Cox Regression | Recurrence After 2 Methods of Atrial Fibrillation Ablation

When P values are lined up for each subgroup, significant and nonsignificant rows are mixed, and it is easy to misread in which population there is a difference. This case study uses 2 methods of catheter ablation for atrial fibrillation. I compared recurrence with Cox regression and read 6 subgroups with tests of interaction and a forest plot.

The study setting, disease, variables, and numbers in this article were newly created by me to explain how the analysis was carried out. The numbers were calculated from a randomly generated dataset and do not reflect the content or circumstances of any actual client project.

827 patients at 2 hospitals, with the type of atrial fibrillation unbalanced between the methods

The study was a retrospective cohort of patients who underwent catheter ablation for atrial fibrillation for the first time at 2 hospitals: 827 patients after exclusion of 3 whose follow-up ended within 3 months after the procedure. The method was radiofrequency in 516 patients and cryoballoon in 311. The outcome was recurrence of atrial arrhythmia more than 3 months after the procedure, and there were 231 recurrences (27.9%) during follow-up of up to 30 months.

The 2 methods were chosen by the physicians, and persistent atrial fibrillation was present in 46.7% with radiofrequency and 26.7% with cryoballoon, an imbalance with a standardized mean difference of -0.42 (Table 1). Left atrial diameter was also smaller and disease duration shorter with cryoballoon.

Table 1. Characteristics of the patients according to method
Radiofrequency (n=516) Cryoballoon (n=311) Standardized mean difference
Age (years) 66.1±10.2 67.3±9.7 0.12
Female 155 (30.0) 110 (35.4) 0.11
Persistent atrial fibrillation 241 (46.7) 83 (26.7) -0.42
Left atrial diameter (mm) 40.9±5.6 39.7±5.7 -0.20
Heart failure 74 (14.3) 42 (13.5) -0.02
eGFR (mL/min/1.73 m²) 67.0±14.9 66.3±15.4 -0.04
Disease duration (years)† 2.1 (1.1 to 4.1) 1.8 (0.9 to 3.1) -0.22
Hospital B 188 (36.4) 166 (53.4) 0.35
* Values are mean±SD or no. (%). The standardized mean difference is cryoballoon minus radiofrequency, and an absolute value above 0.1 is read as an imbalance. eGFR denotes estimated glomerular filtration rate.
† Median (interquartile range).

Handling differences in length of follow-up with Cox regression, and reading subgroups from interactions

The length of follow-up differed among patients, from 8 to 30 months, and a simple comparison of the proportions with recurrence would miss recurrences in patients with short follow-up. I therefore used Cox regression, which handles the time to recurrence, adjusted for 7 factors fixed in advance (age, sex, type of atrial fibrillation, left atrial diameter, heart failure, eGFR, and disease duration), and treated hospital as a stratum.

Six subgroups were fixed in advance: age, sex, type of atrial fibrillation, left atrial diameter, heart failure, and eGFR. Even if P<0.05 in one population and P≥0.05 in the other, that is not evidence that the hazard ratio differs between the 2 populations, so I read the subgroups not from the P values within each population but from the test of interaction between method and subgroup.

Furthermore, when as many as 6 are examined, one may stand out by chance even if there is no difference. I decided in advance not to apply a correction but, in exchange, to treat any row that stood out as an exploratory result, to be confirmed in a subsequent study.

The horizontal axis of the forest plot was on a log scale. Hazard ratios of 0.5 and 2.0 represent differences of the same size, so on an ordinary scale only the side above 1 would be drawn long.

The overall hazard ratio was 1.16, and no interaction was shown

Kaplan–Meier curves of freedom from recurrence up to 30 months after the procedure for the radiofrequency and cryoballoon methods, with the number at risk below
Figure 1. Kaplan–Meier curves of freedom from recurrence, drawn before the Cox regression. Recurrences within 3 months after the procedure are not counted.
Table 2. Results of Cox regression for recurrence
Unadjusted Adjusted for 7 factors
Factor Hazard ratio (95% CI) P value Hazard ratio (95% CI) P value
Cryoballoon (vs. radiofrequency) 0.96 (0.73 to 1.26) 0.76 1.16 (0.88 to 1.54) 0.30
Age (per 10 years) 1.05 (0.92 to 1.21) 0.47
Female 1.02 (0.77 to 1.35) 0.87
Persistent (vs. paroxysmal) 1.90 (1.43 to 2.52) <0.001
Left atrial diameter (per 5 mm) 1.14 (1.01 to 1.29) 0.03
Heart failure 1.42 (1.02 to 1.97) 0.04
eGFR (per 10) 0.91 (0.83 to 1.00) 0.06
Disease duration (per 1 year) 1.05 (1.00 to 1.09) 0.03
* Analysis of 827 patients and 231 recurrences. Both models treated hospital as a stratum.
† The test of proportional hazards with Schoenfeld residuals gave P=0.22 for the method term and P=0.68 for the model as a whole.

Freedom from recurrence at 24 months was 65.0% with radiofrequency and 67.4% with cryoballoon, and the curves almost overlapped (Figure 1). The hazard ratio for cryoballoon was 0.96 (95% CI, 0.73 to 1.26) without adjustment and 1.16 (0.88 to 1.54; P=0.30) after adjustment for the 7 factors (Table 2).

This shift with adjustment is explained by the cryoballoon group having more patients with paroxysmal atrial fibrillation, who have fewer recurrences; the hazard ratio for persistent atrial fibrillation was 1.90 (1.43 to 2.52). The test with Schoenfeld residuals gave P=0.22 for the method term, with no sign of a violation of proportional hazards.

Forest plot of the hazard ratios and 95% confidence intervals for cryoballoon as compared with radiofrequency on a log horizontal axis, for the overall population and 12 rows of 6 subgroups
Figure 2. Hazard ratios overall and in the 6 subgroups. The column on the left shows the number of recurrences and patients, and the value in parentheses in each heading is the P value for interaction. eGFR denotes estimated glomerular filtration rate.

In the subgroups, the P values for the 6 interactions were 0.20 to 0.93, and the difference between the methods could not be said to vary by population (Figure 2). By type of atrial fibrillation as well, the hazard ratios were almost the same, at 1.14 for paroxysmal and 1.19 for persistent.

The row for patients with heart failure, with only 46 recurrences, had a confidence interval of 0.56 to 2.22, with the upper limit 4.0 times the lower, and it lacks the precision to say anything about this population alone. The row for an eGFR under 60 stands out at 1.44, but the interaction had P=0.20, within the range of 1 of 6 departing by chance.

Taken together, the statistical conclusion is that I could not conclude that recurrence differed between the 2 methods, and that anything from 12% fewer to 54% more recurrences with cryoballoon cannot be ruled out. The method was chosen by the physicians, and there may be reasons for the choice beyond the 7 factors adjusted for, so this ratio cannot be read as a difference due to the method itself.

Work performed and estimated fee

Item Details
Analysis method Comparison of the 2 methods with Cox regression (hospital treated as a stratum, proportional hazards checked with Schoenfeld residuals), and the Kaplan–Meier method
Adjustment for background Adjusted by entering 7 factors in the model (age, sex, type of atrial fibrillation, left atrial diameter, heart failure, eGFR, and disease duration)
Additional analyses 6 subgroup analyses and tests of interaction
Tables prepared 1 table of the characteristics according to method, 1 table of the Cox regression results; 2 tables in total
Figures prepared 1 figure of Kaplan–Meier curves, 1 forest plot; 2 figures in total
Deliverables Analysis report (reasons for the choice of methods and discussion of the results), tables of results (Excel), figures (PNG and TIFF), and a draft of the statistical analysis section for the paper. Analysis code is available on request for an additional fee (¥22,000, incl. tax)
Estimated time About 1 week from receipt of the data
Estimated fee ¥154,000 (incl. tax)

The table lists the main analyses only. I also carry out the finer analyses needed to follow a sound statistical process, such as checking distributions and cleaning the data; these, together with after-care with no time limit (answers to your questions after delivery), are included in the fee.

Data Scientist Kenichiro Suzuki

Data ScientistKenichiro Suzuki

Free consultationSee fee examples

Recent Posts

Linear Mixed Model | 6-Minute Walk Distance in COPD Rehabilitation

A linear mixed model comparing 6-minute walk distance at four time points with missing data. How to read a non-significant group-by-time interaction.

Cox Regression and Multiple Imputation | Mortality in Hemodialysis

Cox regression of 8 factors and all-cause mortality in new hemodialysis patients, with multiple imputation for missing grip strength and other values.

ROC Analysis | Comparing the AUCs of 4 Blood Markers for Bacteremia

ROC analysis of the AUCs of 4 markers for bacteremia: all pairs tested with the paired DeLong test and corrected with the Benjamini–Hochberg procedure.

Kruskal–Wallis Test | Fatigue and Pain in 5 Rheumatoid Arthritis Groups

Kruskal–Wallis test with corrected pairwise comparisons of fatigue and pain scores among 5 rheumatoid arthritis treatment groups, plus factor analysis.

RMST | Pancreatic Cancer Chemotherapy When Survival Curves Cross

RMST (restricted mean survival time) comparing 2 first-line chemotherapies for advanced pancreatic cancer when survival curves cross, with adjustment.

Multiple Regression | Factors Associated With Dementia Caregiver Burden

Multiple regression with 8 prespecified factors for burden and quality of life in family caregivers of people with dementia, with principal components.

Multiple Regression | Monthly Antimicrobial Use and Resistance Over Time

Multiple regression of monthly antimicrobial use and the proportion of resistant Pseudomonas aeruginosa, after unit root tests and differencing.

Canonical Correlation Analysis | Psychosocial Scales and Diabetes Self-Care

Canonical correlation analysis of 4 psychosocial scales and 5 self-care behaviors in type 2 diabetes, with the size checked by 10-fold cross-validation.