Paperonium

← Back to the plain-language case

Significant C. difficile reduction in abstract contradicts Table 4: 4 vs 0 events, P = 1

BMC Med (2025) · doi:10.1186/s12916-025-04130-x · PMC12219783

Summary

This retrospective claims-database study compares flomoxef with cefmetazole for hospitalised urinary tract infection. The Abstract, the Results and the Discussion all state that C. difficile infection was significantly less common with flomoxef, and the Results give P = 0.03 and an incidence of 1.1% in the cefmetazole group. The paper's own Table 4 reports 4 cases among 4796 cefmetazole patients (0.1%) and 0 among 673 flomoxef patients, with P = 1. Recomputation from those counts gives a two-sided Fisher P of 1.00, and no standard test comes close to 0.03. The safety claim in the Abstract is therefore not supported by the paper's data. The text also gives the two kidney-failure P values the other way round from the tables, though that result is significant on every test. The preprint of the same study reported 63 vs 2 cases (P = 0.04) under a broader case definition. This fits the text not having been fully updated when Table 4 was revised.

What the paper reports

Abstract, Results:

Moreover, the flomoxef group demonstrated a significantly lower frequency of adverse events such as C. difficile infection and renal failure.

Results, "Safety outcome" (the whole paragraph):

The incidences of AEs related to the gastrointestinal, renal, and hepatic systems and hypersensitivity reactions were analysed (Table 4). Renal failure was the most reported AE in the cefmetazole group, affecting 7.5% of patients. This was followed by hepatic dysfunction, occurring in 2.3% of the cases. Furthermore, gastrointestinal AEs included C. difficile infection in 1.1% of the patients in the cefmetazole group. In the flomoxef group, renal failure (3.6%) was also the most common adverse event, followed by hepatic dysfunction (2.5%) and other forms of kidney dysfunction (0.9%). C. difficile infection (P = 0.03) and kidney failure (P = 0.04) were significantly less common in the flomoxef group than in the cefmetazole group. In the safety analysis of Dataset 3, similar to the findings in Dataset 1, the cefmetazole group showed a significantly higher incidence of kidney failure compared to the flomoxef group (P < 0.01) (Additional File 6).

Discussion, opening paragraph:

The incidence rates of C. difficile infection and renal failure after therapy were significantly lower in the flomoxef group than in the cefmetazole group.

Table 4, "Safety endpoints of Dataset 1" (selected rows; columns are cefmetazole n = 4796, flomoxef n = 673, P value):

Non-C. difficile diarrhoea [% (N)] | 23 (0.5) | 0 (0.0) | 0.14

C. difficile infection [% (N)] | 4 (0.1) | 0 (0.0) | 1

Kidney failure [% (N)] | 360 (7.5) | 24 (3.6) | < 0.01

Other kidney dysfunction [% (N)] | 54 (1.1) | 6 (0.9) | 0.73

Additional File 6, "Safety Endpoints of Dataset 3" (propensity-matched, cefmetazole n = 580, flomoxef n = 580), as printed in the publisher's file:

Clostridium difficile infection [% (N)] | 3 (0·5) | 0 (0·0) | 1

Kidney failure [% (N)] | 36 (6·2) | 20 (3·4) | 0·04

Methods, "Patient outcomes" and "Statistical analyses":

Safety outcomes were analysed using Dataset 1.

For safety outcomes, the frequencies were calculated and descriptively analysed for gastrointestinal, renal, hepatic, and hypersensitive AEs.

Statistical tests were two-tailed, with a p < 0.05 considered statistically significant, and performed using R version 4.2.1 (R Foundation for Statistical Computing, Vienna, Austria).

Additional File 4 defines C. difficile infection as "A04.7 Enterocolitis due to Clostridium difficile and include the prescription of oral vancomycin or metronidazole", recorded from the index date to 1 month.

What the paper's own numbers show

Method. The Methods do not name the test used for the safety outcomes, so for each 2×2 table I computed the three standard two-sided tests: Fisher's exact test, Pearson's chi-square with Yates' continuity correction (the default of R's chisq.test for 2×2 tables), and Pearson's chi-square without correction. The inputs are the event counts and group sizes printed in Table 4 and Additional File 6. As a check on the method, I first recomputed every P value in both tables. Yates' chi-square reproduces each printed value in Table 4 to the printed precision: 0.138 (printed 0.14), 1.00 (1), 2.5 × 10⁻⁴ (< 0.01), 0.727 (0.73), 0.758 (0.76) and 1.00 (1). So the tables appear to have been produced with R's default test, and they are internally consistent. The one exception is a minor point in Additional File 6, noted below. Every number here is printed by recompute.py.txt.

Quantity Location Reported Recomputed from the paper's numbers
C. difficile P, Dataset 1 (4/4796 vs 0/673) Results text P = 0.03, "significantly less common" Fisher 1.00; Yates 1.00; uncorrected chi-square 0.454. One-sided Fisher 0.591
C. difficile P, Dataset 1 Table 4 1 Agrees (Fisher 1.00, Yates 1.00)
C. difficile incidence, cefmetazole Results text 1.1% 4/4796 = 0.083% (Table 4 prints 0.1). 1.1% of 4796 would be 53 cases
Row printed as 1.1% in Table 4 Table 4 Other kidney dysfunction 54 (1.1) 54/4796 = 1.13%
C. difficile P, Dataset 3 (3/580 vs 0/580) Additional File 6 1 Fisher 0.249; Yates 0.248; uncorrected 0.083. Not significant on any test, though not 1
Smallest split that could give P < 0.05 (derived) — With 0 flomoxef cases, at least 28 cefmetazole cases (0.58%) are needed for a two-sided Fisher P < 0.05 (P = 0.042)
Kidney failure P, Dataset 1 (360/4796 vs 24/673) Results text P = 0.04 Fisher 7.2 × 10⁻⁵; Yates 2.5 × 10⁻⁴; uncorrected 1.8 × 10⁻⁴
Kidney failure P, Dataset 1 Table 4 < 0.01 Agrees
Kidney failure P, Dataset 3 (36/580 vs 20/580) Results text P < 0.01 Fisher 0.039; Yates 0.040; uncorrected 0.028
Kidney failure P, Dataset 3 Additional File 6 0.04 Agrees
C. difficile, preprint Table 4 (63/4796 vs 2/673) medRxiv v1 63 (1.3) vs 2 (0.3), P = 0.04 1.31% vs 0.30%; Yates 0.037; Fisher 0.021; uncorrected 0.023

There are four C. difficile cases in 5469 patients. If risk were equal in the two groups, the expected number in the flomoxef group would be 4 × 673/5469 = 0.49. So the observed split, all four in the larger cefmetazole group, is the single most likely outcome under the null, with probability 0.591. That is why the two-sided exact P is 1. The P = 0.03 in the text cannot be reached from these counts with any of the usual tests, one-sided or two-sided. It also cannot be reached from the Dataset 3 counts, where the best case is the uncorrected chi-square at 0.083.

The two kidney-failure P values in the text are the reverse of the tables. The text's "P = 0.04" for Dataset 1 is the Dataset 3 value (Additional File 6; recomputed 0.040). Its "P < 0.01" for Dataset 3 matches Dataset 1 (Table 4; recomputed 2.5 × 10⁻⁴). Both are below 0.05, so the conclusion on kidney failure is unchanged.

Figure

Reported and recomputed P values for the safety outcomes on one log axis

Figure 1. Reported and recomputed two-sided P values for C. difficile infection and kidney failure. Each outcome and dataset has two rows. The upper row shows the values the paper reports: filled orange circles for the Results text and open orange diamonds for Table 4 or Additional File 6. An arrow marks a value reported only as "< 0.01". The lower row shows the values recomputed from the printed counts: the blue dot is Fisher's exact test, and the blue line spans Fisher's test, Yates-corrected chi-square and uncorrected chi-square. The dashed line marks P = 0.05, the paper's threshold. For C. difficile in Dataset 1, the text's P = 0.03 lies on the significant side of the threshold, while Table 4 and every recomputed test sit at or near P = 1. In Dataset 3 every recomputed value is also above the threshold. For kidney failure, the text's values for the two datasets are the reverse of the tables. The direction of the result is unaffected.

Preprint and published versions of the C. difficile result

Figure 2. The C. difficile result in the preprint and in the published article. In the medRxiv preprint (v1, 2024-07-17), C. difficile infection was defined by ICD-10 code A04.7 alone. Table 4 reported 63 (1.3%) vs 2 (0.3%) cases, P = 0.04, which Yates' chi-square reproduces (0.037). In the published article the definition also requires oral vancomycin or metronidazole (Additional File 4), and Table 4 reports 4 vs 0 cases, P = 1. The Abstract, Results and Discussion of the published article still describe a significant reduction. The text's 1.1% and P = 0.03 match the C. difficile row of neither version.

Why it matters

The Abstract names a lower frequency of C. difficile infection as one of two safety advantages of flomoxef, and the Discussion repeats it in its summary of findings. C. difficile infection is a clinically important outcome in antibiotic choice. A reader of the Abstract would conclude that the study found evidence of lower risk with flomoxef. The paper's own table shows four events in total, none of them in the flomoxef group. That count gives no evidence either way: the study has essentially no power for this outcome, and the observed split is what equal risk would most often produce. The Discussion also notes that median hospital stay with flomoxef was about half that with cefmetazole, so adverse events had a shorter in-hospital window in which to be recorded in the flomoxef group. That is a further reason for caution with any between-group difference in these counts.

The error does not affect the paper's formal conclusions in the Abstract and main text, which rest on length of stay, 28-day mortality, carbapenem use and the in vitro data. The kidney-failure difference is significant on every test in both datasets, so the second safety claim stands, apart from the swapped P values. The 1.1% figure overstates the cefmetazole incidence about thirteen-fold (1.1% vs 0.083%).

Explanations considered

A minor, separate point: Additional File 6 prints P = 1 for C. difficile in Dataset 3 (3/580 vs 0/580), where Fisher's test and Yates' chi-square both give 0.25. This does not change any conclusion, since neither value is significant.

Status

As of 2026-10-24:

Suggested correction

  1. Remove the statement that C. difficile infection was significantly less frequent with flomoxef from the Abstract (Results), the Results "Safety outcome" paragraph and the opening paragraph of the Discussion. Report instead that C. difficile infection occurred in 4 of 4796 cefmetazole patients (0.1%) and none of 673 flomoxef patients (P = 1), too few events to compare the groups.
  2. Correct the cefmetazole C. difficile incidence in the Results from 1.1% to 0.1% (4/4796).
  3. Swap the kidney-failure P values in the Results to match the tables: P < 0.01 for Dataset 1 (Table 4) and P = 0.04 for Dataset 3 (Additional File 6).
  4. Name the test used for the safety comparisons in the Methods (the printed values match Pearson's chi-square with Yates' continuity correction). Check the Dataset 3 C. difficile P value in Additional File 6 (3 vs 0 of 580 gives P ≈ 0.25, not 1).
  5. Optionally, state in the Methods that the C. difficile case definition requires oral vancomycin or metronidazole in addition to ICD-10 A04.7, so readers comparing with the preprint can see why the counts differ.