TIME-OF-YEAR VARIATION IN INTRAOCULAR PRESSURE

Abstract

Purpose.

To investigate the relationship between intraocular pressure (IOP) and time of year.

Methods.

During a separate investigation, patients from 2011 to 2018 (dataset A, N=3,041) in an urban, academic facility in Chicago, IL, USA received an examination that included Goldmann applanation tonometry. Regression analyses assessed the relationship between time of year and IOP. Two additional datasets, one collected in a similar manner during 1999 to 2002 (dataset B, N=3,261) and another consisting of all first visits during 2012 to 2017 (dataset C, N=69,858), were used to confirm and further investigate trends.

Results.

For dataset A, peak mean IOP occurred in December/January (15.7 ± 3.7 / 15.7 ± 3.8 mm Hg) and lowest in September (14.5 ± 3.1 mm Hg). The analysis suggested conventional quarterly analysis (Jan to Mar, etc.) can conceal time-of-year relationships due to inadequate statistical power and timing of IOP variation. Multiple linear regression analysis, with a November-to-October reordering, detected an annual, downward IOP trend (P<0.0001). Analysis of dataset B confirmed this trend (P<0.001). Fourier analysis on datasets A and B combined supported a 12-month IOP cycle for right/left eyes (P=0.01/P=0.005) and dataset C provided stronger evidence for an annual periodicity (P<0.0001). Harmonics analysis of dataset C showed a repeating pattern where IOP trended downward around April, and then back upward around October.

Conclusions.

This analysis strongly supports a demonstrable annual, cyclical IOP pattern with a trough to peak variation of ≈1 mm Hg, which has a seasonal relationship.

Précis:

In this study conducted in Chicago, Illinois, U.S.A, intraocular pressure (IOP) level was found to have a subtle, but measurable, annual pattern. Reasonable evidence is presented for a time of year variation in IOP. Adequate numbers of subjects must be studied to detect this small variation.

Introduction

Intraocular pressure (IOP) is a well-known risk factor in the development of open-angle glaucomas,^1–6 and lowering IOP is the only proven method for slowing progression. ^3,7–10 Thus, IOP is not only used diagnostically, but also in evaluating the efficacy of medical and/or surgical treatments. The diurnal variability of IOP has been studied extensively, along with physical and socio-demographic factors in varying populations, yet identified factors only account for about 10% of IOP variability.^11–14 Therefore, much greater understanding of factors influencing IOP is needed.

Previous studies have noted a higher IOP in winter compared to summer months but reports on a potential time-of-year IOP variation have been variable, with limited inclusion of potential confounding factors.^15–24 Reasons for a possible time-of-year IOP variation are unknown, but there has been various speculation on potential mechanisms. For example, it has been postulated that longer daytime hours, which increases melatonin secretion, induce an increase in progesterone androgen secretions that correlate with lower IOP in summer months.^25–30 Seasonal fluctuations have been observed to dampen with use of glaucoma medications.^{18,22,31–34}

Lowering IOP through medical or surgical treatment options has been shown to reduce the risk of progression of glaucoma. According to the Early Manifest Glaucoma Trial (EMGT), every 1 mm Hg of IOP reduction was associated with a risk reduction of 10 to 13%, depending on the analysis.^3,9,35,36 Therefore, even seemingly small IOP differences can be influential on glaucoma-related outcomes.

During an ongoing, multiyear investigation of the long anterior zonule (LAZ) trait, we collected data on numerous demographic, lifestyle, systemic, and ocular variables that included IOP. Although not the goal of the originally planned study, evolving analyses led to more intensive study of IOP, which demonstrated a relationship between IOP and LAZ.³⁷ For that report, our analyses on the LAZ-IOP relationship did not reveal a significant relationship between IOP and the time-of-year variable, but time was structured in a quarterly manner to conform to other conventional analyses (January to March, April to June, etc.). However, due the breadth of the dataset to control for a variety of potential confounding variables and the availability of additional datasets to check for the consistency of certain observations, we further explored the time-of-year variable using novel approaches to build on an existing literature that has suggested that time of year is influential on IOP level. Through these analyses, we further demonstrate the likelihood of an annual periodicity to IOP level, show additional ways to assess it, and suggest some additional characteristics concerning the cyclical variation.

Methods

We used three different datasets for the analyses described herein. Our primary dataset (A), from which our interest in the IOP to time-of-year relationship developed, was recently described in separate reports that addressed IOP among people with the LAZ trait.^37,38 This study excluded patients that had greater than trace LAZ. Subsequent to exploration on that dataset, we used the other two datasets (B, C) to check for certain consistencies and to perform additional analyses afforded by greater numbers of subjects.

The primary dataset (A) used in this present analysis was collected from October 2011 to April 2018 as part of an ongoing investigation of ocular and general health associations with the LAZ trait. For this dataset, consecutive patients presenting to nine different practitioners for a regularly scheduled examination in an urban academic eye care facility in Chicago, IL, were included if they were ≥18 years of age, had their pupils dilated during their visit, had Goldmann tonometry measurements, and agreed to complete a written questionnaire for supplemental information. Among the consenters, subjects were excluded if IOP was not measured or it was not measured via Goldmann tonometry, they were taking IOP-lowering medication, they had comorbidity or a history of ocular surgery that likely influenced IOP including cataract surgery, their maximum LAZ grade was trace, or there was missing information. The data collection time period and length of time of participation varied among investigators, with some contributing subject data over the entire study period and others for shorter timeframes.

The second dataset (B), used for validation of findings, had also been used to study LAZ but was obtained from August 1999 to March 2002.^39,40 That earlier dataset had also been collected on regularly scheduled, consecutive patients belonging to multiple (five) practitioners at the same facility. Patients ≥18 years of age having their pupils dilated were similarly included, but a supplemental questionnaire was not administered, thus resulting in information on fewer variables. Therefore, this second dataset was very similar in nature to dataset A but collected during a much earlier time period. While this dataset was not used to perform identical analyses, it allowed for checking consistency of certain trends that could help further support observations made with the primary dataset A.

The third dataset (C) was derived via an institutional electronic medical record query to collect IOP measurements on first visits of all patients ≥18 years of age seen at the facility between January 1, 2012 and December 31, 2017. Date, patient age, gender, and IOP values with measurement method was collected. Again, this dataset was not used to perform the exact same detailed analyses done with dataset A, but simply to assess certain trends in the IOP data to determine if they were consistent with some of the observations made with dataset A.

We emphasize that datasets B and C were considered supplementary in nature, not initially being aimed to provide an identical analysis to that of the primary dataset A. They were nonetheless considered to provide some value toward the overall analysis because certain IOP patterns could be checked to determine if they were repeatable with other independent data. It was felt that this might help strengthen reliability of earlier observations. For the primary dataset A, we acquired detailed ocular/medical health histories and performed comprehensive ocular examinations that included Snellen visual acuities, pupil testing, motility and binocularity assessment, color vision screening, confrontation visual fields, pre-dilation subjective refraction, slit lamp exam, Goldmann tonometry, and dilated fundus examination that included peripheral assessment with binocular indirect ophthalmoscopy. Consistent with the study site’s educational setting, student clinicians assigned to the Primary Eye Care Service performed initial examinations, including the Goldmann tonometry measurements. For uniformity, we used student measurements for analyses, primarily because faculty investigators often did not also perform tonometry measurements. All students in this clinical setting were experienced in Goldmann tonometry measurement and had passed proficiency examinations on this technique to be certified to participate in patient care. For each subject, we only used tonometry measurements from the single day of study. Further, all were collected in a consistent fashion using slit lamp mounted Goldmann tonometers, which is the standard in this clinical setting. The time of day for IOP measurement was recorded within a few minutes after measurement using an automated time stamp feature in the electronic medical record. Statistical models explored the relationship between time of day and IOP using a variety of categorical groupings of time. Final models adjusted for IOP measurements taken in the AM vs. PM clock hours.

For analysis of the time-of-year variable, we initially calculated monthly means (January to December) and quarterly means (January to March, April to June, July to August, September to December), reviewed their graphical plots, and performed exploratory regression analyses. In addition to categorical groupings, we assessed time of year as a continuous variable. Based on these observations, we then also explored other orderings and groupings to investigate whether traditional cut points might conceal true IOP trends. Other categorical grouping that we explored included quarterly groupings such as December to February, March to May, etc., and we also explored two-month categorizations such as January to February, March to April, etc. Therefore, we attempted to model the data in various ways, traditional and otherwise, in order to better understand any cyclical pattern of variation. In addition to regression modeling, we used Fourier spectral analysis to assess IOP-level periodicity and the Fisher’s g-statistic to assess significance.

For this dataset, we collected information on presence of the LAZ trait and Krukenberg spindle formation because they are highly associated with one another and important to the original dataset collection aims. We adjusted for both variables in this current analysis because we previously showed LAZ to be significantly associated with IOP. As detailed elsewhere, the criterion we used for presence of the LAZ trait was the presence of radially-oriented, possibly pigmented fibers along the anterior capsule with anterior tips extending central to the normal anterior capsular zonular insertion zone, which is located about 1.5 mm anterior to the lens equator.^37,41,42 To ensure designation of definitive LAZ eyes, we excluded subjects who had <5 LAZ fibers, i.e., trace LAZ,⁴³ in at least one eye.

We considered a Krukenberg spindle present when there was any degree of “fine pigment dusting” of the central region of the posterior cornea when it was not possible to “count” individual pigment granules because they were too small, numerous, and coalesced. We did not consider larger, coarse pigment flecks representative of a Krukenberg spindle, even when somewhat central.³⁸

Data on other variables (Table 1) relied on information extracted from the medical record and/or the questionnaire which was completed by patients without assistance. We constructed race categorization based on medical record and questionnaire information, with final assignment to one of the following categories: (1) Black/African American, (2) Asian, (3) Hispanic, Black or White, (4) Non-Hispanic White, (5) other. Education level was determined via questionnaire, using the question: “What is your highest level of education? (1) Less than high school degree, (2) High school degree, (3) Vocational school or some college but no degree, (4) College Associate’s or Bachelor’s degree, (5) College Master’s, Professional, or Doctoral degree.”

TABLE 1.

UNADJUSTED RELATIONSHIPS WITH IOP^* (RIGHT EYES) - CATEGORICAL VARIABLES

Variable	Subject Distribution	Mean IOP (SD)	P-value
Time of Year			0.001
January / February	9.4% / 9.9%	15.7 (3.8) / 15.0 (3.2)
March / April	9.0% / 10.4%	15.1 (3.4) / 15.5 (3.6)
May /June	5.7% / 7.8%	15.4 (3.5) / 14.8 (3.0)
July / August	9.5% / 7.0%	15.0 (3.2) / 14.9 (3.0)
September / October	8.5% / 8.6%	14.5 (3.1) / 14.9 (3.2)
November / December	7.2% / 7.0%	15.3 (3.7) / 15.7 (3.7)
Time of day
AM / PM	51.6% / 48.4%	15.3 (3.4) / 15.0 (3.3)	0.02
Race
African American / Other	82.2% / 17.8%	15.2 (3.4) / 14.8 (3.2)	0.02
Gender
Female / Male	65.1% / 34.9%	15.3 (3.4) / 14.8 (3.4)	<0.001
Education
>High School / ≤High School	37.8% / 62.2%	15.0 (3.2) / 15.4 (3.6)	<0.001
Long anterior zonule trait
Yes / No	3.0% / 97.0%	16.0 (3.4) / 15.1 (3.4)	0.007†
Krukenberg spindle
Yes / No	1.9% / 98.1%	16.0 (3.7) / 15.1 (3.4)	0.07
Beta-blocker medication
Yes / No	10.4% / 89.6%	15.2 (3.3) / 15.1 (3.4)	0.75
Cholesterol medication, current (per record)
Yes / No	15.8% / 84.2%	15.4 (3.4) / 15.1 (3.4)	0.03
Cholesterol medication, ever (per survey)
Yes / No	28.1% / 71.9%	15.4 (3.6) / 15.0 (3.3)	0.006
Steroid medication
Yes / No	7.7% / 92.3%	15.1 (3.5) / 15.1 (3.3)	0.82
Body mass index (kg/m 2 )
Under/Normal vs. Over/Obese	24.5% / 75.5%	14.7 (3.0) / 15.3 (3.4)	<0.0001
Underweight (<18.5)	1.8%	14.2 (3.1)
Normal weight (18.5 – 24.9)	22.7%	14.7 (3.1)	<0.0001
Overweight (25.0 – 29.9)	29.0%	15.1 (3.5)
Obese (≥30)	46.5%	15.5 (3.4)
Smoking
Ever vs. Never	47.9% / 52.1%	15.0 (3.4) / 15.3 (3.4)	0.008
Current	30.0%	14.9 (3.4)
Past	17.8%	15.0 (3.4)	0.02
Never	52.2%	15.3 (3.4)
Alcohol
Ever vs. Never	60.7% / 39.3%	15.1 (3.4) / 15.3 (3.4)	0.11
Current	48.9%	15.1 (3.3)
Past	11.2%	14.9 (3.4)	0.25
Never	40.0%	15.3 (3.4)
Diabetes
Yes / No	21.3% / 78.7%	15.9 (3.5) / 14.9 (3.3)	<0.0001
Hypertension
Yes / No	47.2% / 52.8%	15.4 (3.5) / 14.9 (3.3)	0.001
Cancer history (any site)‡
Yes / No	4.4% / 95.6%	14.1 (3.3) / 15.2 (3.4)	<0.001

^*Abbreviations: IOP, intraocular pressure; kg/m², kilograms per meter-squared; SD, standard deviation

^†Bolded p-values significant at α=0.05 level

^‡Cancer sites, N=134 cases: breast, 41.0%, prostate, 14.2%, colon, 6.7%, lung, 6.0%, uterine/cervix, 5.2% (other cancer sites <5% each)

We used questionnaire information to categorize subjects as a “current smoker,” “former smoker,” or “never smoker,” by asking: “Have you ever been a smoker? (1) Yes, currently, (2) Previously: quit <12 months ago, (3) Previously: quit >12 months ago, (4) Never or rarely: smoked less than a total of 50 cigarettes (2 ½ packs) over my lifetime.” We considered subjects as “former smokers” only if they had quit smoking more than one year. To help improve the accuracy of categorization, we asked subjects what they had smoked (cigarettes, cigars, pipe, other), how much they had smoked, (half-pack increments of cigarettes), the age they started smoking, and when they had stopped, if applicable.

For alcohol use, we used questionnaire information to categorize subjects as a “current drinker,” “former drinker,” or “never drinker” by asking “Do you drink alcohol? (1) Yes, I do currently, (2) Previously: quit <12 months ago, (3) Previously: quit >12 months ago, (4) Never or rarely because I have not drunk alcohol more than 10 times during my life.” We considered subjects as “former drinkers” only if they had quit drinking >1 year prior to the examination, otherwise, they were considered a “current drinker.” To help improve accuracy of categorization, we asked subjects how many days per week they drank (<weekly, 1–2 day, 3–4, 5–7), how much they usually drank (1–2 drinks, 3–4, 5–7, ≥8), how many years of their life they had drunk alcohol, and, if applicable, the age they had stopped drinking.

We considered subjects as having diabetes if they were currently taking prescribed medication or if they had stopped taking it against medical advice. Likewise, we only considered subjects to have a formal hypertension diagnosis if they were currently taking prescribed medication or had stopped against medical advice. As part of the standard work-up in our clinical setting, student clinicians also measured blood pressure prior to pupillary dilation drop instillation, using either automated wrist cuffs or manual arm sphygmomanometers. Body mass index (BMI) was derived using weight and height information (kg/m²) collected via the questionnaire, and it was categorized using standardized guidelines for adults (Table 1).⁴⁴

To explore cholesterol-lowering medications and reported history of elevated cholesterol, we pursued two approaches. For one, we determined which subjects had medical record notation indicating current use of cholesterol-lowering medications, and for the other, we also asked via questionnaire about history of high cholesterol and medication use, i.e., “Have you ever been diagnosed with high cholesterol and taken medication for it?” Because subjects often did not know the names of their medication(s), we did not employ statistical control for specific cholesterol medications or classes.

To explore other potential confounding of IOP relationships, we also collected medical record information concerning concurrent use of oral beta-blockers, as well as any oral, inhaled, or topical corticosteroids. Since subjects frequently could not recall specific names of their hypertensive medications, we conducted subanalyses that excluded subjects with “unknown” hypertensive medications to check stability of results.

For refractive error, we used spherical-equivalent values based on the pre-dilation subjective refraction and excluded eyes with history of refractive surgery or coexisting condition that could influence refractive error, e.g., keratoconus.

Statistical analyses were conducted using the SAS® System, Release 9.3 for Microsoft Windows® (SAS Institute Inc., Cary, NC). In addition to calculating descriptive data on subjects, multiple linear regression was used to model independent variables against the dependent variable, IOP. Stepwise, forward, and backward regression techniques were employed for model building, and we extensively studied variables using a variety of continuous and categorical formats of interest. Assumptions were met for analyses, and variables were checked for correlation and interaction. Fourier analysis was conducted using MATLAB®, Version R2018a for Microsoft Windows® (The MathWorks, Inc., Natick, MA). For datasets A and B, only data from the single study visit was used, and for dataset C, only data from the first clinic visit during the period of interest (2012 – 2017) was used for analyses. Right eye data was used for study purposes in most analyses, with left eye data being used to check for consistency of results. We obtained Institutional Review Board approval for the investigation, we followed the tenets of the Declaration of Helsinki, and subjects provided written informed consent prior to participation.

Results

The recruitment of subjects for the primary dataset A is summarized in Figure 1. Among 4,841 potential subjects, there were 3,620 (75%) who consented to participate and 1,221 who did not consent. Consenters had a mean age of 52 ± 16 (18–95 years) and were 65% female and 84% African American, and non-consenters had a mean age of 55 ± 17 (18–94 years) and were 64% female and 78% African American. Gender was similar between the consenters and non-consenters (P=0.57), but differences relative to race and age were statistically significant (P<0.0001). Among the consenters, 579 (64% female, 85% African American, mean age=59 ± 17, 18–95 years) of 3,620 (16%) were excluded because they did not have Goldmann tonometry measurements, they were taking IOP-lowering medication, they had comorbidity or a history of ocular surgery that likely influenced IOP, their maximum LAZ grade was trace, or there was missing information. The final inclusion group was 65% female, 83% African American, with mean age of 50 ± 15 (18–94) years. Differences were not statistically significant relative to gender (P=0.70) and race (P=0.32) between these inclusion and exclusion groups, but differences in age were statistically significant (P<0.0001).

Fig 1.

Flow diagram summarizing subject participation and inclusion for the primary dataset (Dataset A).

The variables used in the initial analysis, and their unadjusted relationships with IOP, are summarized in Tables 1 and 2. Our group has previously examined most of the univariate relationships in detail via an earlier and smaller version of this dataset.³⁷ Trends for potential explanatory and confounding variables appear similar with our past observations³⁷ (Table 3). This includes slightly higher IOP with female gender, diabetes, presence of the LAZ trait, higher systolic blood pressure and higher BMI. There was a slightly lower IOP in people who had a history of smoking, education more than high school, and any form of cancer. However, whereas previously we examined the time-of-year variable according to standard quarterly categorization (January to March, April to June, July to September, October to December), which did not yield statistical significance for the time-of-year explanatory variable, we explored more novel approaches for this current analysis. When examining the monthly means in the primary dataset A (Table 1), a few things were evident: 1) variation among means was small over the course of the year, and 2) the highest IOP means occurred in December and January (15.7 ± 3.7 and 15.7 ± 3.8 mm Hg) and the lowest in September (14.5 ± 3.1 mm Hg). The low variation indicated that large samples would be necessary for adequate statistical power to prove differences confidently, and the higher means straddling December/January suggested that any IOP time-of-year trends might not be captured using standard calendar quarters.

TABLE 2.

UNADJUSTED CORRELATIONS WITH IOP^* (RIGHT EYES) - CONTINUOUS VARIABLES

Variable	Correlation (r) with IOP	P-value
Age (years)	0.05	0.01
Refractive Error (SE, diopters)	−0.07	<0.001 †
Body mass index (kg/m 2 )	0.13	<0.0001
Systolic blood pressure (mm Hg)	0.17	<0.0001
Diastolic blood pressure (mm Hg)	0.13	<0.0001
Pack years smoking	−0.03	0.08

^*Abbreviations: IOP, intraocular pressure; kg/m², kilograms per meter-squared; mm Hg, millimeters of mercury; SE, spherical equivalent

^†Bolded p-values significant at α=0.05 level

TABLE 3.

MULTIVARIATE ANALYSIS OF IOP^* AS A FUNCTION OF TIME OF YEAR ADJUSTING FOR OTHER VARIABLES

Variable	Coefficient ± SE (P-value)‡
	Right Eyes	Left Eyes
Intercept	10.5 ± 0.54	10.5 ± 0.54
Time of year (order = Nov, Dec, Jan, etc.)	−0.06 ± 0.02 (0.0006)	−0.06 ± 0.02 (0.002)
Gender (female)	0.46 ± 0.13 (0.0004)	0.49 ± 0.13 (0.0001
†Refractive error (per diopter)	−0.10 ± 0.02 (<0.0001)	−0.08 ± 0.02 (0.0003)
Long anterior zonule (LAZ) trait present	0.77 ± 0.36 (0.03)	0.95 ± 0.34 (0.005)
Education >high school	−0.46 ± 0.12 (0.0002)	−0.43 ± 0.13 (0.0006)
Ever smoke	−0.33 ± 0.12 (0.008)	−0.26 ± 0.12 (0.03)
History of cancer (any site)	−0.98 ± 0.29 (0.0009)	−1.10 ± 0.30 (0.0002)
Systolic blood pressure (per 10 mm Hg)	0.29 ± 0.03 (<0.0001)	0.29 ± 0.03 (<0.0001)
Body mass index (per 10 units, kg/m 2 )	0.28 ± 0.08 (<0.0007)	0.22 ± 0.08 (0.006)
Diabetes	0.74 ± 0.15 (<0.0001)	0.69 ± 0.15 (<0.0001)

^*Abbreviations: kg/m², kilograms per meter-squared; mm Hg, millimeters of mercury; SE, standard error

^†Spherical equivalent value

^‡Variables included on if significant at P<0.05 level

To next explore IOP relative to time of year, we examined scatterplots of the monthly means, first with IOP in traditional calendar order, i.e., January to December, and then in alternative orders. Here, with the understanding that any relationship may not be exactly linear, the least squares regression line exhibited a downward trend with greatest slope with monthly reordering of November to October (P<0.0001) (Figure 2). To determine if that trend held year-over-year, we then graphed each of the six 12-month periods in a similar manner. This showed a downward trend for each of the 12-month periods, but only the first two 12-month periods reached statistical significance (P<0.05), apparently due to the lower number of subjects contributing to each year (see Figure S1 in Supplemental Materials).

Fig 2.

Plots using the primary dataset A. (Upper left): Monthly mean plots using a November-to-October ordering, showing a downward IOP trend. (Lower left): Least squares regression line and 95% confidence interval lines calculated using the individual data points rather than monthly means. (Right): Periodogram to test for cyclical IOP patterns, truncating time period from January 2012 to December 2017. A peak is present near the 0.08 frequency, which would be expected with a yearly cycle (6 years, cycles / 72 months = 0.083). Although the dominant peak suggests a 12-month IOP cycle, statistical significance is not reached with this dataset (Fisher’s g-statistic, P=0.21).

After examining the raw scatterplots, we then determined whether the November-to-October linear trend remained statistically significant while adjusting for potential confounders (Table 3). To do this, we explored a variety of regression models, and then using the same basic model as reported previously,³⁷ we found that time of year, modeled as a continuous variable (November to October), appeared independently associated with IOP for both right (P=0.0001) and left eyes (P=0.0003).

To further explore this observation of periodicity, we truncated the data to include January 2012 through December 2017 and then used Fourier analysis to examine the sequential monthly means over this 6-year period. As shown in Figure 2, a dominant peak was present at frequency of about 0.08, which corresponds to a cyclical IOP pattern of about 1 year (6 years, cycles / 72 months = 0.083). This frequency component of 1 cycle/year, however, did not reach statistical significance (P=0.21, Fisher’s g-statistic) due to limitations in the monthly sample size.

To help validate our observations on the first dataset, we used the second independent dataset B, which our group had previously collected at the same institution from August 1999 to March 2002. As shown in Figure 3, the scatterplot of IOP against time of year again shows a significant downward trend (P=0.0003) with monthly data ordered November to October. With the data truncated into two 12-month periods ordered November to October, each subset regression line again trended downward (P=0.003, P=0.045) (see Figure S2 in Supplemental Materials) similar to dataset A. Likewise, Fourier analysis of dataset B showed a dominant peak consistent with 1 cycle/year (frequency ≈ 0.08) (Figure 3), but again statistical significance was not reached with the available sample size (P=0.27). Therefore, to increase subject numbers, we then combined datasets A and B and repeated the Fourier analysis, which then showed statistically significant dominant peaks at a frequency that supported a yearly IOP cycle (Figure 4; right/left eyes, P=0.01/P=0.005).

Fig 3.

Plots using the older dataset B. (Upper left): Monthly mean plots using a November-to-October ordering, showing a downward IOP trend. (Lower left): Least squares regression line and 95% confidence interval lines calculated using the individual data points rather than monthly means. (Right): Periodogram to test for cyclical IOP patterns, truncating time period from January 2000 to December 2001. A peak is present near the 0.08 frequency, which would be expected with a yearly cycle (6 years, cycles / 72 months = 0.083). Although the dominant peak suggests a 12-month IOP cycle, statistical significance is not reached with this dataset (Fisher’s g-statistic, P=0.27).

Fig 4.

Plots using datasets A and B combined. (Top): Scatterplots of right and left eye monthly means using truncated data from dataset B (January 2000 thru December 2001) and dataset A (January 2012 thru December 2017). (Bottom): Corresponding periodograms. Yearly cycles are suggested for both eyes with statistically significant peaks present near the 0.08 frequencies (arrows, P=0.01/0.005).

Finally, for better characterization of IOP pattern, we extended our analysis to the third independent dataset C, consisting of first-visit Goldmann IOP measurements from all patients (N=69,858) seen at the institution from January 1, 2012 to December 31, 2017 (72 months). Therefore, each IOP data point represented that from a single subject. Consistent with datasets A and B, scatterplot again showed downward slope in IOP level for the November-to-October ordering (P<0.0001), and Fourier analysis showed a very dominant peak (P<0.0001) that supported a yearly repeating IOP cycle (Figure 5). With the data truncated into six 12-month periods ordered November to October, each subset regression line again trended downward, with all but the third and fourth 12-month periods reaching statistical significance (P<0.05, see Figure S3 in Supplemental Materials). To verify that dataset C was not influenced by patients with glaucoma who were taking IOP lowering medications, patients with a glaucoma-related diagnoses were removed and the analyses were repeated. Again, the Fourier analysis showed a very dominant peak (P=0.001) that supported a yearly repeating IOP cycle.

Fig 5.

Plots using the dataset C consisting of institutional patient first-visits. (Left): Least squares regression line and 95% confidence interval lines calculated using the individual data points rather than monthly means. A downward IOP trend is observed, consistent with datasets A and B. (Upper right): Monthly means scatterplot for the 72-month time period. (Lower right): Corresponding periodogram shows a dominant peak near a frequency of 0.08 with strong statistical significance (P<0.0001).

Following these analyses, we conducted a harmonics analysis of the monthly means of dataset C after collapsing the monthly means into one year to get a single mean value for each month. This 12-month data was then expanded into a Fourier series after keeping the first harmonic (at 1 cycle per year) or the first two harmonics. These approximations show first-order and second-order harmonics that appear to fit the mean values with greater precision than a simple linear trend (Figure 6). Again, similar patterns were found without the inclusion of subjects using any glaucoma lowering medications.

Fig 6.

Harmonics analysis using monthly mean IOP values of right eye, first-visit Goldmann IOP from patients seen at institution from January 2012 thru December 2017 (dataset C). The 12-month data from each year was expanded into a Fourier series showing approximations after keeping the first order (1 cycle/year) and second order harmonics.

Discussion

This analysis, using multiple datasets, demonstrated a yearly IOP pattern of variation. Although we did not find that IOP was uniformly highest or lowest in specific months with all datasets, measurable trends consistently supported a yearly cycle of IOP variation with certain tendencies. Generally, we did observe that IOP was highest around December or January and lowest around August or September. What seemed to be most consistent among our datasets was that IOP tended to be higher in colder months and lower in warmer months. Although not discernible with the smaller datasets A and B, it was interesting that the larger dataset C suggested a “plateau” of IOP that existed from about October to April, followed by an ensuing trough that developed from May to September. It is evident that we cannot make firm conclusions about this definitive pattern, but it would be interesting to see how even larger datasets from around the world may, or may not, mirror these observations. These results agree with previous studies that have looked at seasonal or monthly IOP variation. Monthly IOP in the Ocular Hypertension Treatment Study¹⁷ demonstrated similar findings, suggesting IOP is higher during the northern winter than the summer.

Although this analysis shows that the magnitude of the time-of-year IOP variation was modest, i.e., about 1 mm Hg on average from peak to trough, it could be significant and could perhaps influence management strategies. It should be kept in mind that this magnitude represents an “average” variation, which could be larger in some individual subjects. Our findings confirm recent analyses by Terauchi et al., that reported winter IOP was higher than summer IOP in both healthy and primary open angle glaucoma eyes. Their results also show IOP fluctuations in and around 1 mm Hg.^45,46 Analysis of datasets A and B excluded patients with IOP lowering medications and patients with a history of cataract surgery, which confirm the findings in their healthy population. In dataset C, the same yearly trend was seen across all patients and with glaucoma patients excluded, hence also excluding those with IOP lowering medications. Interestingly, Terauchi et al.’s studies were carried out in Tokyo, Japan where the weather undergoes seasonal changes that is similar to the seasonal weather changes in Chicago, Illinois, USA, i.e., the location of our study. Future studies may look at populations where weather is consistent year-round to determine if these variations still exist.

As previously noted, lowering IOP through medical or surgical treatment options has been shown to reduce the risk of progression of glaucoma. It is noteworthy that the Early Manifest Glaucoma Trial showed that IOP lowering of just 1 mm Hg yielded a 10 to 13% risk reduction in visual field deterioration.^3,9,35 Further, the Ocular Hypertension Treatment Trial⁴ demonstrated that this amount of IOP lowering resulted in a 10% reduction in the risk of glaucoma conversion. To maintain a level reduction in risk of glaucoma progression, this suggests that, at least for certain patients, management might need to be altered during certain times of the year.

Although the physiologic mechanism for this annual fluctuation in IOP is unknown, several hypotheses have been postulated, including hormonal melatonin secretion from the pineal gland.^{25,26,28–30,47,48} Blood levels of melatonin are highest at night and lowest during the day. Secretion is regulated in the suprachiasmic nucleus of the hypothalamus, which is regulated by light. Melatonin is increased in the summer months with more daylight exposure. Melatonin increases the release of estrogen and progesterone, which have been shown to reduce IOP by increasing aqueous humor outflow.^30,48–51 Depression has also been linked to a decrease in melatonin secretion. Dating back to 1970,¹⁶ Stojek noted variation in seasonal affective disorder and intraocular pressure.^27,28,30 Intraocular melatonin receptors have been identified and may be involved in aqueous humor dynamics.

Exercise has also been shown to have an effect on IOP. Two recent reviews by Zhu et al.⁵² and Risner et al.⁵³ demonstrated that IOP decreases after dynamic exercise, such as jogging or cycling. The effect of isometric exercise, such as weightlifting, on IOP is more controversial. It has been postulated the mechanism by which exercise reduces IOP is quite complex, however, may be associated with lowering of norepinephrine and the release of nitric oxide and endothelin post-exercise, coupled with the gene polymorphism of beta2-adrenergic receptors.⁵⁴ As previously mentioned, Chicago, Illinois, USA undergoes seasonal changes and perhaps IOP is lower in summer months as outdoor activity, such as walking, jogging and cycling, is more accessible and enjoyable.

Recently, Ayaki et al.⁵⁵ reported seasonality of IOP correlated with seasonality of dry eye disease, most notably tear break-up time (TBUT). It was demonstrated that seasonality of IOP was greater with a shorter TBUT, compared with a normal TBUT, and IOP was lower in patients with a shorter TBUT. They hypothesized that ocular surface inflammation, which is worse in winter months, may lead to increased IOP due to high transforming growth factor-beta (TGF-beta). It has been shown that TGF-beta levels in the anterior chamber can lead to dysfunctional contractions of the trabecular meshwork via Rhokinase (RhoA) signaling in early open-angle glaucoma.⁵⁶ This is a cellular pathway that may be further explored in future studies.

In terms of seasonal IOP variation, a relevant question is whether IOP variation magnitude, or trend, may be different from one year to another depending on monthly and daily temperature differences among years. For example, if the temperature in November is warmer or cooler in one year vs. another, does this have any effects on relative IOP measurements? We believe this may be an interesting question to explore but the effects may be extremely small and difficult to measure without careful adjustment for many other variables. The increasing availability of very large datasets may make this a more feasible question to answer, but a potential problem may be that any effects could also be negated by spending most time indoors.

The current analysis supports the hypothesis that IOP has an independent relationship with time of year, and it shows that large datasets may be necessary to detect smaller variations and more precisely model the nature of relationships. Although our initial exploratory approach was linear to detect a recurrent yearly pattern that adjusted for multiple variables, we emphasize that this approach primarily enabled us to initially establish an approximate trend that recurred annually. The advantage of the linear model is that it allowed for “adjustment” of potential confounding variables provided by the primary dataset A. Although datasets B and C did not afford the control of many potential confounding variables, they nonetheless provided consistent support for the trend observed with dataset A. Datasets B and C also helped provide a different analysis using the Fourier analyses, which further supported that IOP fluctuation has an annual periodicity.

We emphasize here that neither the linear regression analyses nor the Fourier analyses provide perfect insight into the nature of our data, but together, these different approaches complimented one another with some confirmatory observations. The advantage of the linear regression analysis on Dataset A was that we were able to model linear characteristics of the annual trend with the monthly reordering, i.e., November to October, while also adjusting for many potential confounding variables such as age, gender, refractive error, blood pressure, smoking, etc. The advantage of the Fourier analysis was that it provided a novel way to confirm the observation of a recurring annual pattern of IOP fluctuation. Alone, the Fourier analysis does not adjust for potential confounding variables, but the consistency of that analysis with the regression analyses adds another level of confirmation of an annual IOP pattern. A strength of Fourier analysis is that it can help identify patterns or cycles in time series data. The information gleaned from this approach may be helpful to future investigations of cyclical IOP variation.

A limitation of this study is that is uses a single facility patient population, which might not be representative of other groups or populations of subjects. Nonetheless, the observations provided give further insight that can be helpful to future investigations into the IOP and time-of-year relationship. It will be interesting to determine if similar observations can be detected in other populations and whether patterns appear related to factors such as temperature, systemic adrenergic variability, other endocrine markers or endogenous substances and ocular surface inflammatory markers. Another limitation of this study is that IOP measurements were taken during the daytime only, i.e., during business hours. Therefore, we cannot comment on IOP variation that may have occurred outside of our normal business hours. We emphasize, however, that we did control for IOP diurnal variation that may have occurred during our hours of measurement. For this, we used the single measurement from each person but adjusted for the time of day according to various categorizations, e.g., hourly, AM vs. PM, etc. These models showed that the time-of-year IOP variation that we observed was not confounded by any daily diurnal fluctuation.

Since this study is cross-sectional in nature, it is unknown how the IOP of our individual subjects varied over time. A longitudinal design where IOP measurements taken at varying times of year from individual subjects would provide this type of information. However, this does not invalidate the observations made from a cross-sectional design, especially given the large numbers of subjects we included who were tested throughout the year and during the entire length of the business day. Further, the strength of cross-sectional designs to make observations that can then be tested with other designs cannot be overstated. It is clear that cross-sectional study designs are often used to initially show associations which form the basis of hypotheses that are later used to demonstrate causation. This is especially true when adequate numbers of subjects and covariables allow adjustment for potential confounding. Findings become even more important when they are consistent with existing knowledge and when general trends are repeatable with different datasets. Collectively, our analyses included each of these features. With our primary dataset A, we were first able to show a definitive relationship between IOP and time of year while incorporating exploration of many potential confounders. Although our supplemental datasets B and C didn’t offer exactly the same control as our primary dataset A, the observational trends in IOP and time of year were consistent and further supported conclusions.

The harmonics analysis used with dataset C is noteworthy in that a cyclical pattern seems to emerge, whereby the main observation is a drop in IOP during warmer months with subsequent return to prior levels with the onset of colder months. This observation shows why a downward “linear” pattern would emerge upon reordering of the months beginning with November. An advantage of reporting all three datasets together is that it more completely shows the nature of the data and how it might vary depending on the size of the dataset. We believe the totality of this information may be helpful to future analyses.

Conclusions

This analysis showed that IOP level had a yearly cyclical pattern. These data further support that time of year may influence IOP level. Although time-related variation in IOP may be small, it is demonstrable with an adequately sized dataset and appropriate methods and may be worthy of further study relative to glaucoma control. We found trough to peak IOP to be ≈1 mm Hg or less among our data. Our most consistent observation about timing of IOP variation was that it appears to trend lower during summer months.