Aims To investigate the day-to-day repeatability of intraocular pressure (IOP) measurements.
Methods A prospective cohort study of untreated patients presenting with primary open-angle glaucoma or ocular hypertension presenting with IOP>21 mm Hg. IOP was measured by masked Goldmann tonometry at 08:00, 11:00 and 16:00 at each of the three weekly visits. After starting travaprost (0.004%) to both eyes, the measurements were repeated for a further three weekly visits. Day-to-day repeatability was estimated before and after commencing medication and reported as the coefficient of repeatability and coefficient of variability.
Results At the 8:00 time point, mean IOPs were 26.1 and 17.9 mm Hg in the eye with higher pressure before and after starting treatment, respectively. Coefficient of repeatability and coefficient of variability were 6.8 mm Hg and 10.0%, respectively, before treatment, and 4.6 mm Hg and 10.5% on treatment. Therefore, before treatment and after starting medication the IOP lay within a range of ±20% of the mean IOP with 95% confidence.
Conclusions The non-therapeutic variability from day to day significantly undermines the precision of IOP estimation and of the estimation of medication effectiveness even when the time of day is standardised in patients with primary open-angle glaucoma/ocular hypertension.
- intraocular pressure
- measurement error
- intraobserver variability
- diurnal rhythm
- treatment medical
- clinical trial
- treatment surgery
Statistics from Altmetric.com
- intraocular pressure
- measurement error
- intraobserver variability
- diurnal rhythm
- treatment medical
- clinical trial
- treatment surgery
It is well known that intraocular pressure (IOP) varies spontaneously over a 24 h cycle.1 Until recently, it was assumed that diurnal curves in an individual are constant so that at any given time IOP is repeatable from day to day.2–4 This assumption underlies the way in which the clinical effectiveness of glaucoma medications is traditionally assessed. Typically, a patient is started on treatment and the IOP rechecked at a subsequent visit at the same time of day. In this time-of-day standardised approach, any change in IOP is then attributed to the medication.
However, there is little in the literature to support this assumption, and where limits of day-to-day repeatability have been quantified, this has only been in healthy eyes.5–8 A clearer understanding of the non-therapeutic variability of IOP in both untreated and treated patients with glaucoma and ocular hypertension (OHT) is needed to determine how best to accurately measure the effectiveness of glaucoma medication.
In this study, we aimed to quantify the day-to-day repeatability of Goldmann tonometry at different times of the day in patients with primary open-angle glaucoma or OHT; first, when they were treatment-naïve, and subsequently after starting IOP-lowering medication. In this manner, we examined the justification for determining glaucoma medication effectiveness using the traditional time-of-day standardised approach.
Details of the enrolment, inclusion criteria and measurement methodology have been published in detail previously.9 In brief, patients with primary open-angle glaucoma (POAG), or OHT with an IOP >21 mm Hg in both eyes at their initial clinical visit who were considered to need bilateral antihypertensive medication were invited to participate. Patients underwent a complete ophthalmological examination, including slit-lamp biomicroscopy, gonioscopy, dilated funduscopy and threshold visual field analysis. All patients were >40 years of age, naïve to treatment, had open drainage angles and no history of ocular surgery or corneal pathology.
Goldmann applanation (Haag-Streit (Bern, Switz)) was used to measure IOP using a single drop of proxymetacaine (0.5%) fluorescein (0.25%) (Bausch and Lomb, UK) for anaesthesia. Each measurement was taken by two investigators—an examiner who equilibrated the mires, but was masked to the IOP value, and a second examiner who recorded the result. Three measurements were taken, and the median of these taken as representative. Between each measurement, the tonometer gauge was reset to 10 mm Hg. The right eye was always measured first. The examiner undertaking pressure measurements was masked to the IOP on previous visits. The same tonometer was used throughout the study, and accurate calibration was confirmed daily.
Following the recruitment visit (V0), each participant had seven further visits at weekly intervals. At each of these visits, IOP was measured at 8:00, 11:00, and 16:00 (±15 min). Diurnal IOP was defined as the mean of IOP at the three time points. The visit regime is shown in table 1. For the first three post-recruitment visits (V1–3) both eyes were untreated. After visit 3, travaprost (0.004%) (Travatan®, Alcon, Fort Worth, Texas, USA) was commenced in the eye with higher IOP. At the fourth visit (V4) medication was only being used in one eye but was commenced in the second eye following that visit. At visits 5, 6 and 7 the patient was using travaprost in both eyes.
For the purposes of this study, we analysed the data from visits 1–3 (neither eye treated) for both first and second eyes. We also analysed data after 1, 2 and 3 weeks of treatment (visits 4–6 for first treated eyes; visits 5–7 for second treated eyes). Intraclass correlation coefficients (ICC) were calculated from repeated measures analysis of variance (ANOVA) to assess the short-term variability in IOP at each of the 3 fixed time points (8:00, 11:00, 16:00). An ICC <0.4 was considered poor agreement; 0.4–0.75 fair to good agreement; >0.75 excellent agreement.10
Repeatability at the same time points on different days was estimated using ANOVA to estimate within-subject (test-retest) SD (Sw) from the square root of the residual mean square.11–13 Two measurements would be expected to lie within 2.77Sw (the coefficient of repeatability (CR) ie, 1.96×√2Sw) of each other with 95% confidence.
Within-subject coefficients of variation (CV) for IOP measurements at each time point were derived from Sw using the log transformation method.12 Measurements would be expected to lie within 1.96 CV of the mean IOP with 95% confidence.
CIs were derived from the SE of Sw (ie, se= Sw /√(2n(m-1)) for n subjects, and m observations per subject). Significance tests were performed by ANOVA except where stated.
Analysis was performed using SPSS V.15.0 (SPSS).
The study was reviewed and approved by the Nottingham Ethics Committee, and was conducted in accordance with the Declaration of Helsinki. All participants provided written informed consent.
Of the 32 patients enrolled, 30 completed all the required visits (19 women; 28 Caucasian; mean age 64.4 years (range 42–88); 16 OHT, 14 POAG). Mean pre- and post-treatment IOPs for each time point are given in table 2. IOPs for eyes with POAG compared with OHT were not significantly different. There was no further reduction in post-treatment IOP after the first week on travaprost at any time point, indicating that maximum effect was achieved within a week.
ICC results indicated nominally good to excellent agreement10 between measurements at three successive visits in the first treated eyes, and fair to good agreement in the second treated eyes (table 2).
Day-to-day repeatability and variability coefficients
Repeatability was quantified using CR and CV for IOP values at three pre- and three post-treatment visits. The results are shown in table 3 for the eyes treated first. A CR value for 8:00 of 6.8 mm Hg indicates that 95% of repeated measurements on the same subject at the same time of the day would be expected to lie within this range. Any difference <6.8 mm Hg in IOP between visits would be indistinguishable from non-therapeutic variation and measurement error. CR ranged between 5.3 and 6.9 mm Hg pre-treatment, and between 4.1 and 4.8 mm Hg on treatment.
Repeatability (Sw) was independent of the time of day before (p=0.25) and after (p=0.62) starting treatment, and the mean IOP before (p=0.53) and after (p=0.55) starting treatment (linear regression results for 8:00). It was also independent of the duration of treatment. The repeatability between 1 and 2 weeks on treatment (V4,5) was no different from that between 3 and 4 weeks on treatment (V6,7). CR for V4,5 at 8:00 was 4.3 mm Hg compared with 4.5 mm Hg for V6,7 (at 8:00, p=0.69) (results at other time points were similar).
Repeatability was also independent of the diagnosis; CR for OHT at 8:00 was 6.7 mm Hg compared with 6.9 mm Hg for POAG (at 8:00, p=0.60).
From the Sw, we deduce that 95% of measurements would be expected to lie within ±4.9 mm Hg (ie, 1.96Sw) or ±20.0% (ie, 1.96 CV) of the mean IOP value at 8:00. At the three time points, this precision ranged between ±3.8 and ±4.9 mm Hg (±14.9% to ±20.5%) pre-treatment, and between ±2.9 and ±3.4 mm Hg (±21.2% to ±23.1%) on treatment.
There were significant reductions in CR after starting treatment, more pronounced for the first treated eyes (p<0.001 at all three time points for first treated eyes), than second treated eyes (at 8:00 p=0.005; at 11:00,p=0.11; at 16:00, p<0.001). However, the coefficient of variability was the same, or slightly greater, after treatment in both first and second eyes. As CV is a proportion of the underlying mean IOP, this is unsurprising, and indicates that precision as a percentage decreased after starting treatment, as the mean IOP was reduced (tables 3 and 4).
In order to determine whether measuring IOP at a standardised time each day improves repeatability, we used the 8:00 IOP for V1, 11:00 for V2 and 16:00 for V3 as a non-time standardised comparison for the pre-treatment group. The results for this analysis and the equivalent after starting treatment are given in table 5. So, for example, CR and CV were 8.3 mm Hg and 12.7%, respectively, compared with 6.8 mm Hg and 10.0% using a fixed 8:00 time point. It can be seen that time-standardised measurements consistently improved repeatability.
Repeatability of diurnal measurements
Using the mean of the three IOP readings (8:00, 11:00, 16:00) at each visit rather than just using a single time point, precision improved from an average of ±18.4% to ±13.3%, and from 20.9% to ±16.1% pre- and post-treatment, respectively.
Our results indicate that the correlation between IOP measurements repeated at the same time of day on three occasions over a 3-week period was nominally ‘good to excellent’10 in patients with untreated POAG/OHT both before and after commencing treatment with travaprost.
However, on closer inspection using quantitative assessment of the degree of variability, agreement appears less satisfactory. CR (within which 95% of repeated values are expected to lie) ranged between 5.3 and 6.9 mm Hg at the same time on different days in untreated patients, and between 4.1 and 4.8 mm Hg after starting treatment. Any difference in day-to-day IOP readings less than these values would be indistinguishable from day-to-day fluctuation and measurement error.
Repeated IOP measurements at the same time on different days lie within a range of ±14.9% to ±20.5% of the mean value with 95% confidence in untreated patients, and between ±21.2% and ±23.1% after starting treatment.
A level of uncertainty in measurement of IOP of the order of 20% is certainly clinically important and needs bearing in mind when interpreting the results of single IOP readings taken in isolation. It also contributes a significant degree of imprecision in assessing the effectiveness of glaucoma treatment, which in this context was around 30%.
There is little data on the day-to-day repeatability of IOP and we are not aware of any studies quantifying between-session repeatability in a controlled, prospective manner in patients with glaucoma/OHT either with or without treatment. Previous reports in healthy subjects have yielded CR in the range 2.3–4.8 mm Hg.5–8 The observation that our values are higher may in part be due to methodological differences, but more likely it is due to variation in the populations studied. All our subjects had POAG with elevated IOP or OHT. However, raised IOP was not the factor responsible, as variance for between-visits pressure was shown to be independent of the mean IOP. This suggests that day-to-day variation in patients with high-tension glaucoma/OHT may be higher than in normal eyes regardless of the level of IOP, and extrapolation from normal subjects inappropriate. Furthermore, the reduction in CR after starting travaprost suggests that treatment may decrease intrinsic IOP fluctuations in patients with POAG/OHT.
Realini et al examined short-term repeatability in patients with treated glaucoma.2 The results were presented in the form of ICCs. The ICC between two visits ranged between 0.45 and 0.71; conventionally interpreted as ‘fair to good’ correlation.10 However, this descriptive interpretation depends on clinical context and can be misleading. In our data, for example, an ICC of 0.85 indicating ‘excellent’ agreement is associated with an uncertainty of ±20%. A range of uncertainty between 16 and 24 mm Hg, for example, would not be considered excellent precision. Furthermore, there are theoretical reasons why use of ICC in this type of study can be misleading. ICC=Sb2/(Sb2+Sw2) where Sb2 is the between-subject SD from the ANOVA. Therefore, unlike other measures of agreement, for example, CV and CR which are only dependent on Sw, ICC depends on the variation between subjects, and so relates only to that sample population. If we simply choose a sample with a wide range of IOP, then the ICC will be higher than for a narrow sample irrespective of the level of agreement. In this study, first treated eyes had a broader spread of IOP than fellow eyes (see SD values in table 2), which accounts for the higher level of ICC for first treated eyes seen in tables 3 and 4, while the more robust estimates of variability (CV and CR) were the same for both eyes.
ICC values that are ‘good to excellent’ in this data disguise a limit of precision in excess of ±20%, which represents a clinically significant level of uncertainty when comparing IOP measurements from visit to visit even at the same time of day.
The strength of this study lies in its prospective nature and a protocol designed to minimise sources of measurement bias. All readings were taken in a standardised, masked manner to avoid inadvertently biasing the results towards falsely high levels of association. In addition, the IOP at the recruitment visit (V0) was discarded for the analysis to avoid the problem of mean regression, which would otherwise tend to exaggerate the variability.14 It is likely that the level of variability in IOP we report in this study would underestimate that which would be seen under typical clinic conditions with multiple observers and instruments in use, and without a stringent research methodology.
This study questions the assumption of repeatability of IOP at any given time of day that underlies the assessment of medication effect in clinical practice. When a patient is started on treatment, and the IOP checked at the same time on a later day, the assumption is that any observed drop in IOP is attributable to the treatment. However, a variability of ±20% is sufficiently large to mislead the clinician into thinking a significant response has been achieved when in fact the treatment has had no effect, or vice versa. Standardising the time of day for repeat visits is shown in this study to improve repeatability, particularly in untreated eyes, as without this, variability was even larger.
One option is multiple pre- and post-treatment measurements on different visits.15 This improves estimation of medication effect,9 but is resource intensive and inconvenient. Furthermore, no attempt has been made to quantify the number of visits required. A second approach is to use diurnal pressure measurements. By averaging three IOP readings over office hours at each visit, we showed significantly improved repeatability, but this again increases inconvenience.
One alternative is the monocular treatment trial in which medication is begun in only one eye and then the relative change of IOP in the two eyes is compared at follow-up16 ,17 This method has not achieved widespread acceptance, but it gave a significantly more accurate assessment of medication effectiveness in this patient group without any additional office visits.9 ,16
In this study, IOP was only measured up to 4 weeks after starting treatment. However, maximum therapeutic effect was seen within the first week as reported previously,18–20 and comparison of treatment in the first fortnight with that of the second showed no evidence of improved repeatability with time.
We have only looked at patients presenting with bilateral OHT or POAG with elevated IOP. While the results were consistent between these two groups, care should be exercised in wider extrapolation, for example, to patients with normal tension glaucoma, or patients already on treatment.
Within these constraints, it is clear that even under ideal conditions, day-to-day variability significantly undermines the precision of IOP measurement and of the estimation of medication effectiveness even when the time of day is standardised. Better ways of eliminating the effects of this variability than are currently available are needed.
This study was supported by an unrestricted grant from Alcon UK. Alcon personnel were not involved in the design or conduct of this research.
Funding AJK has received support from Alcon to travel to scientific meetings and is a member of an Alcon advisory board.
Competing interests None.
Patient consent Obtained.
Ethics approval Ethics approval was provided by Nottingham Ethics Committee.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.