Triage a Table 1 before you trust Table 2
Interactive
Learning objective: by the end of this lab, you will be able to scan a Table 1 for baseline imbalance, decide whether an unadjusted outcome comparison is credible, and name which characteristics threaten validity.
Every observational paper asks you to trust that the groups were comparable before the intervention or exposure. Table 1 is where authors show their work. Read the table below, make your call, then reveal the flagged rows.
| Characteristic | Standard discharge (n=120) | Enhanced pathway (n=115) |
|---|---|---|
| Age, mean (SD) | 71.2 (9.4) | 63.8 (10.1) |
| Female, n (%) | 58 (48%) | 61 (53%) |
| Charlson index, median [IQR] | 4 [3–6] | 2 [1–3] |
| Prior admissions (12 mo), mean (SD) | 2.4 (1.6) | 0.7 (1.0) |
| Diabetes, n (%) | 44 (37%) | 29 (25%) |
Step 1: Ignore the outcome for a moment
Table 2 is where authors show the effect. Table 1 is where they show whether the comparison was fair to begin with. If baseline characteristics diverge on age, comorbidity, or prior utilisation, the outcome gap may reflect who landed in each group, not what the pathway did.
Step 2: Decide what "meaningfully different" looks like
Small numeric noise happens. A seven-year age gap, a two-point Charlson difference, and a threefold gap in prior admissions are not noise. They describe different patient populations wearing the same group labels.
Step 3: Match your scepticism to the imbalance
One mildly imbalanced row might be fixable with adjustment. Several imbalanced rows on factors that drive the outcome mean you should distrust an unadjusted comparison, or at minimum demand propensity matching, regression adjustment, or a study design that fixes the problem upstream.
What to take away
Table 1 is a triage tool, not décor. "Need more info" is a legitimate answer when imbalance is borderline. When age, comorbidity, and utilisation all point the same direction, trusting Table 2 without adjustment is how confounding survives peer review.