Development and initial validation of the Cat HEalth and Wellbeing (CHEW) Questionnaire: a generic health-related quality of life instrument for cats

Abstract

Objectives:

The aims of the study were to define factors that owners consider relevant to the health-related quality of life (HRQoL) of cats, to develop an instrument based on this information, and to evaluate the validity and reliability of the final instrument (the Cat HEalth and Wellbeing [CHEW] Questionnaire).

Methods:

Psychometric research techniques and guidance from the US Food and Drug Administration on outcome measures were used to develop a valid and reliable instrument. Fifty-four cat owners and caregivers participated in the qualitative research, while 1303 cat owners were included in the quantitative validation phase (development dataset, n = 648; validation dataset, n = 655). A random subset of cat owners (n = 391) also participated in test–retest evaluation. Qualitative research was used to generate a draft instrument, which was then subjected to quantitative validation techniques. These included item reduction, domain identification, data quality assessment, and exploratory and confirmatory analysis to develop a final instrument, which underwent confirmatory reliability and validity assessment.

Results:

A draft instrument with 11 domains and 100 items based on qualitative research underwent online quantitative validation testing which refined the instrument to eight domains and 33 items. Confirmatory reliability and validity assessment showed that the final instrument had good validity, was able to discriminate between cats by age and overall health status, and demonstrated good internal and test–retest reliability.

Conclusions and relevance:

The CHEW Questionnaire was developed and validated. Additional research is needed to verify its ability to differentiate cats with and without disease, and to assess its potential as a screening tool.

Introduction

Quality of life (QoL) in people has been defined as ‘a broad multidimensional concept that usually includes subjective evaluations of both positive and negative aspects of life.’ ¹ In people, QoL depends on factors such as social relationships, mental stimulation, health, food consumption, stress and degree of control over one’s life.^1,2 When QoL is viewed from a health perspective, it is commonly referred to as health-related QoL (HRQoL), which restricts this complex concept to issues that relate specifically to an individual’s health.

In people, HRQoL is viewed as being multi-dimensional, and incorporates domains encompassing physical, mental, emotional and social functioning. ³ Measures of HRQoL in people rely either on patient self-report (patient-reported outcomes [PROs]) where patients are available and able to report their own health status, or on clinician or caregiver reports (observer-reported outcomes [OROs]) when patients are unable to self-report their health status (eg, children, cognitively impaired individuals).^4,5 Several commonly used generic (not disease-specific) PROs or OROs (eg, SF-36, ⁶ EuroQoL ⁷ ) have been developed to assess overall HRQoL, monitor decline or improvement in overall health status, compare overall disease burden across disease states and populations, and determine cost-effectiveness of treatment. Disease-specific measures also have been developed to assess the impact of a specific disease state or health issue, such as heart failure, ⁸ arthritis, ⁹ depression ¹⁰ and asthma. ¹¹

The measurement of companion animal QoL or HRQoL has received more attention in recent years, and there are a number of instruments for specific diseases, such as heart disease,^12,13 cancer,^14–17 degenerative joint disease,^18–21 diabetes mellitus, ²² spinal cord injury, ²³ dermatologic disease ²⁴ and epilepsy. ²⁵ However, most of the work on HRQoL in companion animals has focused on dogs, with fewer publications on cats. In companion animals, HRQoL is defined by the owner’s perception of the pet’s HRQoL; therefore, an instrument to assess a pet’s HRQoL is classified as an ORO.

There are a number of excellent reviews of the importance of QoL assessment and methodologies in companion animals.^2,26,27 A generic HRQoL instrument could be useful for optimizing care of dogs and cats, including the impact of preventive healthcare to help maintain health status with age, for early diagnosis of disease or decline in health status, progression of disease, or the change in overall health status with certain interventions. This could be particularly useful in cats because they can be difficult to assess, often hide clinical signs of illness, and are taken to see veterinarians less frequently than dogs. ²⁸ The availability of an HRQoL instrument to differentiate between healthy and sick cats and identify diseases or health issues at an earlier stage could be beneficial for the healthcare of cats.

The QoL instruments developed thus far to study dogs and cats range from a single-answer Likert or visual analog scale to more complicated instruments,^{12–16,18–34} and the methods used to demonstrate validity and reliability vary considerably. ¹⁷ Two publications reported on an instrument that assessed non-physical aspects of QoL in dogs,^29,30 but this instrument was not able to differentiate between sick and healthy dogs. Another study on the development of a QoL survey for healthy dogs with four domains (happiness, physical functioning, hygiene and mental status) and 15 items was reported to have good validity, reliability, internal consistency and responsiveness, and to discriminate across age of the dogs; however, the dogs’ health was not reported. ³¹ Another group developed an online HRQoL instrument for dogs with chronic pain caused by degenerative joint disease. ³³

Two studies with validated surveys to assess HRQoL in dogs with spinal cord injuries ²³ and skin disease ²⁴ were able to distinguish between affected dogs and healthy controls. Another study in dogs with cardiac disease showed that the instrument had good reliability and validity, and was highly correlated with measures of disease severity, but healthy dogs were not included in that study. ¹²

While several HRQoL instruments are available for cats with cardiac disease, ¹³ diabetes, ²² cancer ¹⁴ and degenerative joint disease,^20,21,34 these instruments are intended to be more disease-specific or contain language that may not be applicable to healthy cats (eg, how much has your cat’s heart disease negatively impacted his/her comfort or sociability during the past 7 days by making breathing difficult? Does your pet ever show signs of a low blood sugar?). As with HRQoL instruments for dogs, these are proving to be very useful instruments in animals with various disease conditions, especially as endpoints for clinical trials. Recently, a paper by Bijsmans et al ³⁴ reported results of the validation of a general health QoL tool and its use to compare healthy young cats, healthy older cats and cats with chronic kidney disease (CKD). The tool appeared to have good validity and differences were identified in the average weighted impact score between healthy cats and cats with CKD, although there were no significant differences in the average weighted impact score between healthy older cats and cats with CKD. This tool should be valuable for cats with CKD, although its potential role in healthy cats requires more research.

There is a need for additional research on the domains that contribute to HRQoL in order to evaluate how health status may deviate from that of a healthy cat. The ability to more accurately assess generic HRQoL would be useful to alert veterinarians to potential health issues at an early stage and to screen for medical issues, monitor changes during aging, or objectively assess effects of medical and nutritional interventions.

Whether for specific diseases or general health status, rigorously demonstrating the reliability and validity of a new HRQoL instrument is important to ensure that the instrument measures the intended constructs and that any differences observed between animals are reliable and valid.

In 2009, in response to the pharmaceutical industry’s need for guidelines on developing, validating and employing PROs to assess drug treatment effects on a human patient’s HRQoL, the US Food and Drug Administration (FDA) issued a ‘Guidance for Industry’ document with extensive guidelines for developing HRQoL instruments and the measurement properties that a valid and reliable questionnaire should demonstrate for medical claims and claims support. ³⁵ While the guidance specifically addressed claims made about medical products intended for people, the guidelines are based on well-established psychometric principles that can be reapplied for developing reliable and valid HRQoL instruments for companion animals. In their recent paper, Bijsmans et al used a psychometric validation process in the development of their QoL tool. ³⁴

The goals of the study reported here were to: (1) understand and define the factors that cat owners and caregivers consider relevant to the HRQoL of cats; (2) develop an instrument based on this information; (3) assess face and content validity via cognitive debriefing and refine the instrument for quantitative validation; and (4) evaluate the reliability and validity of the instrument via the application of robust statistical analyses and clinical judgment, leading to a reliable, valid instrument. The development of the Cat HEalth and Wellbeing (CHEW) Questionnaire closely followed the 2009 FDA guidance for the development of an HRQoL instrument. ³⁵

Materials and methods

The stepwise process for development of the instrument (ie, the questionnaire) is shown in Figure 1.

Figure 1.

Summary of the study design for developing a health-related quality of life instrument for cats: the Cat Health and Wellbeing (CHEW) Questionnaire

Qualitative research and item generation

Draft instrument development

A series of four focus groups (total participants = 23) was conducted to identify domains considered important for monitoring and assessing a cat’s HRQoL, which included: play, mood, energy, appetite, physique and coat. The focus group was comprised of 18 owners of healthy adult cats ≥1 year of age and of varying body condition, and five personnel from the The Iams Company, Pet Health and Nutrition Center, Lewisburg, OH, USA, who were caregivers to a population of healthy cats that were also ≥1 year of age and of varying body condition. For the purposes of the study, cats were defined as healthy if they were reported to be healthy by their owners, not receiving any medications to treat a disease and not being fed a veterinary diet. No physical examinations, diagnostic testing or other criteria were used to verify the cats’ health.

Veterinarians and researchers in the area of cat health and nutrition were also consulted and a literature review was conducted to further define domains relevant to measuring a cat’s HRQoL. In addition to defining domains, the language used by cat owners and caregivers in the focus groups to describe the domains was captured for generating items reflective of the descriptions used by cat owners.

From the focus groups, a total of 11 domains was identified initially and these were used to facilitate item generation to ensure that the items (ie, questions) adequately represented all domains. A draft instrument consisting of 155 items representing the 11 domains was created. A recall period of 7 days was used in which owners were asked to consider the cat’s behavior over the past 7 days. Participants used a 6-point Likert scale, where 1 = never and 6 = always, plus a ‘does not apply to me and my cat’ option to respond to each item. Also included were two general questions to determine owners’ perceptions of the overall HRQoL of their cats (5-point Likert scale from excellent to poor) and of the cats’ overall health status (5-point Likert scale from extremely healthy to not at all healthy). These two questions were used for validation purposes and to classify cats according to perceived overall HRQoL and health status, respectively.

Cognitive debriefing interviews

Cognitive debriefing interviews with the draft instrument were conducted to evaluate face and content validity, identify redundancies, refine items and reduce the total number of items. Specifically, the cognitive debriefing interviews were designed to evaluate the face validity (ie, the degree to which a measure is clearly and unambiguously encompassing the construct it is intended to measure) of the items and the instrument as a whole, and the content validity (ie, the extent to which the items in the instrument cover all the elements relevant to a cat’s overall health and wellbeing [ie, the cat’s HRQoL]).

These cognitive debriefing interviews were conducted with a novel group of 26 female cat owners between the ages of 18 and 65 years who owned healthy cats ≥1 year of age. ‘Healthy’ was defined as described on page 691. No physical examinations, diagnostic testing or other criteria were used to verify the cats’ health.

Owners who had previously agreed to take part in qualitative research were identified from a database and invited to participate in the cognitive debriefing interviews. Interviews of approximately 60 mins’ duration were conducted by two trained interviewers using a structured interview guide. Owners were asked to report the cat’s age and to evaluate the body condition of the cat by assigning a body condition score based on visual analysis of diagrams of cats with varying body conditions (very underweight, underweight, normal weight, overweight or very overweight). Each owner completed the draft instrument prior to being interviewed using a think-aloud approach ³⁶ commonly used to assess face and content validity.

The objectives for these cognitive debriefing interviews were to identify items that did not exhibit a common understanding; to ensure the instructions for completing the instrument and available response options were clear; to eliminate ambiguity, reduce redundancy and optimize length; and to identify missing items of importance to cat owners regarding their cat’s HRQoL.

After completing the questionnaire, each owner was asked for her overall opinion of the instrument and if any issues were experienced during completion. Further, they were questioned about their understanding of each item, and whether the wording of each item was optimal or needed to be revised. The appropriateness of the 7 day recall period was also discussed to determine if the time frame was too long or too short for owners to consider the concepts included in the instrument. The results were used to refine the draft instrument before its use in the quantitative validation study.

Phase II of cognitive debriefing, with a novel group of five female cat owners between the ages of 18 and 65 years, was undertaken to further clarify wording and identify additional items for deletion. The process to guide the discussion in phase II was similar to that used in phase I.

Online quantitative validation

After refinement of the instrument, an online quantitative development and validation study, involving 1303 participants, was conducted to develop the final instrument and to further evaluate reliability and validity. Owners of healthy adult cats ≥1 year of age and primarily living indoors (so they could be sufficiently observed) were eligible for the study. ‘Healthy’ was defined as described on page 691. No physical examinations, diagnostic testing or other criteria were used to verify the cats’ health.

Study participants were recruited by a commercial market research company from within its large database that included cat owners who had voluntarily agreed to be contacted regarding research studies (mean age = 49 ± 15 years; female [70%], male [30%]). Owners received ‘points’ for completing the study which could be redeemed against consumer goods. To qualify for the study, an owner had to own fewer than three cats and be the primary caregiver. If an owner owned more than one cat, only one cat was randomly selected by the computer to be the ‘subject’ for the completion of the instrument. Owners were asked to report the cat’s age and to classify the cat’s body condition (very underweight, underweight, normal weight, overweight or very overweight). The questionnaire was available for completion for a 7 day period.

A subpopulation of 400 owners was randomly selected by a computer randomization system to complete the questionnaire a second time 7 days after they first completed the questionnaire to determine test–retest reliability of the instrument; 391 owners agreed to do so.

Item reduction and domain identification analyses

The sample of 1303 cats from the quantitative validation phase was randomly divided into two sets. Analyses were conducted on approximately half of the cats surveyed (referred to as the ‘development’ dataset) to develop a final instrument. The remaining cats’ surveys (referred to as the ‘validation’ dataset) were used to confirm the reliability and validity of the final instrument. For assignment purposes, cats were assigned a random uniform number between 0 and 1. Cats with random numbers <0.5 were assigned to the development dataset and cats with random numbers ⩾0.5 were assigned to the validation dataset.

Initial item reduction on data from the development dataset was performed using standard psychometric analyses^37–40 to determine the most effective subset of items and to identify the underlying factors characterizing feline HRQoL. Items with missing data rates greater than 10% were candidates for elimination. Item frequency distributions were computed by overall health status to examine floor and ceiling effects; ie, the respective clustering of cats at the worst- or best-possible response categories. Items were reverse- transformed (as appropriate) so that response values were consistent across all items from low to high.

To examine floor effects, the proportion of cats with ‘extremely healthy’ or ‘very healthy’ overall health status, with owners who responded with the worst HRQoL value (ie, 1), was calculated. A floor effect in the ‘extremely healthy’ or ‘very healthy’ group reflects areas that are inconsistent with their perceived excellent health status. Items with ⩾10% floor effect were candidates for elimination. Analogously, to examine ceiling effect, the proportion of cats with ‘somewhat healthy’, ‘not very healthy’ or ‘not at all healthy’ overall health status, with owners who responded with the best-possible HRQoL, was calculated. A ceiling effect is undesirable as no possible improvement in score is possible even with improved underlying HRQoL. Items with ⩾50% ceiling effect were considered for elimination. Additionally, an analysis of variance was used to assess the overall ability of an item to differentiate cats with varying overall health status. Items that had relatively low ability (lowest 30% of items) to differentiate were considered for elimination. Clinical perspective was combined with a review of each item’s performance to identify which set of items would be eliminated and which set would be retained for the next phase of analyses.

Exploratory and confirmatory factor analysis

Exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) techniques, using commercial software, ⁴¹ were used to identify a proposed domain structure from the development dataset. The EFAs corresponding to the domains were explored and evaluated based on standard fit statistics (chi-square/df ratio, root mean square error of approximation [RMSEA], comparative fit index [CFI], Tucker–Lewis index and weighted root mean square residual [WRMR]). Since the chi-square goodness-of-fit statistic is a function of the sample size, it was augmented with the other incremental fit indices listed above in order to test the hypothesis of model fit in totality. Researchers investigated the performance of these indices and recommended a multi-index strategy with standard cut-off values. ⁴²

An 11 domain solution appeared to have good fit and most closely matched the original clinical interpretation and hence was used to assess item retention. Items with factor loadings >0.4 on more than one domain (ie, showing complex correlations across multiple domains) and items with factor loadings <0.4 on all domains (ie, having no meaningful correlations on any domain) were considered for removal. CFA models were then conducted to assess the overall fit. Modification indices and substantive knowledge were utilized to identify any further model refinement needed. ⁴³

It was noted that, in some instances, items leading to poor model fit had a different context as compared with the other items in that domain. For example, the item ‘was irritable’ has a negative context while all the other items in that domain (eg, ‘was friendly’ or ‘was relaxed’) have a positive context. Hence, models addressing a potential method effect due to positive and negative contextual wording were explored. In this additional exploration, factors of ‘positive’ and ‘negative’ were also included as part of the final model. Comparison of model fit as well as clinical interpretation led to retaining the model with method effects. The final instrument contained eight domains with a total of 33 items (ranging from three to seven items per domain).

Domain score computation

A domain score for each of the final eight domains was computed using Likert’s method of summated ratings, which equally weights each item and sums them into an overall domain score. Each overall domain score was linearly transformed to a 0–100 metric, with 100 indicating a favorable state (highest QoL), 0 indicating least favorable state (lowest QoL), and scores in between representing the percentage of total possible score achieved.

Confirmatory reliability and validity assessment

The psychometric properties of reliability and validity were assessed for the final instrument using the validation dataset. A CFA was conducted to confirm the factor structure of the final instrument. Average variance extracted was computed to assess discriminant validity. ⁴⁴ For each of the eight domains, classical known-groups validity was assessed by evaluating the ability of the domain score to differentiate cats by: 1) overall health status; 2) age; and 3) the owner’s perception of the cat’s overall QoL. For each of the eight domains, internal consistency reliability and test–retest reliability were assessed using Cronbach’s alpha and intraclass correlation coefficients. The intraclass correlation coefficients were computed for the subset of cats with test and retest responses based on variance components from a repeated measures analysis of variance model.

Results

Qualitative phase

Analysis of the qualitative data from focus group interviews revealed 11 potential domains that owners considered important for evaluating a cat’s HRQoL. The domains, and the language and expressions owners used to describe the visual and behavioral signs, provided the basis for item generation, leading to the development of 155 items in the draft questionnaire. Cognitive debriefing interviews (phase I) led to the elimination of 50 items because of poor wording, ambiguity or redundancy, leaving 105 items. However, a second round of interviews (phase II) with five cat owners was deemed necessary because it was unclear whether several additional items should be eliminated. Phase II interviews led to the elimination of five additional items.

The remaining 100 items demonstrated good face validity. Owners did not propose any new or additional items which indicated that the instrument demonstrated good content validity. In addition, owners stated that they did not have any difficulty completing the instrument related to the instructions or the response options. With respect to the recall period, owners felt that less than 7 days would be too short a period to observe some behaviors and that a period longer than 7 days was unnecessary.

Based on these debriefing interviews, items were refined and a revised instrument with 100 items comprising the 11 original domains (with each domain containing more than three items, as is normally required for CFA) was used in the online quantitative validation study.

Quantitative validation study

Demographic and baseline characteristics

A total of 1303 cat owners completed the quantitative validation study, distributed approximately evenly across cat age groups (Table 1). Of the 1303 owners, 391 completed the questionnaire a second time 7 days later to determine test–retest reliability of the questionnaire. Overall, cats were generally reported to be healthy, with 95% of owners responding that their cats had very good or excellent health. Cat demographics for the test–retest group were similar to those of the overall group (Table 1).

Table 1.

Demographics of cats (n = 1303) included in the quantitative validation study of the Cat HEalth and Wellbeing (CHEW) Questionnaire

		Overall		Test–retest dataset
Characteristic	Level	n	%	n	%
Age group	1–3 years	291	22	75	19
	4–7 years	421	32	115	29
	8–10 years	290	22	101	26
	>10 years	301	23	100	26
	Total	1303	100	391	100
Overall health	Not at all healthy	2	0	1	0
	Not very healthy	6	0	2	1
	Somewhat healthy	69	5	16	4
	Very healthy	536	41	180	46
	Extremely healthy	690	53	192	49
	Total	1303	100	391	100
Overall quality of life	Poor	1	0	1	0
	Fair	7	1	0	0
	Good	32	2	10	3
	Very good	392	30	126	32
	Excellent	871	67	254	65
	Total	1303	100	391	100
Body condition	Very underweight	101	8	26	7
	Underweight	389	30	118	31
	Normal weight	640	49	185	47
	Overweight	162	12	57	15
	Very overweight	11	1	5	1
	Total	1303	100	391	100

Of the original population of 1303 cats in the dataset, 648 were randomly assigned to the development dataset and 655 were randomly assigned to the validation dataset. The development and validation data subsets were similar to the original population in terms of their demographic characteristics (data not shown).

Item reduction: data quality and item distributions

Data quality was assessed by examining missing data rates, floor and ceiling effects, and ability to differentiate cats by overall health status. Thirty-eight items were eliminated for exhibiting several of these criteria or an extreme in any one of these criteria, and where there was no clinical rationale for retaining the item for the next phase of the analysis.

Domain identification

Identification of the final model, domains and items within a domain was based on EFA and CFA. Standard multidimensional factor structures and models allowing for method effects based on item contextual wording were examined along with clinical considerations, in general, and relative to an owner-reported outcome assessment, in particular. An EFA with Geomin rotation was conducted on the remaining 62 items with good data quality. Solutions corresponding to eight to 11 rotated factors were evaluated. Acceptable fit statistics were observed for each of these solutions. An 11 factor solution provided the best fit and most closely matched the original clinical conceptual model and hence was identified to take to the next stage (chi-square/df ratio = 1899/1264 = 1.50, RMSEA = 0.03, CFI = 0.99, Tucker-Lewis index = 0.98, WRMR = 0.03). Twenty-five items were identified for deletion due to poor loading on all factors or cross-loading on more than one factor. The resultant EFA indicated nine domains (Mobility, Play, Engagement, Emotion, Coat, Eyes, Energy, Fitness and Appetite) comprised of 37 items.

It was noted that two closely related domains, Play and Energy, were comprised solely of items of either positive contextual wording or negative contextual wording. For example, all the items related to Energy had a negative connotation (eg, ‘seemed tired’, ‘was sluggish’, ‘seemed lazy’) while all the items related to Play had a positive connotation (eg, ‘was frisky’, ‘was playful’, ‘wanted to play with people’). From the perspective of an ORO tool, it was questioned whether Play and Energy belonged to a single underlying construct or truly reflected two separate underlying constructs as indicated by the EFA. Additionally, two items, originally part of the Energy domain but with positive wording (‘lively’, ‘full of energy’), were identified for deletion due to cross-loading significantly with the Play domain. Hence CFA models were explored to allow for the presence of a wording method effect (ie, were multiple factors from the EFA arising not due to an underlying construct, but rather due to an artificial effect due to wording context?).

CFA models were investigated to assess the model fit of the resultant EFA model but with the Energy and Play domain combined (simplified to the single term, Energy). Forty-one items were included in this analysis; the 37 items identified for retention from the EFA model, the two additional items noted above that loaded on both the Energy and Play domains, as well as two additional items that were retained based on clinical consideration (‘coat or fur looked thick and full’, ’felt bony’). Two additional factors were included to encompass the effect due to ‘positive’ and ‘negative’ item wording. Modification indices and factor loadings were assessed to identify particular items with lack of fit (eg, ‘was happy’), as well as to enable shortening of the resultant questionnaire (eg, the item ‘was playful’ was considered very similar to the items ‘wanted to play with people’ and ‘played with toys’, and hence was the only one retained in the final model). A final instrument consisting of eight domains and 33 questions was identified and analyses to confirm the reliability and validity were conducted utilizing the independent ‘validation’ subset of the data.

Confirmatory reliability and validity assessment

Construct validity

A CFA was conducted to assess construct validity (the appropriateness of the underlying model; Table 2). The final model was shown to have good fit (chi-square/df ratio = 1037/434 [2.4], RMSEA = 0.046 [95% confidence interval [CI]: 0.042–0.050], CFI = 0.99, Tucker-Lewis index = 0.98, WRMR = 0.91).

Table 2.

Confirmatory validation parameter estimates^* for the Cat HEalth and Wellbeing (CHEW) Questionnaire

	Item	Domain factor	Positive effect factor	Negative effect factor
Mobility	Got up slowly when he/she was lying down	0.70	–	0.44
	Had difficulty getting up when he/she was lying down	0.89	–	0.29
	Movements were slow	0.79	–	0.44
	Movements were stiff	0.84	–	0.28
Emotion	Was affectionate	0.66	0.61	–
	Was friendly	0.80	0.48	–
	Was irritable	0.59	–	0.10
	Was approachable	0.69	0.45	–
	Was relaxed	0.78	0.32	–
Energy	Was playful	0.81	0.32	–
	Was frisky	0.74	0.43	–
	Was lively	0.91	0.33	–
	Was full of energy	0.93	0.29	–
	Was sluggish	0.80	–	0.39
	Did not have the energy to play	0.75	–	0.47
	Seemed tired	0.72	–	0.48
Engagement	Greeted me when I returned from being away †	0.50	–	0.61
	Was curious about his/her surroundings	0.79	–	0.34
	Observed everything going on around him or her	0.87	–	0.28
Eyes	Eyes looked alert	0.88	0.19	–
	Made eye contact with me	0.80	0.42	–
	Eyes looked bright	0.96	0.24	–
	Eyes looked clear	0.95	0.18	–
Coat	Coat or fur looked shiny	0.87	0.09	–
	Coat or fur felt soft	0.88	0.17	–
	Coat or fur looked dull	0.72	–	0.32
	Coat or fur looked thick and full	0.78	0.14	–
Appetite	Had a good appetite	0.91	0.17	–
	Enjoyed his/her food	0.94	–0.02	–
	Seemed satisfied after eating	0.85	–0.03	–
Fitness	Felt bony	0.66	–	0.18
	Body felt muscular	0.82	0.01	–
	Looked athletic or fit	0.94	–0.02	–

^*Eight factor model with additional factors for positive and negative wording effect

^†Retained in the CHEW Questionnaire despite a high method effect

Convergent validity

The majority of items within the eight domains had high loadings (⩾0.7) and exhibited significantly greater loading on the conceptual domain vs method effect, thus supporting convergent validity for the domains being measured. There were a few cases of relatively low loading or high method effect. However, these items were retained due to clinical considerations and a desire to have a sufficient number of items (⩾3) for each domain.

Clinical known-groups validity

The instrument was evaluated for its ability to discriminate between cats known to differ by age, overall health status and overall QoL. For example, it would be expected that cats that are older, that have poor overall health status or poor overall QoL would have lower scores for the respective domains. As age increased, a decrease in mean domain score was observed, indicating a decreased HRQoL for each domain (Figure 2a). As overall health (Figure 2b) and overall QoL (Figure 2c) increased, an increase was observed in mean domain scores, indicating greater HRQoL for each domain. For all domains and comparisons, a statistically significant linear trend was observed (all P <0.0002).

Figure 2.

Discriminant ability of the Cat HEalth and Wellbeing (CHEW) Questionnaire by domain. Statistically significant linear trends (P <0.0002) were observed for all domains across categories of (a) cat age, (b) overall health status and (c) overall quality of life (QoL). Data are expressed as mean ± SEM. For age, cats are divided into 1–3 years (black bars), 4–7 years (hatched bars), 8–10 years (grey bars) and >10 years (white bars). For simplicity of the figure showing overall health status (b), cats classified as ‘not at all healthy’ and ‘not very healthy’ were combined into a single group (black bars) and cats classified as ‘very healthy’ and ‘extremely healthy’ were combined into a single group (white bars), with cats classified as ‘somewhat healthy’ shown in hatched bars, although they were analyzed as five separate groups. Similarly, for overall QoL (c), ‘poor’ and ‘fair’ were combined for simplicity (black bars), as were ‘very good’ and ‘excellent’ (grey bars); cats classified as having ‘good’ QoL are represented by the hatched bars

Scale reliability

Internal consistency reliability coefficients (Cronbach’s alpha and composite reliability) were computed to estimate scale reliability at the domain level (Table 3). For all eight domains in the CHEW Questionnaire, both of these reliability estimates exceeded the usual threshold value of 0.7, indicating good internal consistency. Test–retest reliability estimates were conducted for 391 cats (Table 3). Acceptable intraclass correlation coefficients (⩾0.7) were observed for most domains, indicating consistency of responses across the two administrations of the instrument; slightly lower test–retest reliability was seen for the domains, Eyes and Appetite.

Table 3.

Scale reliability estimates from confirmatory and validity assessment of the Cat HEalth and Wellbeing (CHEW) Questionnaire using the validation dataset (n=655)

Reliability indices	Mobility	Emotion	Energy	Engagement	Eyes	Coat	Appetite	Fitness
Cronbach’s alpha (n = 655)	0.84	0.87	0.91	0.79	0.91	0.79	0.86	0.70
Composite reliability (n = 655)	0.86	0.87	0.91	0.81	0.92	0.80	0.87	0.74
Test–retest reliability (n = 391)	0.70	0.79	0.84	0.69	0.64	0.78	0.56	0.84

Discussion

Similar methods to those described in the 2009 FDA guidance on outcome measures were used to develop and validate the CHEW Questionnaire for HRQoL in cats. ³⁵ This instrument was developed using owner- generated concepts and language and is an owner-reported, multidimensional assessment consisting of eight domains and 33 items that can be completed online. The physical, mental, emotional and social dimensions of health that are recognized as important for assessing human HRQoL (as opposed to the even more complex concept encompassed by QoL) via general HRQoL instruments, such as the SF-36, were also identified by owners/caregivers as important for determining the overall HRQoL of cats and shown to be reliable and valid measures in the CHEW Questionnaire. These dimensions as assessed in the CHEW are: physical (Mobility, Eyes, Coat, Fitness and Appetite domains), mental and emotional (Emotions and Energy domains), and social functioning (Engagement domain).

Many of these domains were similar to those identified in a QoL questionnaire in cats with and without degenerative joint disease, ²⁰ which included active transition, eating/ drinking, grooming/scratching, playing/ hunting, rest/observe and social. That study reported that activities classified as being inactive were more important for a cat’s QoL than active activities, thus emphasizing the importance of assessing different domains impacting an owner’s assessment of a cat’s QoL. ²⁰ The current study included items within the domains that encompassed both active and inactive activities.

In terms of demographics, the cats owned by study participants represented a broad distribution of ages between 1 to >10 years and, in general, owners perceived the health and HRQoL of their cat to be very good or excellent. The overall high HRQoL and health status of the cats included within this study were expected as the questionnaire was developed in a population of apparently healthy cats (ie, cats reported by their owners to be healthy and not receiving medical treatment or being fed a veterinary diet).

Using the target audience to identify items and domains, such as in the Benito et al study, ²⁰ is important. Including cat owners in the initial qualitative phase of the current study ensured that the perspective and language of the target audience served as the foundation for the item generation and domain identification in developing the instrument. Following construction of the draft instrument, cognitive debriefing with additional cat owners helped eliminate ambiguity in the interpretation of each item, initiated the item reduction needed to reduce the burden placed on respondents, and refined the wording, resulting in a draft instrument with 100 items for larger scale, online quantitative testing and validation. During the quantitative validation phase, items were removed due to high missing data rates, high floor and ceiling effects, and poor ability to differentiate cats across owner-reported health states. This was followed by EFA and CFA, which were used to finalize domains and the ‘best set’ of items for each domain, resulting in a final instrument containing eight domains and 33 items.

Discriminant (clinical known-groups) validity was assessed through differences in response between groups with known variation; ie, groups differing in age, HRQoL and health status. For example, there were age-related declines in Mobility domain scores, which support previous findings of increasing prevalence of mobility impairment with age.^45–47 Similar patterns were observed within the other domains (ie, decreasing HRQoL with increasing age and increasing HRQoL with increasing overall health and overall QoL). This supports good discriminant validity for the CHEW Questionnaire.

The domains and items of the final CHEW Questionnaire appear to be clinically relevant and encompass the primary components of HRQoL for humans and animals: the more subjective assessment of affective state and a more objective assessment of specific external parameters indicative of physical/functional health.^48,49 The advantage of including items assessing the affective state, which encourage owners to judge an animal’s experiences or subjective feelings through behavioral observation ⁵⁰ (eg, ‘my cat seemed tired’ or ‘my cat was relaxed’), is that the affective state of an animal may significantly influence its HRQoL and owners perceive that this is important for determining their cat’s overall HRQoL. ⁵¹ However, assessments of affective states are open to individual owner interpretation and are measured indirectly through specific behavioral patterns yet to be sufficiently defined.^48,50 By using an ORO, one assumes that owners are able to observe, understand and interpret their cat’s behaviors, which may or may not be true. By contrast, the questions of physical health, such as ‘my cat’s movements were stiff’ and ‘my cat’s body felt bony’, are more established indices of health and HRQoL, can be readily and directly observed by the owner and are less subjective. Together, an HRQoL instrument that encompasses both mental (affective) and physical parameters offers the opportunity to evaluate an outcome multifactorially in a reliable and robust manner.

There are additional limitations to the current study which are important to address. By their nature, content and face validity rely on qualitative research and cannot be quantified. Thus, it is always possible that concepts or items that are important for the HRQoL of cats were missed. Assessment of QoL in companion animals involves OROs and an owner’s ability (or inability) to observe cats and their behavior has important effects on the results. Also, since one cannot communicate directly with the animal, concepts deemed important to an owner may or may not be as important to the cat. In humans, OROs are discouraged, with PROs being preferred, even in children. ³⁵ However, there are situations where OROs are necessary for humans and they are certainly required to be able to assess HRQoL in companion animals. In children, when the use of OROs is necessary, further issues must be addressed, such as the use of proxy (where the observer must make inferences about the child’s experience) and observation (which is based on direct observation and therefore preferred). ⁵² Clearly, veterinary clinicians and investigators are even more limited in assessing HRQoL in companion animals, but consideration of the many issues involved in the assessment of HRQoL through OROs is critical to having valid instruments and results.

Cats in this study were defined as ‘healthy’ if they were reported by their owners to be healthy, and not receiving medications or being fed a veterinary diet, rather than being assessed clinically. Results showed that, of the 1303 participants, a small number of cats (n = 8; 0.6% of the total) were categorized by the owner as being ‘not very healthy’ or ‘not at all healthy’, despite the inclusion criterion of the study that cats should be healthy. Although this has the potential to affect the results, the small number makes this unlikely. Thus, from an owner perspective, using this definition of ‘healthy’ appears to work well in defining a ‘real world’ owner-perceived healthy population. However, it is important to note that no physical examinations, diagnostic testing or other criteria were used to verify the cats’ health. Both this limitation and the low number of cats that were reported to be unhealthy limits full assessment of the instrument. Evaluation both in cats that are more definitively verified to be healthy and in a less healthy population of cats is needed.

Intraclass correlation coefficients for the test–retest validation suggested that the results were consistent (>0.7) for most domains. The exceptions were the Eyes and Appetite domains, in which lower values were observed. This may have been due to the small number of items in these domains; or these may be domains that are more subjective in general. Although participants responded that 7 days was appropriate for the retest period, this may not have been long enough in apparently healthy cats. Additional research would help to optimize timing of the retest period.

Further studies are needed to explore the reliability, validity and discrimination ability of the CHEW Questionnaire in other populations of cats and to determine whether the instrument is able to differentiate cats with and without disease. Finally, further item reduction could lead to the development of a shorter (five to 10 item) inventory for use by veterinarians as a screening tool for early disease or other issues of concern to cat owners.

Key points

• A generic HRQoL instrument, such as the CHEW Questionnaire, may have a role in veterinary research and clinical practice in the future, as it does in human clinical practice, to screen for medical issues, monitor progress over time and facilitate patient-centered care.^53,54

• Although much additional research is needed, validated instruments have the potential to provide a tool that veterinarians and owners can use to monitor the ongoing HRQoL of cats – for early diagnosis of disease or to identify a decline in health status, progression of disease, or a change in overall health status with interventions, such as diet, medication or surgery.