Visual Functioning and Quality of Life under Low Luminance: Evaluation of the German Low Luminance Questionnaire

Abstract

Purpose.: To validate the German-translated Low Luminance Questionnaire (LLQ), a vision-related quality of life scale assessing mainly mesopic and scotopic functioning, and to determine the relationship between the severity of vision impairment, ocular conditions, and low luminance–related visual functioning.

Methods.: In all, 274 participants, 184 patients with visual acuity <6/12 or a long-standing symptomatic eye condition and 90 controls, were recruited from an outpatient clinic at a German eye hospital. Participants underwent a clinical examination and completed the German LLQ and VF-14 scales. The validity and psychometric properties of the scales were assessed using Rasch analysis exploring key indices, such as instrument unidimensionality, discriminant ability, and targeting of item difficulty to patient ability. Multivariate analyses of low luminance functioning were adjusted for conventional visual functioning (VF-14 scores).

Results.: The 30-item German LLQ initially displayed poor fit to the Rasch model. Following Rasch-guided iterative adjustments to the scale, a 23-item LLQ emerged as a valid and unidimensional scale. Visual functioning under low luminance consistently declined with worsening vision loss. Compared with patients with no vision impairment, those with mild or moderate/severe vision impairment recorded significantly poorer low luminance functioning scores (mean change, −6.33 and −16.62; P = 0.032 and P < 0.001, respectively). Age-related macular degeneration and cataract were independently associated with low luminance visual functioning, as was worse self-reported health.

Conclusions.: Low luminance functioning is considerably compromised in visually impaired patients even at the mild spectrum of visual acuity loss. Additionally, the impact of age-related macular degeneration and cataract on patients' low luminance functioning is substantially independent of vision impairment.

The effect of vision impairment on visual functioning and vision-related quality of life (VRQoL) has been well documented. ^{1 –3} Evidence suggests that vision impairment has an impact on activities of daily living, falls, and mobility ^{4 –6} as well as emotional well-being. ^7,8 The impact of vision impairment from the patient's perspective is usually assessed using a validated patient-reported outcome (PRO) measure. The assessment of functional ability is generally assumed to occur in good lighting conditions. ^9,10 Rarely are patients asked to rate their functional ability in difficult lighting conditions, and most questionnaires contain only a very limited number of items pertaining to this. This is important because visually impaired patients may exhibit satisfactory levels of visual functioning in optimal lighting, but can report poorer functioning in suboptimal lighting conditions, such as at dusk and dawn, in the dark or glare. ¹¹ Those with retinal disorders such as age-related macular degeneration (AMD) ^{12 –14} or rod and cone dystrophies ^15,16 may be particularly affected. Furthermore, some evidence suggests that progressive vision loss from AMD could be predicted by poor self-reported night vision. ¹⁷  

However, evidence regarding the specific effect of low luminance on patients' visual functioning is limited. One study has shown that contrast and glare sensitivity, stereoacuity, and visual fields, independent of visual acuity, are significant risk factors for self-reported visual disability in elderly populations. ¹⁸ Similarly, findings from the Smith-Kettlewell Institute longitudinal study of vision function and its impact among the elderly showed that older people with good visual acuity were effectively visually impaired when performing everyday tasks in suboptimal lighting conditions. ¹¹ Furthermore, participants' visual ability under nonideal conditions could not be predicted from their visual acuity alone. ¹¹  

Limited research in the area of low luminance–induced functioning may be related to a lack of suitable PROs. One instrument, the Low Luminance Questionnaire (LLQ), has recently been developed to measure the impact of mesopic (in intermediate levels of illumination) and scotopic (in low levels of illumination) vision impairment on visual functioning ⁹ and has been shown to be internally valid and reliable ⁹ and responsive to interventions. ¹⁰ However, it has not been subjected to validation using modern psychometric theory. 

Therefore, in this study, we first determined the validity, reliability, and measurement characteristics of the German-translated LLQ using Rasch analysis. We then investigated the relationship between the severity of vision impairment, the main causes of vision loss and visual functioning under low luminance in a sample of German outpatients. 

Materials and Methods

Recruitment

This cross-sectional, observational study took place from March until May 2009 at the Department of Ophthalmology, University of Bonn, Germany, where patients were recruited from outpatient clinics. Inclusion criteria were: having presenting visual acuity worse than 6/12 or a long-standing symptomatic eye condition; 18 years of age or older; and ability to converse, read, and write in German. Participants underwent a complete ophthalmic examination including presenting and best-corrected visual acuity; biomicroscopy; intraocular pressure measurements; and funduscopy. Further diagnostic tests (fluorescein angiography, optical coherence tomography, and electrophysiology) were performed as appropriate in each individual case. All patients were administered the German LLQ, and sociodemographic characteristics and medical history were obtained using two trained interviewers. Sample size calculations were based on good practice in validating psychometric instruments that require at least the number of items times five participants for a thorough evaluation. ¹⁹ Ethical approval was obtained from the ethics committee of the University of Bonn. Informed consent was obtained from every participant before the interview. The study adhered to the tenets of the Declaration of Helsinki. 

The Low Luminance Questionnaire

The original LLQ was developed by Owsley and colleagues ⁹ to assess VRQoL under low luminance in patients with AMD. It contains 32 items summarized into six subscales (Table 1). Each item has five to six response options targeting difficulty or frequency using Likert scaling, ranging from “no difficulty at all” or “none of the time” to “stopped doing because of your vision.” Two additional response options represent nonapplicable or missing data: “stopped for other reasons” and “don't do.” The original LLQ questionnaire was translated into German, and back-translated by a professional translator. After a pilot test with 10 German patients to assess comprehensibility and appropriateness of each question, items 18 and 19 were removed because patients felt they were not relevant or appropriate to their daily lives. This final draft was once again reviewed by both a clinician and the professional translator. 

Table 1.

 

Structure, Item Content, and Subscales of the Original Low Luminance Questionnaire (LLQ) and the Rasch-Validated German LLQ-23

Tool	Item	Subscales	German LLQ-23
LLQ	1. Difficulty seeing in bright sunlight	Extreme Lighting	Omitted-Rasch
	2. Difficulty seeing in fluorescent lighting		Omitted-Rasch
	3. Difficulty seeing faces in hallway with direct sunlight behind the face		Omitted-Rasch
	4. Difficulty reading menus in dimly lit restaurants
	5. Difficulty reading newspaper without good lighting
	6. Upset because of difficulty seeing while driving in the rain at night		Omitted-Rasch
	7. Difficulty reading material printed on dark or colored material
	8. Difficulty seeing dark-colored cars at night
	9. Because of vision, bothered moving around in darkened theaters	Mobility
	10. Because of vision, difficulty going to nighttime social events		Omitted-Rasch
	11. Because of vision, dependent on others to move around at night or under poor lighting
	12. Because of vision, concerned about falls at night
	13. Difficulty seeing colors at night
	14. Difficulty seeing furniture in dimly lit rooms with dark floors
	15. Difficulty seeing at night	General dim lighting
	16. Difficulty seeing in poor lighting conditions
	17. Difficulty with depth perception at night
	18. Difficulty seeing in candlelight		Omitted-FGD
	19. Difficulty seeing when visiting other homes because of lack of light		Omitted-FGD
	20. Difficulty seeing under kitchen counters, in cabinets, or in closets because of lack of light
	21. Difficulty with peripheral vision under poor lighting conditions	Peripheral vision
	22. Difficulty with peripheral vision at night
	23. Difficulty with peripheral vision in bright sunlight
	24. Difficulty reading street signs when driving at night	Driving
	25. Difficulty due to headlights of oncoming cars while driving at night
	26. Limited driving in the rain at night because of vision		Omitted-Rasch
	27. Limited driving at night due to vision		Omitted-Rasch
	28. Difficulty seeing while driving at dusk/dawn due to glare
	29. Worry about making a mistake at a social event due to poor vision in poor lighting	Emotional distress
	30. Feel bad/depressed about ability to see at night/in poor lighting
	31. Feeling bad/depressed because poor vision at night/in poor lighting keeps you from doing the things you want to do
	32. Feel bad/depressed due to inability to help others as much as you want to because of poor vision at night/in poor lighting

 Rasch, Rasch analysis; FGD, focus group discussion.

Visual Functioning: The VF-14

The VF-14 contains items related to the degree of difficulty in performing 14 vision-dependent activities (e.g., reading, watching television). ²⁰ We used the German VF-14 to assess conventional visual functioning, which we then used in our analyses to control for and differentiate between conventional and low luminance visual functioning. The psychometric evaluation using Rasch analysis and the VF-14 data for this sample have been described in detail elsewhere. ²¹  

Psychometric Validation of the German LLQ

Rasch analysis is a modern psychometric method that mathematically describes the way respondents interact with test items. In the Rasch model, the probability of a correct response is modeled as a logistic function of the difference between person ability (person measure) and item difficulty (item measure). Rasch analysis therefore provides a strict model that the structure of the responses should satisfy, rather than a simple statistical description of the responses. ^{22 –25} Estimation methods are used to obtain estimates from matrices of response data based on the model and raw ordinal scores are thus transformed into data that approximates interval-level measurement (expressed in log of the odds units, or logits). A high person measure (in logits) indicates a high level of the assessed latent trait (e.g., low luminance functioning). To ease interpretation, the rating scale of the LLQ was reversed so that patients with a high level of low luminance functioning were given high scores. Rasch analysis also provides insight into the psychometric properties of a scale, such as how well items fit the underlying latent trait being measured, how well items discriminate between the respondents, how well item difficulty targets person ability, and the appropriateness of the response scale used. ²⁶  

To test the psychometric properties of LLQ questionnaire, we fitted the data to the Rasch model using commercial software (Winsteps software version 3.68, Chicago, IL) ²⁷ and the Andrich rating scale model. ²⁸ Three rating scales were applied to this questionnaire because there were three sets of response options with differing characteristics. Several key indicators for fit to the Rasch model were investigated. First, response categories were checked for disordered thresholds via visual inspection. Disordered thresholds can result when certain response options are underused by participants. This may be the source of significant misfit and require collapsing of response categories to improve the model fit. The capacity of the scale to discriminate between different strata of person ability was then assessed by the Person Separation Index (PSI) and Person Reliability (PR) coefficients (minimum value of 2.0 and 0.8, respectively). ²⁹  

The Rasch model requires that scales measure a single underlying trait, or are unidimensional, which is assessed by several “fit” statistics. How well each item fits the underlying trait is represented by “infit” mean square standardized residuals (MNSQ), measured as item fit or misfit. A value of 0.7–1.3 is considered acceptable, and lower or higher values may indicate redundancy or unacceptable variation in the responses, respectively. The second statistic is the principal component analysis (PCA) of the residuals, which tests for local independence in the scale. Ideally, the first factor should explain at least 50% of the variance and the eigenvalue of the second component should not exceed 2.4, because this suggests the existence of a second dimension (i.e., the scale does not function as a single measure). If multidimensionality exists, the PCA standardized residual loadings (i.e., loading value of >0.4) on the first contrast are examined to determine whether certain items are loading together. The content of any positively loading items identified is assessed for evidence to split the scale and form a conceptually relevant second dimension. ²⁸  

Further measures of unidimensionality include norm reference correlational analyses, which involves “item split” whereby items are split into two scales based on their positive and negative loadings onto the first factor, and graphically depicted dimensionality by cross-plotting of the split scales. For the correlational analysis, we obtained the reliability measures for both split scales (items with positive and negative loadings; R₁ and R₂), and then correlated the person measures of both half-scales, using Pearson's correlation coefficient (C). If the resulting value of C/√(R₁ × R₂) approximates 1.0, this indicates that both halves of the scale essentially measure the same construct, and thus are unidimensional. ³⁰ Visually, the person measures from each half of the scale should display a similar trend in a cross-plot, with very few persons being located off the diagonal distribution. ³⁰  

The targeting of the instrument is determined through visual inspection of the person–item maps and the difference between person and item mean logits. A difference of >1.0 logits indicates suboptimal targeting. Differential item functioning (DIF) was assessed to explore whether sample subgroups systematically respond differently to items, despite having similar underlying ability. A DIF contrast of >1.0 logits for an item indicates notable DIF and indicates the presence of interpretation bias. Finally, person location scores (in logits) were recalibrated to a 0 to 100 scale to make the results easier to interpret. Rescaling the Rasch scale, which is theoretically indefinite at both ends, to 0 to 100 is controversial. However, in practice, the scale is never unlimited at either end and much easier to understand ranging from 0 to 100, because this conforms with most original questionnaire scales and is generally more intuitive. ³¹  

Statistical Analysis

Commercial statistical/analytical software (SPSS Version 17.0; SPSS Science, Chicago, IL) was used to analyze the data. Descriptive statistical analyses were performed to characterize the participants' sociodemographic and clinical characteristics and ability to function under low luminance. Low luminance functional ability, as assessed by the composite score of the LLQ, was the main outcome examined. Multiple linear regression models were conducted to determine the independent factors associated with the LLQ score. The covariables adjusted for in the multivariable analyses were those found to be univariately associated with low luminance visual functioning. All models were adjusted for reported conventional visual functioning (i.e., the VF-14 scores). Visual acuity was categorized into three categories: normal presenting vision in the better eye (≤0.3 logMAR); mild visual impairment (0.3 < logMAR < 0.5); and moderate to severe visual impairment (logMAR ≥ 0.5). 

Results

Sociodemographic and Clinical Characteristics of the Participants

A total of 274 participants were recruited comprising 184 (67.2%) patients with varying degrees of vision impairment and 90 (32.8%) controls (no vision impairment and no eye condition). Participants' mean ± SD age was 59.4 ± 21.76 years and there were more female (58.4%) than male participants (Table 2). Participants' self-reported health was generally poor, with 40.5% (n = 111) reporting only poor to fair health. Most reported having at least one nonocular comorbidity, with approximately two-thirds reporting having hypertension (59.7%), cardiovascular diseases (46.2%), and diabetes (30.3%). 

Table 2.

 

Sociodemographic and Clinical Characteristics of the Study Participants (n = 274)

Variable	n (%)	LLQ-23 Score
		Mean ± SD	P
Total sample	274 (100%)	51.12 ± 24.53
Participants			<0.001
Cases	184 (67.2%)	40.67 ± 19.60
Controls	90 (32.8%)	77.41 ± 13.47
Sex*			0.066
Female	160 (58.4%)	48.24 ± 25.14
Male	108 (39.4%)	54.02 ± 22.58
Age, y*			<0.001
≤60	111 (40.5%)	60.37 ± 25.50
>60	142 (51.8%)	41.15 ± 18.91
Education*			<0.001
None/primary school	101 (36.9%)	39.95 ± 21.19
Some secondary	62 (22.6%)	45.01 ± 21.32
Secondary completed	57 (20.8%)	70.39 ± 18.62
University/other higher education	52 (19.0%)	63.05 ± 23.61
Work status*			<0.001
Not working	158 (57.7%)	41.40 ± 20.70
Working	107 (39.1%)	67.78 ± 21.39
Vision impairment			<0.001
None	161 (58.8%)	64.77 ± 20.33
Mild	49 (17.9%)	42.20 ± 15.80
Moderate/severe	64 (23.4%)	28.11 ± 17.05
Eye conditions			<0.001
None	90 (32.8%)	77.41 ± 13.47
AMD	54 (19.7%)	36.72 ± 15.80
DRP	28 (10.2%)	31.99 ± 17.69
Glaucoma	15 (5.5%)	47.40 ± 20.52
Other retinal disease	51 (18.6%)	46.30 ± 21.20
Cataract and corneal disease†	9 (3.3%)	37.96 ± 17.48
Other	27 (9.9%)	44.38 ± 21.85
Self-rated general health*			<0.001
Very good to excellent	69 (25.2%)	76.81 ± 16.43
Good	91 (33.2%)	55.61 ± 21.06
Fair to poor	111 (40.5%)	36.33 ± 18.76
Other nonocular comorbid conditions
Hypertonia	111 (59.7%)	39.98 ± 19.06	0.470
Diabetes	56 (30.3%)	36.94 ± 19.50	0.080
Pulmonary diseases	38 (20.4%)	36.32 ± 22.90	0.101
Cardiovascular diseases	85 (46.2%)	35.98 ± 18.59	0.003
CNS diseases	21 (11.4%)	32.32 ± 14.96	0.038
Restricted mobility	67 (36.2%)	35.77 ± 19.39	0.009
Depression and other mental illness	17 (9.2%)	35.45 ± 25.18	0.260
Neoplastic diseases	28 (15.2%)	39.11 ± 19.33	0.649
Other chronic diseases	66 (36.5%)	38.97 ± 19.79	0.295

  PP values in bold indicate statistical significance.

*  Data incomplete.

†  60% of this category comprises patients with cataract.

Of the 184 cases, 38.6% (n = 71) had normal presenting vision in the better eye (≤0.3 logMAR); 26.6% (n = 49) were considered mildly vision impaired (0.3 < logMAR < 0.5); and 34.8% (n = 64) were considered moderately or severely visually impaired (logMAR ≥ 0.5). The main cause of vision loss was AMD, followed by “other retinal diseases.” 

Psychometric Validation of the German LLQ

The validity, reliability, and unidimensionality of the German LLQ were explored using Rasch analysis (Table 3). The 30-item LLQ scale had disordered thresholds in two of its three rating scales, suggesting that use of certain response categories was suboptimal. Consequently, categories 2 and 3 (“completely blind under these conditions” and “a lot of difficulty”) were collapsed for the first rating scale and categories 4 and 5 (“a little difficulty” or “some difficulty”) were collapsed for the second rating scale, which resulted in five final response options for each. The 30-item LLQ displayed excellent discriminant ability with PSI and PR values of 3.76 and 0.93, respectively, indicating that it was able to discriminate between five strata of low luminance ability. However, the targeting of the instrument was suboptimal, with a 1.83 logit difference in person and item means, suggesting that participants had a higher level of ability than the average of the scale items (i.e., 0 logits). Moreover, although the PCA for the first factor explained >60% of the variance, the unexplained variance in the first contrast of the residuals was 3.1, suggesting evidence of multidimensionality. Similarly, three items (Items 1, 2, and 20) demonstrated substantial misfit (MNSQ > 1.3). 

Table 3.

 

Fit Parameters of the LLQ Scale Compared with Rasch Model Requirements

Parameter	Rasch Model	LLQ-30	LLQ-23	LLQ-23 without Extremes
Disordered thresholds	No	Yes	No	No
Number of misfitting items	0	3	1	1
Person Separation Index	>2.0	3.76	3.74	4.65
Person Reliability	>0.8	0.93	0.93	0.96
Difference between person and item means	<1.0	1.83	2.06	1.65
Variance by 1st factor	>50%	69.6%	72.1%	72.1%
PCA (eigenvalue for 1st contrast)	≤2.4	3.1	2.2	2.2
Differential item functioning (DIF) (Item number [DIF contrast])
Sex	<1.0	na	None	None
Age group (≤50 y, >50 y)	<1.0	na	None	None
Vision impairment, yes, no	<1.0	na	None	None

 na, not assessed.

The standardized residual loadings in the PCA were initially examined to identify potential other dimensions in the scale. Although a group of six items (mostly pertaining to driving) loaded together and were removed, the fit statistics of the overall scale did not improve. Moreover, the misfitting items remained and further examination of the PCA loadings revealed additional groupings. Iterative removal of these items finally resulted in a substantially shortened scale (data not shown) and thus this line of analysis was abandoned. 

Instead, the three misfitting items initially found ^1,2,20 were removed, which improved the fit statistics, but unearthed four subsequent misfitting items (Items 3, 8, 24, 30). Removal of three of these items (Items 3, 8, and 24, but not 30) further improved the fit to the Rasch model (Table 3), resulting in a 24-item LLQ scale. DIF for age, sex, and severity of vision impairment was investigated for the LLQ-24. Notable DIF (1.36) was found for vision impairment for item 25 “limiting driving at night” and this item was deleted, resulting in a final 23-item LLQ scale (Table 4). 

Table 4.

 

Differences in the 23-Item LLQ Overall Score by Age, Sex, Education, Work Status, Categories of Vision Impairment, Self-Reported Health Rating, and Eye Condition in Multivariate Modeling, Adjusted for Reported VF-14 Scores

Variable	RC (95% CI)	P
Age (y)		0.708
≤60 (Reference)	48.50 (44.11 to 52.88)
>60	1.20 (−5.11 to 7.52)
Sex		0.119
Female (Reference)	47.48 (43.86 to 51.11)
Male	3.23 (−0.84–7.29)
Education
Trade, TAFE, or University (Reference)	52.01 (46.70 to 57.33)
Secondary completed	−3.87 (−10.82 to 3.08)	0.273
Some secondary completed	−5.84 (−12.47 to 0.79)	0.084
None/primary school only	−1.96 (−8.12 to 4.19)	0.531
Work status		0.388
Working (Reference)	50.59 (45.64 to 55.55)
Not working	−3.00 (−9.82 to 3.83)
Categories of VI
Normal (Reference)	56.75 (52.96 to 60.53)
Mild	−6.33 (−12.10 to −0.57)	0.032
Moderate/severe	−16.62 (−22.49 to −10.76)	<0.001
Self-reported health rating
Very good to excellent (Reference)	55.20 (48.38 to 62.01)
Good	−4.61 (−12.04 to 2.83)	0.223
Poor to fair	−13.67 (−22.08 to −5.31)	0.001
Eye condition
None (Reference)	55.29 (49.23 to 61.35)
AMD	−9.63 (−18.65 to −0.61)	0.037
Diabetic retinopathy	−9.51 (−19.52 to 0.50)	0.062
Glaucoma	−0.16 (−10.85 to 10.54)	0.977
Other retinal diseases	−4.76 (−13.00 to 3.48)	0.256
Cataract and corneal diseases	−12.29 (−23.89 to −0.70)	0.038
Other	−7.03 (−16.02 to 1.96)	0.124

 Adjusted means are given for the reference categories. RC, regression coefficient; CI, confidence interval; VI, visual impairment. Bold coefficients represent independent variables found to be statistically (P < 0.05) associated with the LLQ overall score. Bold and underlined coefficients represent independent variables found to be statistically (P < 0.05) associated with the LLQ overall score and clinically meaningful.

Unidimensionality of the 23-item scale was explored using PCA, explained variance, factor loadings, and item MNSQ In- and Outfit values for the first factor. PCA of fewer than three eigenvalues in the first factor is indicative of unidimensionality, given that anything fewer than three items does not constitute a separate dimension. The number of items loading together on a separate factor is indicated by the rounded eigenvalue of the first contrast (i.e., an eigenvalue of <2.5 indicates that 2 items are loading onto that particular factor). ³⁰ The eigenvalue of the first contrast of the German LLQ-23 is 2.2, indicating that only 2 items load onto this factor. According to the PCA loadings, the two items that loaded onto the first factor (loadings of 0.65 and 0.57) had an MNSQ In- and Outfit of <1.0 (0.88, 0.79 and 0.93, 0.84, respectively). MNSQ values of <1.0 indicate “local intensification of the measured dimension” described by Linacre, rather than constituting a separate dimension. ³⁰ Moreover, the raw variance explained by the measures was 72.1%, the total raw unexplained variance was 27.9%, and the unexplained variance in the first factor was 2.7%. Random simulated data that fit the Rasch model are expected to have a second dimension (first contrast) of up to 4% raw unexplained variance, which is more than we observe in these data. ³⁰ Furthermore, the total raw variance explained by the measures was >10-fold larger than the variance explained by the first contrast, which represents excellent fit to the Rasch model. ³⁰  

In addition, the norm reference correlational analysis for the LLQ-23 indicated that the scale was unidimensional, because the value obtained was 0.8, which approximates one. Furthermore, the person measures of the positive and negative halves of the scale display a similar trend in the cross-plot, with only a negligible number of persons being located off-diagonal, thus providing further supporting evidence for unidimensionality (Fig. 1). Taken together, the fit statistics of the LLQ-23 and the additional in-depth exploration into the PCA indicate that the scale is valid, reliable, and unidimensional. ³⁰  

Figure 1.

 

The cross-plot of person measures of the positively and negatively loading items of the German LLQ-23 demonstrates a similar trend of both halves and indicates that the whole scale is unidimensional.

The LLQ-23 also had excellent discriminant ability and no notable DIF. However, the targeting of the scale was poor, most likely because a large proportion of the sample (58.8%) was not visually impaired and therefore had a higher ability than that required by the average item difficulty. Therefore, persons with extreme scores at the high spectrum of ability level (n = 21) were thus removed from the analysis, which markedly improved the targeting as well as the PSI and PR scores (person–item–map displayed in Fig. 2). Collectively, these results show that the final LLQ-23 is a unidimensional, reliable, and valid scale to assess low luminance functional ability in this population. 

Figure 2.

 

Person–item–map in logits for the LLQ-23 scale with extremes removed.

The participants' mean ± SD overall score for the LLQ-23 was 51.12 ± 24.53. In linear regression models, independent significant predictors of vision functioning were considered to be clinically meaningful if the confidence interval limits of their beta coefficients were >12.5 or < −12.5, which is approximately half the SD of the overall mean. This is generally considered to be a useful estimate of a clinically meaningful difference, as shown in similar studies. ^32,33  

Relationship between Vision Impairment and Vision-Related Quality of Life

Low luminance functioning decreased steadily with worsening vision. Participants with mild (mean ± SD, 42.20 ± 15.80) and moderate/severe (mean ± SD, 28.11 ± 17.05) vision impairment reported significantly poorer low luminance functioning compared with participants with normal vision (mean ± SD, 64.77 ± 20.33, P < 0.001) (Fig. 3 and Table 2). In addition to vision impairment, ocular conditions, age, education, work status, suffering from cardiovascular disease, CNS disease, or restricted mobility, and self-rated general health were also univariately associated with the overall low luminance functioning score (all P < 0.05; Table 2). 

Figure 3.

 

Relation between better eye visual acuity (x-axis) and reported low luminance functioning, using the LLQ-23 scale with extremes removed (y-axis).

Mean ± SD of the VF-14 visual functioning score after Rasch-guided estimation of interval level measurements was 52.55 ± 19.57 for cases and much better for controls (91.67 ± 12.87; P < 0.001). Vision functioning was significantly poorer in persons with mild as well as moderate/severe visual impairment (both P < 0.001) compared with persons with no visual impairment (Table 2). Detailed information regarding the Rasch results and the impact of visual impairment and ocular conditions on visual functioning as measured with the VF-14 has been reported elsewhere. ²¹ Both measures (i.e., the LLQ-23 and the VF-14) were highly correlated (Pearson correlation coefficient 0.701, P < 0.001). 

In adjusted general linear models, controlling for conventional visual functioning (VF-14 scores), only severity of vision impairment, the presence of certain ocular diseases, and self-rated general health remained independently associated with the overall low luminance functioning score (Table 4). Vision-specific low luminance functioning consistently declined with worsening vision impairment as indicated by the model's β-coefficient. For example, compared with people with no vision impairment, those with mild or moderate/severe vision impairment recorded significant declines in LLQ scores (β-coefficients −6.33 and −16.62; P = 0.032 and P < 0.001, respectively). Based on our estimates of clinically meaningful differences (change of >12.5 or ≤12.5), this independent association was clinically significant for those with moderate/severe vision impairment. The presence of AMD and cataract and corneal diseases were also independently associated with worse low luminance functioning (Table 4). Compared with those with “very good” self-reported health, patients with “poor/fair” self-reported health reported poorer low luminance functioning (β-coefficients −13.67; P = 0.001; Table 4). This association was clinically meaningful. 

Discussion

We investigated the relationship between vision impairment and low luminance functioning in a clinical sample of German outpatients. We first established the psychometric properties of the German version of the LLQ (30 items) using Rasch analysis. After collapsing disordered thresholds of the response options, removing seven misfitting items, and dropping participants with extreme scores, we found the LLQ-23 to be a valid, reliable, and unidimensional scale to assess low luminance functioning. Further analyses showed that vision impairment, age-related macular degeneration, cataract and corneal diseases, and poorer self-reported health were associated with poorer low luminance functioning. Even at mild levels of vision impairment, participants reported poor low luminance functioning. 

It has been suggested that older people in particular may effectively be visually impaired when performing everyday tasks involving low and changing light levels, stereopsis, glare, and low contrast, despite having good visual acuity. ¹¹ Low luminance functioning has only been specifically investigated from the patient's perspective in a small number of studies. ^9,10  

Our finding of an independent association of retinal diseases such as AMD with low luminance functioning is not unexpected. ^{9,10,12,17,34 –36} Owsley and associates ³⁶ found that low luminance functioning as assessed by the LLQ was more responsive to changes in rod-mediated dark adaptation in AMD than general vision-specific functioning as assessed by the National Eye Institute Visual Function Questionnaire. Furthermore, patients in this study reported a greater impact on low luminance functioning than vision-specific functioning in daytime conditions. ³⁶ Our association between cataract and corneal diseases and low luminance functioning is also supported by available evidence. Cataract has repeatedly been associated with decreased contrast sensitivity and increased visual disability with glare, particularly at night. ^{37 –39} Such reduced low luminance functioning may be partly caused by increasing senile lens opacification, ⁴⁰ but other unknown neural processes also seem to play a role in contrast sensitivity and low luminance functioning in the ageing eye. ^{37 –39}  

Our finding that ocular conditions such as AMD and cataract were independently associated with poor low luminance functioning suggests that visual functions such as contrast sensitivity and mesopic and scotopic function may be as important as visual acuity in determining patients' level of functional disability. Controlling for conventional visual functioning (VF-14 scores), no other ocular conditions were associated with low luminance functioning in this sample. Not having measured visual fields, glare, or mesopic function, our ability to infer further associations with reported low luminance functioning is limited. 

The independent association between self-reported health and low luminance functioning found in our study suggests that other factors such as general health and comorbidity may play a role in patients' functional ability under low luminance, in addition to visual impairment. Indeed, studies have shown that a considerable proportion of the variance in reported visual functioning and VRQoL is explained by general health. ⁴¹ However, measurements of visual functioning and VRQoL are not distorted by this and they still remain a valid concept that can be measured using visual functioning and VRQoL scales. ⁴¹  

A key strength of our study is the use of Rasch analysis, an important step in modern scale validation, to assess the psychometric properties of the German LLQ and ensure comprehensive assessment of low luminance functioning. Several aspects of our Rasch analysis deserve comment. First, the resulting 23-item scale contains a mixture of functioning and emotional items. This could be a result of our approach of deleting misfitting items to obtain a fitting solution rather than being guided principally by PCA loadings. However, when PCA loadings were used to identify and remove potential additional dimensions, the scale was substantially reduced and still contained a mixture of functioning and emotional items. Other analyses to resolve this issue were used, including analyzing the emotional and functional items as separate scales (regardless of item fit statistics or PCA loadings), but were similarly unsuccessful. Finally, unidimensionality of the German LLQ-23 was confirmed by several different methods. Rasch-analyzed unidimensional QoL scales containing both functioning and emotional items, although uncommon, have been reported in the literature. ^42,43 The LLQ-23 initially had poor targeting of item difficulty to participant ability, most likely due to the large number of controls in the study with no visual impairment. We substantially improved the targeting of the scale by removing a small number of participants with perfect scores. However, the targeting was still suboptimal and inspection of the person item map suggested that there were too few items of adequate difficulty to challenge the more able participants. Finally, as a result of disordered thresholds, we collapsed two of the LLQ's rating scales from six to five response categories. This is in line with previous findings that ophthalmic questionnaires function optimally with no more than five response categories. ⁴⁴ Overall, therefore, additional improvements in future research could be made to the LLQ's item content, item structure, wording, and response options to further improve its measurement properties. 

As with all psychometric scales, content of instruments developed in other countries may not be appropriate in Germany, which we addressed by including patient feedback on content and wording after translation and back-translation. ⁴⁵ Other strengths include our clinical sample, which represents a continuum of visual acuity and major eye diseases. Conversely, our study was limited by a moderate sample size, which included a number of patients with no significant vision impairment; future studies would benefit from a greater proportion of patients with low vision. Mesopic and scotopic visual functioning or glare was not assessed. Vision impairment under low luminance conditions could well have different associations with LLQ response characteristics in the German version and should be examined in future studies using the German LLQ-23. 

In conclusion, our study found that the German LLQ-23 was a valid, reliable, and unidimensional tool to assess low luminance functioning in this sample. Using the LLQ-23, we found that mild and moderate/severe vision impairment, the presence of AMD, cataract and corneal diseases, and worse self-rated general health were associated with poorer low luminance functioning. PROs, which include functioning in suboptimal and optimal lighting conditions, may better capture patients' vision-related activity limitation. 

Footnotes

 Supported in part by Operational Infrastructure Support from the Victorian Government (Consultation, Education, and Research Associates); and Research Fellowship Grant FI-1540/5-5 of the German Research Foundation (RPF).

Footnotes

 Disclosure: R.P. Finger, None; E. Fenwick, None; C. Owsley, None; F.G. Holz, None; E.L. Lamoureux, None