Why Analyses of Achieved Sodium Correction Rates in Hyponatremia Studies Do Not Identify Causal (Treatment) Effects

Introduction The optimal correction rate for hyponatremia has long been debated. Correct too slowly and life-threatening brain edema may occur. Correct too quickly and osmotic demyelination may cause irreversible damage. Several observational studies have attempted to resolve this clinically relevant question. A recent meta-analysis and a recent large cohort study concluded that rapid correction is associated with reduced risk for neurological events and mortality.1,2 The central question is whether these studies support a causal interpretation of their results, that is, that pursuing rapid correction improves outcomes, and whether such findings should inform clinical guidelines. Given the observational nature of the studies, baseline confounding is an important concern, but we argue that the causal interpretation is further limited by two other issues. First, analyses that examine the associations between achieved sodium correction rates and outcomes are susceptible to selection bias and time-varying confounding. Second, they compare achieved correction rates rather than predefined treatment strategies. In practice, physicians can only implement treatment strategies designed to achieve particular correction rates; they cannot enforce a certain correct rate. We will first describe these inherent limitations. We then outline how observational studies that emulate randomized trials (i.e., target trials3) can, in principle, address the causal question of interest. Why Current Observational Studies Fail To Do the (Causal) Job Current observational studies on hyponatremia typically compare outcomes across achieved sodium correction rates. This gives rise to three important constraints that limit their ability to provide causal evidence about optimal hyponatremia management. First, analyzing patients according to achieved correction rate introduces bias beyond residual baseline confounding. Patients are classified by the achieved sodium correction rate, that is, a post-treatment variable (sodium 24 hours after baseline or change in sodium) that reflects response to therapy. Patients who fail to achieve faster correction are often more severely ill at 24 hours after baseline or have poorer physiological reserve, factors that also influence mortality. As a result, more severely ill patients are included in the low-correction rate group because correction rate is directly related to disease severity. This analysis therefore introduces bias because of time-varying confounding, which can make fast correction appear beneficial, even when the treatment strategy itself would have no true advantage. This bias cannot be resolved by adjustment for baseline covariates alone because it arises through postbaseline covariates (such as disease severity at 12 and 24 hours). This time-varying confounding occurs even in a perfectly conducted randomized trial if patients were analyzed according to achieved correction rate rather than the assigned treatment strategy, highlighting that it is conceptually distinct from baseline confounding. This bias could potentially be remedied by adjusting for time-varying confounding, although it is likely very hard to adjust for all relevant time-varying variables such as disease severity. In addition, patients who die before the second measurement cannot be included in the analysis, which introduces selection bias.4 Second, the results of these studies are not clinically actionable. Patients are categorized according to the sodium correction rate actually achieved, which represents a downstream consequence of treatment and patient health status rather than the treatment itself. In practice, clinicians select strategies with the aim of influencing rates but cannot prescribe a specific correction rate per se. Consequently, these studies address a question that cannot readily translate into real-world decision-making. Even if one argues that intended correction rates are the true therapeutic target, this ultimately leads back to a more fundamental question: Which treatment strategies can reliably achieve those correction rates, whether rapid or gradual? Third, the associations have an ambiguous interpretation. Patients with the same achieved correction rate may have received different treatment strategies, rarely specified or standardized in observational studies. For example, an 8–12 mEq/L per 24 hours increase might result from boluses of hypertonic saline, a continuous infusion, or combinations with adjunctive medications. In addition, different protocols may be in place regarding frequency of sodium monitoring and strategies to handle overcorrection. Each strategy may have a different probability to reach the intended correction rate, may have a different risk of overcorrection, and therefore potentially a different causal effect on outcomes studied. This underlines that different strategies, even when targeting the same correction rate, should be investigated for potential treatment heterogeneity. In addition, validity depends on the appropriate selection and measurement of outcomes. End points must not only be well-defined and accurately ascertained but also plausibly related to the exposure of interest. In hyponatremia, outcomes such as osmotic demyelination syndrome more directly meet these criteria, although their ascertainment remains challenging and may be prone to misclassification when based on administrative codes or inconsistently obtained imaging. By contrast, all-cause mortality is often driven by underlying disease processes that also influence correction rates and prognosis. Without adequate information on the timing and circumstances of outcomes, observed associations may reflect illness severity rather than a causal effect of the correction strategy. The Target Randomized Trial To clarify which causal question can meaningfully be answered, it is useful to consider what an ideal randomized controlled trial on hyponatremia correction rate would look like. Such a trial would randomize patients to distinct treatment strategies, each designed to achieve a specific correction goal (e.g.,