Haigis a0, a1, a2 — What the Three Constants Actually Do

The Haigis formula was one of the earlier widely adopted IOL power formulas to explicitly model effective lens position using multiple biometric inputs and separate constants. Understanding what each constant does — and what it cannot do — is essential for surgeons who use Haigis or interpret its outputs, and for understanding why constant optimization matters in any formula system.

The problem with global calibration

Most IOL power formulas use a single constant that primarily acts as a global calibration term, adjusting predictions to match a surgeon's average outcomes. This approach corrects mean prediction error across a population but cannot correct systematic variation in error that occurs across different biometric subgroups. A formula calibrated to perform well on average may still show consistent bias in short or long eyes if the relationship between measured biometry and actual postoperative lens position varies in ways the single constant cannot capture.

Haigis et al. (2000, Graefes Archive) described a three-constant model in which preoperative ACD and axial length each contribute independently to the ELP prediction, weighted by separate regression terms. This allows the formula to be personalized not just for average performance but for performance across the biometric range.

What each constant does

a0 is an offset term. It shifts the entire ELP prediction up or down by a fixed amount, independent of the patient's biometry. Its function is similar to adjusting the A-constant in SRK/T — it corrects the mean prediction error without changing the relationship between biometry and predicted ELP. Because it does not depend on biometric variation to be estimated, a0 is the only constant that can be meaningfully optimized from a small dataset.

a1 and a2 are regression terms that adjust how measured ACD and AL, respectively, contribute to the predicted ELP within the model. They are modeling parameters — not direct representations of physiological mechanisms. In practice, these terms allow the formula to account for systematic patterns in how biometry relates to actual lens position in a specific surgeon's hands with a specific lens platform — patterns that a global offset alone cannot correct.

Population defaults and their limitations

When surgeon-specific optimization has not been performed, commonly used default starting values for a1 and a2 are 0.268 and 0.238 respectively, with a0 derived from available lens data. These defaults represent average relationships between biometry and ELP across a population and perform acceptably for most eyes in the normal biometric range.

The limitation of population defaults is that they cannot account for systematic biases in individual surgical technique, specific lens platforms, or biometric measurement systems. A surgeon whose outcomes show a consistent pattern across a biometric subgroup cannot correct this with a0 alone if the underlying cause involves the biometry-ELP relationship encoded in a1 or a2.

What optimization requires

Optimizing a0 requires only that outcomes data be available — any dataset sufficient to estimate a mean prediction error is adequate. Optimizing a1 and a2 requires not just a larger dataset but one with sufficient variation in ACD and AL. Haigis (2004) established that full three-constant optimization typically requires a substantial dataset — often 100 or more cases and preferably more — with a reasonable spread of biometric values. Lam et al. (2010) found that full three-constant optimization provided measurable improvement over a0-only optimization, particularly at biometric extremes, though generalizability depends on the cohort and measurement platform.

Why Haigis still matters alongside newer formulas

Modern formulas including Barrett Universal II, Kane, and Olsen often show strong performance in published comparisons and in some cohorts outperform Haigis. Several practical considerations keep it relevant:

• Transparency — the three-constant model is mathematically explicit and interpretable, making the rationale for constant personalization visible in a way that more opaque modern models do not
• Established databases — ULIB and IOLCon maintain optimized Haigis constants for many lens models across multiple biometers, providing a better starting point than manufacturer defaults
• Post-refractive variant — the related Haigis-L variant adapts the framework for post-myopic refractive surgery eyes

Haigis remains a capable formula across a broad biometric range, particularly when constants are well-optimized. The related Haigis-L variant adapts the same framework for post-myopic refractive surgery eyes. Its value is greatest in practices that have invested in systematic outcomes tracking and in surgeons who want a formula whose behavior they can understand and directly influence through personalization.