Reassessment of the Collaborative Normal-Tension Glaucoma Study: Statistical Evidence and Implications for Current Management

Abstract

Introduction

The Collaborative Normal-Tension Glaucoma Study (CNTGS) is frequently cited as evidence that a 30% reduction in intraocular pressure (IOP) slows progression in normal-tension glaucoma (NTG). This study re-examines the statistical methodology of CNTGS to assess how its conclusions are supported by the data.

Methods

This study reviews the CNTGS design with emphasis on survival analysis methodology, including the definition of time zero, censoring rules, and intention-to-treat (ITT) versus per-protocol comparisons. Particular attention is given to the post hoc redefinition of baseline and the handling of cataract-related visual decline, assessing their impact on the reported treatment effect.

Results

CNTGS shifted the analytical baseline for the treatment group to the point of IOP stabilization, thereby excluding early progression events and introducing immortal time bias. Additionally, cataract-related visual decline, more frequent in the treatment group, was censored rather than treated as a competing risk or time-dependent covariate. These methodological choices reduced the number of counted progression events in the treatment arm. Although the adjusted per-protocol analysis yielded a statistically significant treatment effect, this effect disappeared under the original ITT analysis, which included all randomized eyes from time zero and all progression events.

Conclusion

The potential treatment benefit reported in CNTGS depended largely on post hoc analytical modifications, whereas the original ITT analysis did not support a statistically significant effect of IOP reduction. These findings highlight the importance of transparent survival analysis methods and strict adherence to ITT principles in future NTG trials. Well-designed prospective studies that avoid immortal time bias and model treatment-related events appropriately are needed to clarify the true role of IOP reduction on NTG management.

Key Summary Points

Why carry out this study?

Normal-tension glaucoma (NTG) is a subtype of primary open-angle glaucoma characterized by optic nerve damage and visual field loss despite normal intraocular pressure (IOP). Although several mechanisms have been proposed, including vascular dysregulation, increased IOP sensitivity, neurodegenerative processes, and altered cerebrospinal fluid dynamics, the underlying pathophysiology of NTG remains unclear, and no reliable animal model exists.

The Collaborative Normal-Tension Glaucoma Study (CNTGS), published in 1998, has long served as the central evidence supporting IOP reduction as an effective treatment for NTG. Given advances in statistical methodology and ongoing discussion about the interpretation of its results, a critical and structured reassessment of the CNTGS is needed to understand how its conclusions follow from the original data and analytic choices.

What was learned from the study?

This study re-evaluates the methodology and statistical analyses of the CNTGS and identifies key post hoc analytical modifications, including a shift of baseline to the point of IOP stabilization and selective censoring of cataract-related visual decline. These choices substantially influenced the estimated treatment effect.

When the CNTGS data are interpreted under the original intention-to-treat (ITT) framework, which follows all randomized eyes from randomization and includes all progression events, no statistically significant effect of IOP reduction is observed.

These findings reinforce the importance for strict adherence to ITT methodology, transparent handling of time-dependent processes, and appropriate modeling of treatment-related events.

Future clinical studies on NTG would benefit from prespecified analysis plans, unbiased time zero definitions, and appropriate competing-risk or time-dependent modeling to clarify the therapeutic role of IOP reduction.

Introduction

Glaucoma is traditionally associated with an optic nerve neuropathy thought to be caused by elevated intraocular pressure (IOP). The term normal-tension glaucoma (NTG) is therefore somewhat counterintuitive for this poorly understood optic neuropathy. NTG was likely first described by Von Graefe in 1850, when he reported a patient with glaucomatous optic disc cupping despite normal IOP [1]. Since then, a large body of literature has accumulated attempting to clarify the mechanisms underlying NTG [2–4].

NTG is currently classified as a variant of primary open-angle glaucoma (POAG). It is estimated to account for approximately 30–40% of POAG in Western populations and up to 90% in East Asian populations [5, 6]. Diagnosis is based on the glaucomatous disc appearance and corresponding visual field defects. Splinter hemorrhages appear to be more frequent in NTG than in POAG, suggesting a possible vascular contribution [7–9]. Compared to POAG, the scotomas in the visual field are located closer to fixation early in the disease, involving shorter axons that lie nearer to the optic nerve (ON) head [10]. Despite extensive research, no single unifying theory has explained glaucomatous optic disc excavation in the absence of elevated IOP, and no adequate animal model for NTG exists.

The first question in order to elucidate the pathophysiology of NTG is about the location from where the damage takes its starting point. Many authors argue that retinal ganglion cells are likely the first site of damage in NTG [11], although definitive histological or physiological evidence remains limited. Other authors suggest that some patients with NTG may have increased sensitivity to IOP, but this concept also lacks biological substantiation.

Another unresolved question is whether NTG is an isolated ocular disease or whether it is part of a more generalized pathological process. One prominent hypothesis is vascular dysregulation, which focuses on impaired blood supply to the ON head leading to hypoxic damage to the optic disc. One of the unanswered questions in this concept is why such patients never display cotton wool spots on the ON head, as cotton wool spots are the hallmark finding of ischemic damage to axons seen, for example, in anterior ischemic optic neuropathy (AION). Although there is evidence in the literature that vascular dysregulation may play a role in the pathophysiology of NTG, it does not seem to be the final explanation.

A growing body of literature in recent years suggests that glaucoma may involve neurodegenerative mechanisms comparable to those observed in other neurodegenerative diseases [12, 13]. Several studies have reported an increased risk of developing Alzheimer’s disease (AD) in patients with NTG [14]. Further, an intriguing recently published paper reports that a substantial number of patients with AD manifest primarily with symptoms related to the secondary visual pathway, such as simultanagnosia, space perception deficit, agraphia, and object perception deficits [15]. The anatomical sites responsible for these neurological symptoms are situated in the “where and what stream” region, specifically within the parietal-occipital area. Other reports describe diffuse brain damage in patients with NTG as well as impaired cerebrospinal fluid (CSF) dynamics along the ON [16, 17]. These observations raise the question of whether NTG originates primarily in the retinal ganglion cells, as traditionally assumed, or rather in the brain, progressing in an anterograde manner towards to the optic disc. In this line of thought NTG would then primarily be a neurodegenerative process of the ON and the damage might be related to an impaired cleaning function of CSF and interstitial fluid within the subarachnoid space and in the ON parenchyma.

Given the complexity and incomplete understanding of NTG pathophysiology, careful evaluation of the evidence supporting current therapeutic strategies is essential. Although IOP is within a normal range in NTG, clinical management typically focuses on lowering IOP by medical or surgical means. This approach is primarily based on the findings from the Collaborative Normal-Tension Glaucoma Study (CNTGS) published in 1998 [18, 19].

A substantial challenge to the presumed central role of IOP was raised by Lee et al. [20], who compared the progression of visual field loss in patients with NTG with very low IOP (10–15 mmHg) to a group of patients with higher IOP (15–21 mmHg). They concluded “the incidence of defined progression and progression rates using MD, PSD, and VFI did not differ between the 2 groups, even though mean IOP and the decrement of IOP form baseline were significantly higher in group B than in group A.” These results suggest that, for some patients, difference in IOP within the statistically normal range may not meaningfully alter the course of the disease.

Because CNTGS remains the principal evidence base for IOP-lowering therapy in NTG, a rigorous reassessment of its statistical methodology is warranted. Earlier critiques, including that of Sommer [21], highlighted several methodological and clinical limitations. The present study builds on these observations by systematically examining the CNTGS design, analytical choices, and interpretations, with particular emphasis on survival analysis principles and intention-to-treat (ITT) methodology, in order to determine whether the widely accepted clinical conclusions are supported by the underlying statistical evidence.

The aim of this study is to provide a structured re-evaluation of the statistical and methodological framework of CNTGS, focusing on how analytical decisions such as baseline definition, censoring rules, and deviations from ITT principles influenced the reported treatment effect and its subsequent interpretation in the NTG literature.

Methods

This study analyses the methodology and statistical decisions underlying the CNTGS to evaluate the conclusions of its research.

Standard Approach to Modeling and Analyzing Time-to-Event Data in Clinical Trials

The traditional approach for analyzing the effect of an intervention on the time until a specified outcome is survival analysis. This methodology is widely used to estimate and compare the time-to-event distributions across treatment groups. Typically, survival functions are estimated using the Kaplan–Meier (KM) method, which provides a non-parametric estimate of the probability of survival over time, accommodating censored observations. To statistically assess differences between two KM survival curves, the log-rank test is commonly employed, as it tests the null hypothesis that there is no difference in survival between the groups over time [22].

When comparing treatment groups, it is often essential to adjust for patient-related factors that may influence the time to the outcome of interest. A widely used multivariate model for this purpose is the Cox proportional hazards (PH) model, which estimates the hazard ratio between groups while adjusting for potential confounding variables or prognostic factors [23]. This approach assumes that the ratio of hazard rates between groups is constant over time, providing a flexible model for adjusting for multiple covariates without needing to specify the baseline hazard.

Applying This Framework to the CNTGS

In the context of evaluating the effect of IOP reduction on disease progression in NTG, a carefully designed survival analysis should be structured as follows:

1. Defining treatment and control groups:

   The treatment group would consist of patients with NTG undergoing a surgical intervention to reduce IOP by a specified percentage (e.g., 30%) through surgical or medical means. The control group, in contrast, would consist of patients with NTG who do not receive any IOP-lowering intervention.

2. Outcome measure specification:

   A precise and clinically relevant outcome measure must be established to consistently assess disease progression. In this setting, the primary outcome could be defined as visual field deterioration, measured by changes in the Humphrey Visual Field Analyzer or equivalent. Clear criteria should determine the threshold for significant visual field reduction, so the occurrence of progression events can be reliably recorded across both groups.

3. Identifying prognostic factors and adjusting for confounding:

   On the basis of clinical knowledge of NTG, key prognostic factors—such as baseline IOP, optic nerve health, age, and the presence of systemic conditions like hypotension—should be included in the Cox PH model. These variables, known to influence NTG progression, can be included as covariates to adjust the hazard ratio for IOP reduction’s effect on visual field preservation. For instance, systemic hypotension has been associated with an increased risk of glaucomatous progression due to reduced optic nerve perfusion [24].

This structured approach ensures that survival analysis fully captures the impact of IOP reduction on NTG progression. With these principles in place, one can systematically evaluate how the CNTGS study design and analysis align—or deviate—from standard methodology and assess potential sources of bias in the reported treatment effect.

Ethical Approval

This study is a methodological reassessment based exclusively on previously published, anonymized data from the CNTGS. No new human or animal research was conducted. According to the regulations of ETH Zurich and the University of Basel, analyses based solely on published data do not require institutional ethics committee approval.

Results

Baseline Shift and Outcome Definition

A pivotal methodological deviation in the CNTGS design was the post-randomization shift of the baseline timepoint for survival analysis in the treatment arm. In the primary 1998 report, time to progression for treated eyes was calculated not from randomization—as is standard in ITT analyses—but from the point of “IOP stabilization”, i.e., after achieving the 30% IOP reduction target. In contrast, untreated control eyes were followed from the moment of randomization. This asymmetry introduced a run-in period for the treatment group, during which 11 early progression events in treated patients were excluded from analysis because they occurred before IOP-lowering threshold was met. By resetting the survival clock post hoc, the analysis systematically favored the treatment arm: the estimated mean time to progression was inflated to 2688 days (around 7.4 years) in the treated eyes versus 1695 days for controls. In absolute terms, only 7 of 61 treated eyes (12%) reached a defined endpoint (visual field or optic disc deterioration), compared to 28 of 79 control eyes (35%) [18]. This discrepancy illustrates how redefining the “baseline” after randomization can introduce immortal time bias, artificially extend estimated survival, and obscure early failures of therapy.

Selective Censoring of Cataract-Related Progression

Another key methodological choice in the CNTGS was the selective censoring of eyes that developed cataracts during follow-up. Cataract formation was a frequent adverse effect of the intensive IOP-lowering treatment (particularly in surgically treated eyes), occurring in 23 of the treated eyes (38%) vs only 11 of control eyes (14%) [18]. Lens opacities usually cause generalized visual field depression and not focal scotomas as typically seen in glaucoma. The investigators elected to censor follow-up in any eye once a cataract induced at least a 2-line loss of visual acuity [19]. Importantly, this meant that any glaucomatous progression occurring after (or concurrently with) significant cataract was no longer counted as an endpoint in the primary survival analysis. By treating cataract-induced visual decline as a reason to censor rather than as an event or competing risk, the analysis disproportionately removed progression events from the treatment arm. In the companion ITT analysis, 22 of 66 treated eyes showed progression when all events were included, virtually the same rate as controls (31 of 79). However, with cataract-related losses censored, only 8 of 66 treated eyes were counted as progressing, versus 21 of 79 controls [19]. This censoring strategy, while addressing one confounder, omitted many treatment-arm failures from consideration, yielding an overly optimistic view of treatment efficacy. Statistically, cataract development would more appropriately be handled as a competing risk or as a time-dependent covariate rather than through simple right-censoring. Standard right-censoring assumes that the censored cases are random and unrelated to outcome—an assumption clearly violated here. By censoring these informative events, the CNTGS skewed its survival analysis in favor of treatment, ultimately overstating its efficacy. A summary of the key reported CNTGS outcome measures under the different analytic frameworks is provided in Table 1, illustrating how baseline shifting and cataract censoring altered the number of progression events counted in each study arm.

Table 1.

Summary of outcome measures reported in the Collaborative Normal-Tension Glaucoma Study (CNTGS) (1998a/b) under the intention-to-treat (ITT) and post hoc (per-protocol) analytic frameworks

Measure	Control group	Treatment group	Notes
Randomized eyes (ITT population)	79	66
Progression events under ITT	31/79 (39%)	22/66 (33%)	All events included, time zero = randomization
Eyes progressing before IOP reduction target was reached	N/A	11	Excluded from per-protocol analysis after baseline shift
Number of eyes used in per-protocol survival curves	79	61	Treatment arm excludes early progressors and eyes not achieving 30% reduction
Progression events after baseline shift (per-protocol)	28/79 (35%)	7/61 (12%)	Time zero = IOP stabilization in treatment arm
Eyes developing cataract during follow-up	11 (14%)	23 (38%)	Treatment substantially more affected
Progression events counted after cataract censoring	21/79 (27%)	8/66 (12%)	Cataract-related visual decline censored from analysis

All values are taken from published CNTGS reports. The table is intended to clarify how changes in baseline definition and censoring rules affected the number of progression events counted in each arm

Post Hoc Statistical Testing and the Emergence of “Significant” Treatment Effects

The combination of baseline shifting and cataract censoring created conditions in which standard survival analysis favored the treatment group. In the first paper’s per-protocol analysis, the 5-year progression-free survival was approximately 60% in untreated patients versus approximately 80% in treated patients, yielding a visually substantial separation of the KM curves [19]. A log-rank test (and corresponding Cox model likelihood-ratio test) on these adjusted KM curves found this difference statistically significant (reported p values < 0.001) [18]. However, these significant results emerged only after implementing the post hoc analytical adjustments. The originally planned ITT comparison—analyzing groups as randomized from time zero and counting all progression events—showed no significant benefit of treatment: progression rates were “indistinguishable” between groups (approximately one-third of patients in both arms over 5–7 years, p = 0.21) [19]. Only after redefining the analytical baseline to IOP stabilization and censoring cataract-associated losses did a treatment effect emerge. This sequence of analyses demonstrates that the statistically significant treatment effect reported in CNTGS depended on retrospective modifications to the analytical framework that are difficult to justify within standard principles of survival analysis. Even at the time the study was conducted, the ITT approach was well established as the primary basis for estimating treatment effects in randomized trials, and the problems associated with post-randomization exclusions, altered time origins, and outcome-dependent censoring were already recognized in the statistical literature. By shifting the baseline after randomization and censoring a treatment-related event such as cataract formation, the analysis departed from these principles and created conditions that favored the treatment arm. These deviations do not imply inappropriate intent but they complicate the interpretation of treatment efficacy and highlight the need for prespecified analytical plans and unbiased starting points in future NTG trials.

Discussion

The CNTGS remains a landmark trial in the clinical management of NTG, but its findings are more nuanced than often acknowledged. Although its post hoc analyses suggested a benefit of IOP lowering, the unadjusted ITT analysis—the most reliable method in randomized trials—failed to show a significant effect.

Interpretation, Bias, and Cautious Conclusions

Although the investigators acknowledged the conditional nature of their findings, the subtleties of the analysis can be difficult to appreciate without close examination of the methodological choices. The CNTGS authors concluded that “intraocular pressure is part of the pathogenic process in NTG” and that “IOP-lowering therapy ‘free of adverse effects’ would be expected to be beneficial” [18]. This wording implicitly recognizes that the observed treatment effect depended on analytical adjustments that excluded treatment-related adverse events, particularly cataract formation, which materially influenced progression assessment. The companion paper explicitly noted the discrepancy between the ITT and post hoc analyses, attributing this to the higher incidence of cataracts in the treatment arm [19]. The authors further emphasized that nearly half of untreated patients did not progress at all over 5–7 years of follow-up, and that indiscriminate treatment could subject patients to risks without proven benefit [25]. In this light, CNTGS is best understood as providing a conditional signal of benefit under an analytically modified framework, rather than conclusive evidence of universal therapeutic efficacy. The trial did not demonstrate a net benefit under the conditions of randomization and complete event capture, which is the interpretive standard in modern clinical trial methodology.

Propagation of Ambiguity into Clinical Interpretation

Despite the authors’ careful framing, aspects of the trial’s analytic complexity have been simplified in subsequent literature. Over time, the CNTGS findings have increasingly been cited as if affirmative evidence that lowering IOP prevents progression in NTG, whereas the original evidence base is more conditional. For example, Kim et al. introduced their 12-year follow-up study by describing the CNTGS as “the most compelling evidence for the effect of lowering IOP in NTG,” noting that a 30% IOP reduction “halved the incidence of progression at 3 years” [26], emphasizing the per-protocol results without equal attention to the null ITT findings.

Similarly, a 2014 investigation by Lee et al. opened with the statement that “it is generally accepted that a significant reduction in IOP suppresses the progression of visual field damage in NTG patients,” referencing CNTGS among the supporting trials [20].

Such citation patterns illustrate how selective emphasis over time can influence clinical interpretation. The phenomenon is not unique to CNTGS and is well described in the medical literature as citation drift. Here, a trial demonstrating potential benefit under specific analytic modifications has, in some contexts, come to be cited as if it had shown clear-cut benefit under ITT conditions. This underscores the importance of evaluating analytic decisions alongside outcome estimates and of conveying methodological caveats transparently when translating trial findings into clinical practice.

Implications for NTG Management

The findings of this reassessment do not argue against IOP-lowering therapy outright. Rather, they highlight the uncertainty inherent in the evidence supporting routine IOP reduction for all patients with NTG. Given that a substantial proportion of untreated patients in CNTGS did not progress over several years, and that the ITT comparison did not demonstrate a statistically significant treatment effect, management decisions should continue to be individualized, incorporating factors such as disease stage, documented progression, patient-specific risk profiles, and quality-of-life considerations.

Limitations

This study is based entirely on published data from CNTGS, without access to individual-level time-to-event information. As a result, no re-estimation of survival curves, hazard ratios, or alternative models was performed. All numerical comparisons rely on data reported in the original publications. Nevertheless, the methodological critique presented here is grounded in established statistical principles and does not depend on reanalysis of raw data.

Conclusions

Given that the pathophysiology of NTG remains unresolved—potentially involving systemic and local vascular dysregulation, neurodegenerative processes, or both—future studies must avoid analytical shortcuts and adopt transparent, ITT-based designs. Until such rigorously constructed evidence is available, clinical guidelines should reflect the existing ambiguity of the effect of IOP reduction and advocate for individualized, risk-based treatment decisions rather than universal surgical IOP-lowering mandates, often at the cost of quality of life of the patients.

In addition, the methodological issues highlighted in this reassessment point to several priorities for future NTG trials. These include prespecified analysis plans, consistent use of randomization as time zero for both treatment and control groups, and appropriate handling of treatment-related events such as cataract formation using competing-risk or time-dependent approaches rather than censoring. Trials designed along these lines would allow a more reliable estimation of the true therapeutic value of IOP reduction in NTG and help clarify which patients, if any, are most likely to benefit from intervention.

Author Contributions

Hanspeter E. Killer, Achmed Pircher, and Daniel J. Stekhoven conceived the study. Daniel J. Stekhoven performed the statistical assessment and methodological analysis. All authors contributed to drafting and revising the manuscript, approved the final version, and agree to be accountable for all aspects of the work.

Funding

No funding or sponsorship was received for this study or for the publication of this article. The Rapid Service Fee was funded by the authors.

Data Availability

Data sharing is not applicable to this article because no new datasets were generated or analysed. All information used in this reassessment is derived from previously published CNTGS reports cited within the manuscript.

Declarations

Conflict of Interest

Hanspeter E. Killer, Achmed Pircher, and Daniel J. Stekhoven have nothing to disclose.

Ethical Approval

This study is a methodological reassessment based exclusively on previously published, anonymized data from the CNTGS. No new human or animal research was conducted. According to the regulations of ETH Zurich and the University of Basel, analyses based solely on published data do not require institutional ethics committee approval.