Methods

An initial draft of the CDI was created by a veterinary internist (KEC), a veterinary epidemiologist (AR), and a human disease epidemiologist (AF). The instrument was modeled after current human multimorbidity indices^32–34 and adapted for use in canine patients. Human literature on comorbidity and multimorbidity indices was reviewed to identify disease domains applicable to dogs. Common chronic canine ailments that often lack a definitive diagnosis (eg, chronic gastrointestinal disease) were also identified. These chronic diagnoses, certain of which contained subcategories (eg, “hemangiosarcoma” under the category of “cancer”), were then incorporated into the index. The CDI was designed to be used by veterinarians to score the multimorbidities of individual dogs through review of the dog’s medical records without necessarily having to evaluate the dog firsthand in a clinical setting.

The current study was completed in 2 phases. During the initial phase, primary care veterinarians were recruited via convenience sampling throughout the US and were asked to participate in a pilot study to evaluate the first iteration of the CDI described above from January 11, 2021, through February 15, 2021. Participants were asked to use the CDI to review veterinary medical records from within their own practices. Records suitable for use in the pilot study were chosen by veterinary technicians, veterinary practice managers, or investigators who were not participating in the pilot study but who had legal access to the records at each participating site. In order to be eligible for inclusion, the record had to be contained within a veterinary electronic medical record system and had to be from a primary care service (eg, private general practice or primary care service at a veterinary teaching hospital). The records were further screened to include only dogs that were at least 8 years old and had a history of at least 3 separate visits during which the animal was examined by a veterinarian. Additionally, the most recent documented visit must have occurred within 2 months of the start of the pilot study.

The CDI was accessed online through a commercial survey instrument platform (SurveyMonkey). Participants were able to access the tool and complete it independently within their clinical setting. The tool presented a structured series of diagnostic categories covering common chronic conditions in dogs. For each condition, participants were prompted to classify whether the diagnosis was present or not. Participating veterinarians were asked to review records and complete the CDI a total of 3 times at 1-week intervals. Each participant completed the CDI for the same record twice, in the first and third weeks, respectively. In the second week, each participant completed the CDI for a different record, which had been scored by another participant during the first week. At the conclusion of the pilot study, participants were queried via email for any positive or negative feedback regarding the CDI.

Participant feedback was coupled with input from 2 veterinary general practitioners who had not participated in the study in order to revise the data collection instrument. Changes to the instrument included the addition of definitions for each indexed disease to enhance the clarity and uniformity of diagnostic determinations. Response choices were broadened to allow participants to indicate whether each disease recorded in the CDI was a current or past diagnosis or had never been diagnosed or if the diagnosis could not be determined from the medical record. Questions accompanying each diagnosis also included limited relevant diagnostic and treatment inquiries (eg, did the reportedly Addisonian dog have an ACTH stimulation test performed? Was the reportedly diabetic dog receiving insulin?) to improve confirmation that the disease of interest had been definitively diagnosed. Last, modifications to the layout of the instrument were made to streamline the participant’s experience, and the CDI was migrated to a different commercially available online survey platform (Qualtrics).

During the second phase of this study, conducted from May 3, 2021, through June 7, 2021, a new group of nonspecialist veterinarians were selected by convenience sampling to apply the revised CDI instrument to multimorbidities in medical records different from those used in the first pilot but using the same criteria as described above. At the conclusion of the study, the new participants were also queried via email for any positive or negative feedback regarding the revised CDI via the same survey instrument.

Results

In the initial phase of the study, 16 nonspecialist veterinarians (9 primary care veterinarians in private practice and 7 veterinary small-animal rotating interns) participated in the use of the CDI instrument to score morbidities. Intra- and inter-rater agreements for the first pilot can be found in Table 1. Canine chronic bronchitis (κ = −0.05), canine cognitive dysfunction (κ = −0.07), and congestive heart failure (κ = −0.05) showed no inter-rater agreement according to the aforementioned scale. Chronic gastrointestinal disease (κ = 0.14; P = .40), osteoarthritis (κ = 0.20; P = .13), and periodontal disease (κ = 0.18; P = .21) were found to have slight inter-rater agreement. Chronic allergic/inflammatory condition (κ = 0.36; P = .06), malignant neoplasia (κ = 0.25; P = .09), and hypothyroidism (κ = 0.33; P = .09) had fair agreement, and each showed a trend toward significance. Morbidities with moderate agreement and statistically significant correlations included diabetes mellitus (κ = 0.47; P = .001), hyperadrenocorticism (κ = 0.50; P = .005), obesity (κ = 0.51; P = .002), and recurrent urinary tract infection (κ = 0.42; P = .01). Chronic kidney disease (κ = 0.84; P = .001) and seizure condition (κ = 1.00; P = .001) had very high/perfect agreements that were statistically significant.

Intrarater reliability in the initial study showed overall greater agreements than inter-rater assessments. Malignant neoplasia (κ = 0.59; P = .001) and hypothyroidism (κ = 0.54; P = .03) had moderate agreement. Morbidities found to have substantial agreement included chronic allergic/inflammatory conditions (κ = 0.72; P = .001), canine chronic bronchitis (κ = 0.64; P = .006), diabetes mellitus (κ = 0.64; P = .006), chronic gastrointestinal disease (κ = 0.62; P = .001), obesity (κ = 0.74; P = .001), and periodontal disease (κ = 0.78; P = .001). Chronic kidney disease (κ = 0.86; P = .001), canine cognitive dysfunction (κ = 1.00; P = .001), congestive heart failure (κ = 1.00; P = .001), hyperadrenocorticism (κ = 0.86; P = .001), osteoarthritis (κ = 0.81; P = .001), seizure condition (κ = 1.00; P = .001), and recurrent urinary tract infection (κ = 0.83; P = .001) had almost perfect intrarater agreement. Hypoadrenocorticism was not assessed in the initial study.

The second pilot, which utilized the revised CDI, included 18 new nonspecialist veterinarians (8 primary care veterinarians in private practice, 6 primary care veterinarians in academic practice, and 4 veterinary small-animal rotating interns) as participants. Inter- and intrarater agreements were also calculated for the second study (Table 1). Periodontal disease was the only condition with inter-rater agreement worse than chance agreement (κ = −0.24; P = .31). Chronic allergic/inflammatory condition (κ = 0.67; P = .004), chronic kidney disease (κ = 0.61; P = .005), chronic gastrointestinal disease (κ = 0.77; P = .001), and obesity (κ = 0.77; P = .001) were found to have statistically significantly better agreement than would be expected by chance. The remainder showed very high or perfect agreement: malignant neoplasia (κ = 0.82; P = .001), canine chronic bronchitis (κ = 1.00; P = .001), canine cognitive dysfunction (κ = 1.00; P = .001), congestive heart failure (κ = 1.00; P = .001), diabetes mellitus (κ = 1.00; P = .001), hyperadrenocorticism (κ = 1.00; P = .001), hypoadrenocorticism (κ = 1.00; P = .001), hypothyroidism (κ = 1.00; P = .001), osteoarthritis (κ = 1.00; P = .001), seizure condition (κ = 1.00; P = .001), and recurrent urinary tract infection (κ = 0.87; P = .001).

The analysis of data from the second pilot also showed that intrarater concordance was higher than would be expected by chance (κ and P values). Canine chronic bronchitis (κ = 0.64; P = .004), chronic kidney disease (κ = 0.77; P = .001), chronic gastrointestinal disease (κ = 0.61; P = .005), hyperadrenocorticism (κ = 0.64; P = .004), obesity (κ = 0.68; P = .002), and recurrent urinary tract infection (κ = 0.75; P = .001) had substantial agreement. Chronic allergic/inflammatory condition (κ = 1.00; P = .001), malignant neoplasia (κ = 0.82; P = .001), canine cognitive dysfunction (κ = 1.00; P = .001), congestive heart failure (κ = 1.00; P = .001), diabetes mellitus (κ = 1.00; P = .001), hypoadrenocorticism (κ = 1.00; P = .001), hypothyroidism (κ = 1.00; P = .001), osteoarthritis (κ = 1.00; P = .001), periodontal disease (κ = 0.85; P = .001), and seizure condition (κ = 1.00; P = .001) all had almost perfect agreement.

Discussion

We report on the construction and validation of a CDI for canine patients, the first veterinary-focused instrument designed to standardize the study of multimorbidity in companion dog populations. This tool’s development capitalized on the available human multimorbidity literature and the expertise of several veterinary and epidemiological specialists to develop a list of clinically relevant canine morbidities. Through the execution of 2 pilot studies, the instrument was edited and refined to ensure that veterinarians from various backgrounds could accurately and repeatedly extract the same multimorbidity data from veterinary electronic medical records. Though developed for use in a specific cohort of dogs, veterinarians will be able to use this tool more broadly to capture and standardize the discussion and analysis of canine multimorbidity, which will lead to a better understanding of the impact of multimorbidity on the health of companion dogs.

Expanding the existing knowledge of multimorbidity is particularly pertinent in veterinary medicine given the increasing prevalence of senior dogs and the link between age and multimorbidity. In the US, older dogs account for a substantial percentage of the companion dog population.¹⁹ In 2022, 1 survey estimated that senior dog–owning households accounted for 52% of dog-owning homes, whereas households owning puppies less than 1 year old dropped from 13% to 9% in the same period.³⁵ While types of morbidities and general breed dispositions are relevant to disease development,^19–20 it has been suggested that body size and breed have little effect on the overall number of morbidities likely to develop, whereas age was found to be a major risk factor associated with disease accumulation.¹⁷

Human multimorbidity indices have improved the ability to predict mortality, hospital admissions, healthcare costs, and future quality of life. In 1 study⁹ exploring accuracy in assessing human patient prognosis, a multimorbidity index was found to be statistically more accurate than many then-existing prognostic measures and demonstrated overall greater accuracy determining measures of illness severity. Given the extensive similarities between companion dogs and humans, the anticipation is that similarly useful tools can be created for canine patients. The CDI reported here is the first step toward this aim that we hope will serve as the foundation of a canine multimorbidity index.

The implementation of a canine multimorbidity index in veterinary medicine would meet the growing need to provide more accurate and timely predictions for canine patients with numerous chronic morbidities. With such information, veterinary practitioners would be able to better predict patient prognosis, gauge future quality of life, assess treatment efficacy, and anticipate costs of veterinary care. This would empower veterinarians to approach new diagnoses with not only short-term diagnostic and interventional goals but also long-term treatment and monitoring priorities that could lessen the impact of the multimorbid conditions and ultimately prolong the quantity and quality of life. This would also enable veterinarians to have thoroughly informed discussions with dog owners to set realistic prognostic and financial expectations when new diagnoses are made. Furthermore, since companion dogs share their human owners’ physical and social environments and succumb to comparable diseases, dogs have increasingly been utilized as ideal research models for aging and disease progression in humans.^{16–18,29,30,36} As such, the additional expectation is that canine multimorbidity indices could be further molded into a translational tool for investigating similar outcomes in humans.

While multimorbidity indices are valuable tools, limitations do exist. All currently available multimorbidity indices presume that an accurate diagnosis has been made and that an accurate measure of prognosis exists. In veterinary medicine, a definitive diagnosis is not made in all cases for diverse reasons, including limitations in access to care; once made, not all veterinary diagnoses are accompanied by a strongly evidence-based prognosis. Additionally, many multimorbidity indices utilized in human research consist of thousands of morbidities, making their use in clinical settings difficult.⁹ A lack of consensus regarding terminology and diagnostic requirements for many diseases makes the use of such an exhaustive list of diagnoses impractical in veterinary medicine. No single vocabulary standard exists for disease names in veterinary medicine, with some catch-all terms being unavoidable.³⁷ This challenge was noted in our first pilot study, in which participants offered feedback that there was confusion regarding CDI component requirements. The result of the confusion was poor inter-rater agreement. Definitions of each morbidity within the instrument were provided in the revised version of the CDI, leading to marked improvement in agreement between veterinarians’ diagnoses in the second pilot. This highlights the need for a clear, shared disease description for each morbidity included in this, or any other, multimorbidity tool intended for veterinary use.

One of the challenges with a novel research topic is the diversity of approaches by investigators, leading to difficulty in comparing study results. There has been considerable variation in the use of several foundational terms and concepts related to multimorbidity, such as the distinction between comorbidity and multimorbidity and the definition of a disease as “chronic.”¹⁰ By thoroughly detailing the development and contents of this CDI, we hope to enable future investigators to build upon the foundation we have created and to also facilitate the comparison of our approach with others’.

Understanding canine multimorbidity is important because of its relevance to the aging companion dog population, its ability to begin demystifying the links between aging and disease, and its potential to provide predictive outcome measures that mirror those produced by currently utilized human multimorbidity indices. In this first step toward harnessing the predictive power of multimorbidity for veterinary patients, the created and validated CDI discussed here produced a list of clinically relevant canine morbidities, standardized the content to produce excellent intra- and inter-rater agreement, and set the stage for the creation of a true canine multimorbidity index. In the future, utilization of a multimorbidity index in canine medicine will not only enable veterinary practitioners to accurately predict clinically valuable outcomes for their patients but will also provide insightful translational results that will equally benefit human patients.

Disclosures

The authors have nothing to disclose. No AI-assisted technologies were used in the composition of this manuscript.

Funding

This research is based on publicly available data collected by the Dog Aging Project, which was supported by U19 grant AG057377 from the National Institute on Aging, a part of the NIH, and by additional grants and private donations. These data are housed on the Terra platform at the Broad Institute of the Massachusetts Institute of Technology and Harvard. This manuscript is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

ORCID

A. Ruple https://orcid.org/0000-0002-5223-0217