting floor inflated the ratio without any physiological change. Registry studies using SHR, or any derived ratio, should report the frequency of input values at their recording bounds and test how their handling affects the results [17].

10. SHR in patients with and without a documented SMuRF

Patients without a documented SMuRF had lower chronic glycaemia by definition, and their SHR distribution was narrow: most had an SHR slightly above 1.0, and very few had a high value. Their median SHR was similar to that of SMuRF-positive patients, but their mean SHR was lower because extreme values were largely confined to SMuRF-positive patients. The two comparisons are therefore consistent. Among patients with HbA1c below 6.5%, those with at least one SMuRF had a higher median SHR (1.55 vs 1.27), suggesting that the classical risk factors may be associated with a larger acute glycaemic response even when chronic glycaemia is normal. Because of the capture-period issue and the uncertain glucose timing, this should be regarded as hypothesis-generating.

11. Limitations

This was a single-centre retrospective analysis of patients selected for invasive management, and patients who died before catheterisation were not included. The most important limitation is that all patients without a documented SMuRF were recorded in one 4-month period, which makes SMuRF status inseparable from the recording period; this limits the prevalence estimate and all between-group comparisons. SMuRF status was based on documented history, and information on previous screening and pre-admission medication was not available. The timing and fasting status of the glucose sample could not be verified, and the timing of Killip assessment and echocardiography relative to PCI was not uniformly recorded as described in Section Endpoints. Several continuous variables showed heaping at recording bounds; sensitivity analyses did not change our conclusions, but source verification was not possible and was limited to a sample. The registry multivessel field was internally inconsistent, and angiographic disease extent beyond the treated vessels was not available. Follow-up after discharge was not available, and in-hospital mortality was low (3.1%), limiting power. Unmeasured confounders include inflammatory markers, lipoprotein(a) and socioeconomic status.