Construction of diagnostic prediction model for canine nasal diseases using less invasive examinations without anesthesia

Abstract

Advanced imaging techniques under general anesthesia are frequently employed to achieve a definitive diagnosis of canine nasal diseases. However, these examinations may not be performed immediately in all cases. This study aimed to construct prediction models for canine nasal diseases using less-invasive examinations such as clinical signs and radiography. Dogs diagnosed with nasal disease between 2010 and 2020 were retrospectively investigated to construct a prediction model (Group M; GM), and dogs diagnosed between 2020 and 2021 were prospectively investigated to validate the efficacy (Group V; GV). Prediction models were created using two methods: manual (Model 1) and LASSO logistic regression analysis (Model 2). In total, 103 and 86 dogs were included in GM and GV, respectively. In Model 1, the sensitivity and specificity of neoplasia (NP) and sino-nasal aspergillosis (SNA) were 0.88 and 0.81 in GM and 0.92 and 0.78 in GV, respectively. Those of non-infectious rhinitis (NIR) and rhinitis secondary to dental disease (DD) were 0.78 and 0.88 in GM and 0.64 and 0.80 in GV, respectively. In Model 2, the sensitivity and specificity of NP and SNA were 0.93 and 1 in GM and 0.93 and 0.75 in GV, respectively. Those of NIR and DD were 0.96 and 0.89 in GM and 0.80 and 0.79 in GV, respectively. This study suggest that it is possible to create a prediction model using less-invasive examinations. Utilizing these predictive models may lead to appropriate general anesthesia examinations and treatment referrals.

Dogs can develop various diseases in the nasal cavity. Among these, neoplasia (NP), sino-nasal aspergillosis (SNA), and non-infectious rhinitis (NIR) (such as lymphoplasmacytic), have been reported as commonly occurring conditions [21, 23, 35]. Furthermore, secondary rhinitis is caused by foreign body (FB) and dental diseases (DD) [21, 23, 35]. Previous studies have reported that NP and NIR are the most common nasal diseases, followed by SNA and FB [21, 27]. Canine nasal disease is diagnosed based on signalment and clinical signs (such as nasal discharge and epistaxis), followed by physical examination (such as the presence or absence of nasal airflow) and radiography without general anesthesia. Although studies comparing clinical signs and radiographical findings in nasal diseases have been reported, diagnosis based on these examinations is difficult because various nasal diseases show similar clinical signs and radiographic findings [15, 21, 29, 30, 35]. Therefore, in many cases, advanced imaging tests, such as computed tomography (CT), magnetic resonance imaging (MRI), and endoscopy, are performed under general anesthesia. In previous reports, CT, MRI, endoscopy, and rhinoscopy have been more useful than radiography for the diagnosis of canine nasal diseases [7, 9, 15, 18, 28, 33].

Thus, the differentiation and definitive diagnosis of nasal diseases often require advanced imaging tests under general anesthesia. However, immediate examination under general anesthesia is not always feasible in every case due to various reasons, including limited availability of testing facilities, owner’s concerns about anesthesia risks, and financial constraints. In particular, NP is a progressive disease that requires prompt diagnosis because the stage and presence or absence of metastasis at the time of diagnosis and the histopathological tape of NP affect the prognosis and treatment [1, 4, 14, 16, 36]. In addition, the definitive diagnosis of NP is based on histopathological examination, which must be performed under general anesthesia. Sino-nasal aspergillosis is a chronic disease; however, as the disease progresses, it infiltrates the brain, which can be life-threatening. Therefore, it is necessary to make an accurate and appropriate presentation. In addition, a definitive diagnosis of SNA requires histopathological examination, similar to that of NP; the major treatment method is nasal immersion using antifungal drugs, which must be performed under general anesthesia. Rhinitis secondary to a foreign body is generally an acute disease, and clinical signs do not improve as long as a foreign body is present. Furthermore, endoscopic examination under general anesthesia is required for diagnosis and treatment by removing foreign bodies. Therefore, these diseases require early examination under general anesthesia. In contrast, while NIR and DD require examination under general anesthesia for a definitive diagnosis, there are many cases in which provisional and therapeutic diagnoses using anti-inflammatory and antibacterial drugs are performed.

In veterinary clinical practice, there are many cases of suspected of nasal disease; however, the differentiation of the disease is often difficult because of similar clinical signs and radiographic findings, as described above. In addition, because of the complexity of the canine nasal cavity and overlying bony structures, radiography does not reliably provide the detailed information needed to identify the exact cause of chronic nasal disease compared to advanced imaging examinations, such as CT and MRI [5, 28, 32,33,34,35]. Therefore, currently, some imaging tests are performed without radiography or when no differential disease is present. In addition, the necessity of anesthesia examination becomes unclear because of the unpredictability of any nasal disease, and the owner may not want it. Therefore, it is necessary for general veterinarians, not specialists, to accurately determine differential diseases from clinical signs and radiography and to present the need for examinations and treatments under general anesthesia to owners. This study aimed to construct a diagnostic prediction model for canine nasal diseases using less invasive examinations that do not require anesthesia, such as signalment, clinical signs and duration, and radiographic findings.

MATERIALS AND METHODS

Study design

In this study, we retrospectively investigated patients with suspected nasal diseases who visited the Department of Respiratory Medicine, Veterinary Medical Teaching Hospital at Nippon Veterinary and Life Science University (VMTH-NVLU) between February 1, 2010, and June 31, 2020, to construct a diagnostic prediction model (Group of model; GM). Furthermore, we prospectively investigated patients who visited the VMTH-NVLU and DVMs Animal Medical center Yokohama between July 1, 2020, and December 31, 2022, to validate the usefulness of the prediction model (Group of validation; GV). Among these cases, we included only cases in which one or more clinical signs such as sneezing, nasal discharge, and epistaxis were observed at the first visit, radiography was performed, and advanced imaging techniques such as CT, MRI, and endoscopy were performed and confirmed definitive diagnosis. Additionally, dogs suspected of having nasal symptoms due to causes other than nasal diseases such as epistaxis associated with thrombocytopenia and hypertension (based on preliminary examinations), were excluded. For NP, only cases with a definite diagnosis confirmed by the presence of intranasal soft tissue on CT or MRI [7, 18, 28], as well as histopathological examination through tissue biopsy were included. For SNA, only cases with a definitive diagnosis were included, which was confirmed by the presence of intranasal cavitations and non-contrast-enhanced fungal plaques on CT and MRI [18, 33, 34], intranasal cavitations, plaques, and necrotic material observed during endoscopy, and histopathological examination using tissue biopsy. In NIR, only cases meeting the following criteria were included: accumulation of nasal discharge without clear evidence of clear bone destruction or mass on CT and MRI, accumulation of nasal discharge and edema, redness, and irregularity of nasal mucosa in cases where an endoscopic examination was possible, and absence of evident mucosal bacterial infection confirmed by histopathological examination or bacterial culture [9, 15, 18, 20, 38]. For DD, only cases in which nasal discharge around the roots and root osteolysis were confirmed on CT were included [35]. Rhinitis secondary to a foreign body included only cases with a history (acute nasal symptoms and clinical signs such as vomiting, regurgitation, and tingling) and confirmed by CT, MRI, and endoscopy and extracted by endoscopy or nasal flashing [6, 17, 24].

Data collection

The investigated items included signalments, clinical signs, duration, and radiographical findings. Signalments included breed, sex and age, and ages were classified into the following four categories; 0–1 years, 2–5 years, 6–11 years, and >12 years. Palate ptosis, sneezing, reverse sneezing, presence or absence of nasal discharge (unilateral or bilateral), characteristics of nasal discharge (serous, purulent, or hemorrhagic), epistaxis (unilateral or bilateral), starter, facial deformity, and snoring were included as clinical signs. For clinical signs, only symptoms that were present at the first visit were evaluated through questionnaires and history taking. Therefore, the clinical signs observed during the period in which the chief complaint of the visits was observed were included. Additionally, even if the symptoms were unilateral but changed to bilateral at the time of the visit, the patients were evaluated as having bilateral disease. The duration of clinical signs was defined as the period from the onset of symptoms to the time of the first visit and was classified into the following five categories: <1 months, 1–3 months, 3–6 months, 6–12 months, and >1 year. Radiographic images were blindly evaluated by two veterinarians (YN and AF). Radiographic images were evaluated in two directions, the dorsal–ventral and lateral views, but there were a few cases in which only one direction could be evaluated due to a lack of imaging records or imaging technique errors, such as rolling. Regarding the radiation dose used in this study, the tube voltage ranged from 48–72 kV, the tube current was 200 mA, and the mAs value ranged from 3.2–5.0 mAs depending on the body weight. The investigated items of radiographic images included increased opacity in the nasal cavity (unilateral or bilateral), laterality, osteoclasia (destruction of the turbinate, destruction or irregularity of the cribriform plate, destruction or displacement of the nasal septum, and destruction of other bones, including the nasal bone, maxilla, and frontal bone), increased opacity of the frontal sinus, and decreased opacity of the periodontal membrane. Laterality indicates a difference in the strength of the increased opacity in the left and right nasal cavities. Therefore, when a unilateral increased in opacity was observed or when a difference in the strength of opacity was observed in the left and right nasal cavities, even when an increased opacity was observed bilaterally, it was evaluated as positive. The presence of osteoclasia was considered positive if destruction was observed in one or more of the aforementioned bones.

Diagnostic prediction model

In this study, diagnostic prediction models were constructed using the GM results of the two methods. The first method (Model 1) involved manually combining items that were characteristically recognized for each nasal disease. With this method, it is difficult to evaluate radiographic findings of the nasal cavity in brachycephalic breeds, such as French bulldogs and Chihuahuas [28]; therefore, a prediction model was constructed separately for meso- and dolichocephalic breeds. Additionally, two predictive models were created for neoplasia and rhinitis, including NIR, DD, and FB, owing to the small number of brachycephalic breeds. The second method (Model 2) used least absolute shrinkage and selection operator (LASSO) logistic regression analysis to generate automatic variable selection. Table 1 presents the explanatory variables used in logistic regression analysis. Each prediction model was validated using GV.

Table 1. Variables used for diagnostic prediction model.

	Items
Breed	Brachycephalic
	Mesocephalic and dolichocephalic
Sex	Male
	Female
Age	0–1 years
	2–5 years
	6–11 years
	≥12 years
Clinical signs	Palate ptosis
	Sneezing
	Reverse sneezing
	Nasal discharge
	Unilateral
	Bilateral
	Serous nasal discharge
	Purulent nasal discharge
	Hemorrhagic nasal discharge
	Epistaxis
	Starter
	Facial deformity
	Snore
Duration of clinical signs	<1 months
	1–3 months
	3–6 months
	6–12 months
	>1 years
Radiographic findings	Unilateral increased opacity
	Bilateral increased opacity
	Laterality
	Osteoclasia
	Destruction of turbinate
	Destruction of cribriform plate
	Destruction of nasal septum
	Destruction of other bone (nasal bone, maxilla, and frontal bone)
	Increased opacity of the frontal sinus
	Decreased opacity of periodontal membrane

Statistical analyses

The performance of the diagnostic prediction model was evaluated based on the sensitivity and specificity of the diagnostic results for each group. In the LASSO logistic regression analysis, the optimum cutoff value was selected based on the Youden index in the Receiver Operating Characteristic (ROC) curve, and the sensitivity, specificity, C-statistic, and log-odds ratio for the variables were calculated. All analyses were performed using Prism, version 9.00 statistical software (GraphPad Software, San Diego, CA, USA) and R, version 4.2.2 (https://cran.r-project.org).

RESULTS

Animals

In GM, 103 dogs met the inclusion criteria. Regarding the details of the disease, NP (n=38, 36.9%) was the most common, followed by NIR (n=33, 32.0%), DD (n=15, 14.6%), FB (n=12, 11.6%), and SNA (n=5, 4.9%). In addition, there were cases in which secondary bacterial rhinitis occurred in NP or SNA, but there was only one primary disease, and they were classified according to this diagnosis. Miniature dachshunds (n=31, 30.1%) were the most common, followed by Chihuahuas (n=12, 11.7%) and Toy poodles (n=10, 9.7%). The remaining breeds were as follows: 7 mixed breeds, 6 Golden retrievers, 4 Miniature Schnauzers, 3 Labrador retrievers, Pembroke Welsh Corgis, and Pomeranians; 2 Shiba Inus, Pugs, Papillons, Beagles, Boston terriers, and Rottweilers; and 1 French bulldog, Italian greyhound, Airedale terrier, Kaninchen dachshund, Shih Tzu, Jack Russell terrier, Staffordshire bull terrier, Siberian husky, Flat-coated retriever, Border collie, Yorkshire terrier, and Wire fox terrier. In GV, 86 dogs met the inclusion criteria. Regarding the details of the disease, NP (n=44, 51.2%) was the most common, followed by NIR (n=21, 24.5%), FB (n=10, 11.6%), DD (n=9, 10.5%), and SNA (n=2, 2.2%). Miniature dachshunds and Toy poodles (n=10; 11.6%) were the most common breeds, followed by mixed breeds (n=9; 10.5%). The remaining breeds were as follows: 8 Chihuahuas, 6 Shiba Inus, 4 Miniature Schnauzers and Jack Russell terriers; 3 Labrador retrievers, Kaninchen dachshunds, and French bulldogs, 2 Pembroke Welsh Corgis, Pomeranians, Beagles, Italian greyhounds, and Shetland sheepdogs; and 1 Boston terrier, Golden retriever, Pug, Border collie, Yorkshire terrier, Miniature pinscher, Maltese, Petit basset griffon vendeen, Doberman, Toy Manchester terrier, Standard poodle, Cavalier King Charles Spaniel, Australian Shepherd, West Highland White Terrier, Alaskan Klee Kai, and English cocker spaniel.

Signalment, clinical signs, duration, and radiographical findings

Details of the signalment, clinical signs, duration, and radiographic findings observed in each nasal disease were shown in Table 2. In terms of breeds, GM had 82 mesocephalic and dolichocephalic breeds and 21 brachycephalic breeds, whereas GV had 70 mesocephalic and dolichocephalic breeds and 16 brachycephalic breeds. In GM, there were 60 males (39 castrated) and 43 females (31 spayed), whereas GV included 47 males (38 castrated) and 39 females (38 spayed). For GM, the median age for all nasal diseases was 122 months (range: 10–198 months), whereas GV had a median age of 132.5 months (range: 18–187 months). In GM, in terms of age for each disease, most cases were over 6 years of age for NP (n=38, 100%), SNA (n=4, 80.0%), NIR (n=27, 81.8%), and DD (n=15, 100%). In contrast, FB tended to occur more frequently in patients under the age of 6 years in GM (n=9; 74.9%). The most common clinical sign of nasal discharge was observed in 88.3% (n=91) of the GMs. Additionally, sneezing (n=77, 74.8%) and reverse sneezing (n=48, 46.6%) were also common. The localization of nasal discharge in the GM was more unilateral in the NP (n=21, 55.3%), SNA (n=5, 100%), and FB (n=9, 75.0%) groups, whereas it was more bilateral in the NIR (n=27, 81.8%) and DD (n=9, 60.0%) groups. Regarding nasal discharge in the GM, serous and purulent nasal discharges were more common than hemorrhagic nasal discharge in SNA (n=4, 80.0% and n=4, 80.0%, respectively), NIR (n=26, 78.8% and n=28, 84.8%, respectively), DD (n=11, 73.3% and n=13, 86.7%, respectively), and FB (n=8, 66.6% and n=7, 58.3%, respectively). In contrast, hemorrhagic nasal discharge was more common than serous or purulent nasal discharge in patients with NP (n=19; 50.0%). In GM, epistaxis and facial deformities tended to be more common in patients with NP (n=15, 39.5% and n=8, 21.1%, respectively) than in those with other diseases. In addition, in GM, palate ptosis was observed only in patients with NP (n=4, 10.5%) and not in those with other diseases. Regarding the duration of clinical signs in GM, there were relatively few cases in which more than 1 year had passed (n=11, 10.6%), and no obvious difference was observed in other periods. For each disease in GM, more than half of NP (n=25, 65.8%), DD (n=11, 73.4%), and FB (n=7, 58.4%) cases occurred within 3 months, whereas SNA (n=5, 100%) and NIR (n=24, 72.7%) accounted for more than 3 months. On radiographic findings, increased opacity of the nasal cavity was most common in 85.4% (88/103) of patients in GM. Furthermore, in GM, increased opacity of the frontal sinus and osteoclasia was observed in 76.7% (79/103) and 62.1% (64/103), respectively. Increased opacity was mostly bilateral in all diseases (n=71, 68.9%). Osteoclasia was predominantly observed in the NP (n=26, 68.4%) and SNA (n=4, 80.0%), but it was also present in more than half of the cases with NIR (n=23, 69.7%) and FB (GM: n=6, 50.0%). In terms of osteoclasia, in GM, destruction of the nasal turbinate was the most common in all diseases (n=61, 59.2%), the nasal septum was mainly observed in NP (n=10, 26.3%) and SNA (n=1, 20.0%), and destruction of other bones (nasal bone, maxilla, and frontal bone) was observed only in NP (n=9, 23.7%). As presented in Table 2, most of the signalment, clinical signs, and radiographic findings in GV were similar to those in GM.

Table 2. Details of signalment, clinical signs and duration and radiographic findings observed in each nasal disease.

	Items	Neoplasia				Sino-nasal aspergillosis				Non-infectious rhinitis				Rhinitis secondary to dental disease				Rhinitis secondary to foreign body				Total
		Group M (n=38)		Group V (n=44)		Group M (n=5)		Group V (n=2)		Group M (n=33)		Group V (n=21)		Group M (n=15)		Group V (n=9)		Group M (n=12)		Group V (n=10)		Group M (n=103)		Group V (n=86)
		n	%	n	%	n	%	n	%	n	%	n	%	n	%	n	%	n	%	n	%	n	%	n	%
Breed	Brachycephalic	29	76.3	36	81.8	5	100	2	100	28	84.8	17	81.0	12	80.0	8	88.9	8	66.6	7	70.0	82	79.6	70	81.4
	Mesocephalic and dolichocephalic	9	23.7	8	18.2	0	0	0	0	5	15.2	4	19.0	3	20.0	1	11.1	4	33.4	3	30.0	21	20.4	16	18.6
Sex	Male	17	44.7	19	43.2	4	80.0	2	100	21	63.6	14	71.4	10	66.7	5	55.6	8	66.6	7	60.0	60	58.3	47	54.7
	Female	21	55.3	25	56.8	1	20.0	0	0	12	36.4	7	28.6	5	33.3	4	44.4	4	33.4	3	40.0	43	41.7	39	45.3
Age	0–1 years	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1	8.3	1	10.0	1	0.9	1	1.2
	2–5 years	0	0	2	4.5	1	20.0	1	50.0	6	18.2	0	0	0	0	1	11.2	8	66.6	1	10.0	15	14.6	5	5.8
	6–11 years	22	57.9	20	45.5	3	60.0	1	50.0	24	72.7	17	81.0	11	73.3	4	44.4	2	16.8	3	30.0	62	60.2	45	52.3
	≥12 years	16	42.1	22	50.0	1	20.0	0	0	3	9.1	4	19.0	4	26.7	4	44.4	1	8.3	5	50.0	25	24.3	35	40.7
Clinical signs	Palate ptosis	4	10.5	3	6.8	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	4	3.9	3	3.5
	Sneezing	25	65.8	26	59.1	4	80.0	2	100	26	78.8	17	81.0	13	86.7	8	88.9	9	75.0	9	90.0	77	74.8	62	72.1
	Reverse sneezing	13	34.2	9	20.5	1	20.0	1	50.0	20	60.6	9	42.9	7	46.7	3	33.3	7	58.3	5	50.0	48	46.6	27	31.4
	Nasal discharge	29	76.3	23	52.3	5	100	2	100	32	97	21	100	14	93.3	8	88.9	11	91.7	10	100	91	88.3	64	74.4
	Unilateral	21	55.3	20	45.5	5	100	2	100	5	15.2	7	33.3	5	33.3	6	66.7	9	75.0	6	60.0	45	43.7	41	47.7
	Bilateral	8	21.1	3	6.8	0	0	0	0	27	81.8	14	66.7	9	60.0	2	22.2	2	16.8	4	40.0	46	44.7	23	26.7
	Serous nasal discharge	16	42.1	12	27.3	4	80.0	1	50.0	26	78.8	15	71.4	11	73.3	4	44.4	8	66.6	6	60.0	65	63.1	38	44.2
	Purulent nasal discharge	10	26.3	10	22.7	4	80.0	2	100	28	84.8	17	81.0	13	86.7	6	66.7	7	58.3	9	90.0	62	60.2	44	51.2
	Hemorrhagic nasal discharge	19	50.0	15	34.1	3	60.0	2	100	4	12.1	6	28.6	4	26.7	6	66.7	7	58.3	5	50.0	37	35.9	34	39.5
	Epistaxis	15	39.5	21	47.7	2	40.0	0	0	0	0	0	0	1	6.7	2	22.2	0	0	1	10.0	18	17.5	24	27.9
	Starter	16	42.1	17	38.6	2	40.0	0	0	8	24.2	6	28.6	1	6.7	3	33.3	4	33.4	5	50.0	31	30.1	31	36.0
	Facial deformity	8	21.1	15	34.1	1	20.0	0	0	0	0	0	0	1	6.7	0	0	0	0	0	0	10	9.7	15	17.4
	Snore	16	42.1	8	18.2	3	60.0	1	50.0	12	36.4	5	23.8	5	33.3	3	33.3	7	58.3	5	50.0	43	41.7	22	25.6
Duration of clinical signs	<1 months	9	23.7	15	34.1	0	0	0	0	1	3.1	2	9.1	3	20.0	3	33.3	4	33.4	1	10.0	17	16.5	21	24.4
	1–3 months	16	42.1	13	29.5	0	0	1	50.0	8	24.2	5	22.7	8	53.4	2	22.2	3	25.0	4	40.0	35	34.0	25	29.1
	3–6 months	5	13.2	10	22.7	3	60.0	1	50.0	11	33.3	4	18.2	0	0	3	33.3	3	25.0	3	30.0	22	21.4	21	24.4
	6–12 months	7	18.4	5	11.4	1	20.0	0	0	7	21.2	7	31.8	2	13.3	1	11.2	1	8.3	1	10.0	18	17.5	14	16.3
	>1 years	1	2.6	1	2.3	1	20.0	0	0	6	18.2	3	18.2	2	13.3	0	0	1	8.3	1	10.0	11	10.6	5	5.8
Radiographic findings	Unilateral increased opacity	7	18.4	7	15.9	1	20.0	1	50.0	5	15.2	1	4.8	0	0	1	11.1	4	33.4	1	10.0	17	16.5	11	12.8
	Bilateral increased opacity	24	63.2	32	72.7	4	80.0	1	50.0	25	75.8	17	81.0	12	80.0	6	66.7	6	50.0	5	50.0	71	68.9	61	70.9
	Lateralitya	19	50.0	24	54.5	2	40.0	2	100	12	36.4	4	19.0	0	0	2	22.2	6	50.0	2	20.0	39	37.9	34	39.5
	Osteoclasiab	26	68.4	30	68.2	4	80.0	2	100	23	69.7	7	33.3	5	33.3	4	44.4	6	50.0	5	50.0	64	62.1	48	55.8
	Destruction of the turbinate bone	24	63.2	27	61.4	4	80.0	2	100	22	66.7	6	28.6	5	33.3	3	33.3	6	50.0	5	50.0	61	59.2	43	50.0
	Destruction of the cribriform plate bone	18	47.3	16	36.4	2	40.0	1	50.0	14	42.4	2	9.5	2	13.3	1	11.1	3	25.0	2	20.0	39	37.9	22	25.6
	Destruction or displacement of the nasal septum	10	26.3	14	31.8	1	20.0	1	50.0	3	9.0	0	0	0	0	1	11.1	0	0	1	10.0	14	13.6	17	19.8
	Destruction of other bonesc	9	23.7	10	22.7	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	9	8.7	10	11.6
	Increased opacity of the frontal sinus	29	76.3	33	75.0	5	100	2	100	26	78.8	15	71.4	11	73.3	6	66.7	8	66.6	5	50.0	79	76.7	61	70.9
	Decreased opacity of periodontal membrane	6	15.8	4	9.1	1	20.0	0	0	11	33.3	2	9.5	7	46.7	3	33.3	6	50.0	1	10.0	31	30.1	10	11.6

^a Laterality indicates that there is a difference in the strength of the increased opacity in the nasal cavity. Even bilateral increased opacity were included as positive if there was a difference in strength between the left and right nasal cavity. ^b Osteoclasia was included as positive if destruction at least one of the nasal turbinate bone, cribriform plate bone, nasal septum, or other bones was observed. ^c Other bones included the nasal bone, maxilla, and frontal bone.

Diagnostic prediction model

Model 1: Based on the similarity of clinical signs and radiographic findings (Table 2), the cases were classified into the following three categories: NP and SNA, NIR and DD, and FB. For NP and SNA, the first condition was defined as age ≥6 years, duration of clinical signs <3 months, presence of hemorrhagic nasal discharge or epistaxis, and presence of osteoclasia based on radiographical findings (Fig. 1). Fourteen patients with NP and SNA met this condition, and none of them had other diseases. Cases that did not meet the first condition included as a plus case that met at least one of the following (i) −(vi) conditions: (i) palate ptosis presence, (ii) facial deformity and osteoclasia presence, (iii) epistaxis presence, (iv) age ≥6 years, duration of clinical signs <1 years, laterality and osteoclasia presence, (v) destruction of the nasal septum presence, and (vi) destruction of other bone presence (Fig. 1). These secondary conditions included 16 of 20 for NP and SNA and 9 of 48 for other diseases. As a result, the sensitivity and specificity of NP and SNA were 0.88 and 0.81, respectively (Table 3). When this prediction model was validated with GV, 35 dogs with NP and SNA and 7 with other diseases met this condition, and the sensitivity and specificity were 0.92 and 0.78, respectively (Table 3). In brachycephalic breeds, the condition was defined as age ≥6 years, duration of clinical signs <6 months, and presence of one or more of the following four conditions: palate ptosis presence, facial deformity presence, hemorrhagic nasal discharge presence, or epistaxis presence. Seven dogs with NP and one with other diseases met this criterion (Fig. 2A). As a result, the sensitivity and specificity of NP were 0.78 and 0.92, respectively (Table 3). In GV, 5 dogs with NP and 2 dogs with other diseases met this condition, with a sensitivity and specificity of 0.63 and 0.75, respectively (Table 3).

Fig. 1.

Model 1 generated diagnostic prediction model for neoplasia (NP) and sino-nasal aspergillosis (SNA) in non-brachycephalic breeds. The prediction model positively included 88.2% (30/34) of NP and SNA and 18.8% (9/48) of the other diseases in Group M and 92.1% (35/38) and 21.9% (7/32) in Group V, respectively.

Table 3. Sensitivity and specificity in diagnostic prediction models for each group using manual and LASSO logistic regression analysis.

		Group M		Group V
		Sensitivity	Specificity	Sensitivity	Specificity
Manually	Neoplasia and sino-nasal aspergillosis	0.88	0.81	0.92	0.78
	Non-specific rhinitis and rhinitis secondary to dental disease	0.78	0.88	0.64	0.80
	Rhinitis secondary to foreign body	0.75	0.90	0.29	0.90
	Neoplasia (brachycephalic)	0.78	0.92	0.63	0.75
	Rhinitis (brachycephalic)	0.83	0.67	0.75	0.75
LASSO logistic regression analysis	Neoplasia and sino-nasal aspergillosis	0.93	1	0.93	0.75
	Non-specific rhinitis and rhinitis secondary to dental disease	0.96	0.89	0.8	0.79
	Rhinitis secondary to foreign body	0.75	0.92	0.2	0.97

Fig. 2.

Model 1 generated diagnostic prediction models for neoplasia (NP, A) and rhinitis (B) in brachycephalic breeds. (A) The prediction model positively included 77.8% (7/9) of NP and 8.3% (1/12) of rhinitis in Group M and 62.5% (5/8) and 25.0% (2/8) in Group V, respectively. (B) The prediction model positively included 83.3% (10/12) of rhinitis and 33.3% (3/9) of NP in Group M and 75.0% (6/8) and 25.0% (2/8) in Group V, respectively.

In NIR and DD, the first condition was defined as age ≥6 years, palate ptosis absence, sneezing or reverse sneezing presence, serous or purulent nasal discharge presence, epistaxis absence, and osteoclasia absence (Fig. 3). This condition was met in 11 dogs with NIR and DD and 1 with other diseases. Among those who did not meet the first condition, those who met 4 or 5 out of 6 of the first condition were advanced to the second condition. The second condition was defined as the duration of clinical signs >1 months, palate ptosis absence, sneezing or reverse sneezing presence, serous purulent nasal discharge presence, epistaxis absence, facial deformity absence, bilaterally increased opacity presence, and destruction of the nasal septum and other bones absence (Fig. 3). As a result, 20 dogs with NIR and DD and 4 with other diseases met the second condition, with a sensitivity and specificity of 0.78 and 0.88, respectively (Table 3). When this prediction model was validated with GV, 16 dogs with NIR and DD and 9 with other diseases met the conditions, with a sensitivity and specificity of 0.64 and 0.80, respectively (Table 3). In brachycephalic breeds with rhinitis, the condition included palate ptosis absence, facial deformity absence, sneezing or reverse sneezing presence, serous or purulent nasal discharge presence, hemorrhagic nasal discharge absence, and epistaxis absence; six dogs with rhinitis and one with other diseases met the condition (Fig. 2B). Of the cases that did not meet the first condition, cases with palate ptosis absence, absence of facial deformity, absence of epistaxis, and meeting the following conditions (i) or (ii) were included as positive: (i) duration of clinical signs >6 months or (ii) age <6 years (Fig. 2B). As a result, 4 dogs with rhinitis and 2 with other diseases were included in the second condition, and the sensitivity and specificity were 0.83 and 0.67, respectively (Table 3). When this prediction model was validated with GV, 6 dogs with rhinitis and 2 with other diseases met this condition, with a sensitivity and specificity of 0.75 and 0.75, respectively (Table 3).

Fig. 3.

Model 1 generated diagnostic prediction model for non-infectious rhinitis (NIR) and rhinitis secondary to dental disease (DD) in non-brachycephalic breeds. The prediction model positively included 77.5% (31/40) of NIR and DD and 11.9% (5/42) of the other diseases in Group M and 64.0% (16/25) and 20.0% (9/45) in Group V, respectively.

In FB, the first condition was age <6 years, unilateral nasal discharge presence, and epistaxis absence (Fig. 4). This condition was observed in four dogs with FB and three with other diseases. Among cases not meeting the first condition, cases were included as positive if they met the following conditions: duration of clinical signs <1 months or >1 years, palate ptosis absence, sneezing or reverse sneezing presence, serous or purulent nasal discharge presence, epistaxis absence, facial deformity absence, and destruction of cribriform plate, nasal septum, and other bones absence. (Fig. 4). As a result, the second condition included 2 dogs with FB and 4 with other diseases, with a sensitivity and specificity of 0.75 and 0.90, respectively (Table 3). When this predictive model was validated by GV, 2 dogs with FB and 6 with other diseases met this criterion, with a sensitivity and specificity of 0.29 and 0.90, respectively (Table 3).

Fig. 4.

Model 1 generated diagnostic prediction model for rhinitis secondary to foreign body (FB) in non-brachycephalic breeds. The prediction model positively included 75.0% (6/8) of FB and 9.5% (7/74) of the other diseases in Group M and 28.6% (2/7) and 9.5% (6/63) in Group V, respectively.

Model 2: For NP and SNA, variables for model building were selected, and the log-odds ratios for variables calculated by LASSO logistic regression analysis included epistaxis and 1.14, destruction of other bones on radiographic examination and 0.58, hemorrhagic nasal discharge and 0.12, age >12 years and 0.09, and starter and 0.07. The ROC curve constructed with these variables had a C-statistic of 0.96, a sensitivity of 0.93, and a specificity of 1 at the cutoff value (0.46) chosen based on the Youden index (Fig. 5A, Table 3). Evaluation of the prediction model using GV gave a sensitivity and specificity of 0.93 and 0.75, respectively (Fig. 5D, Table 3). For NIR and DD, bilateral and 1.14 and purulent nasal discharge and 0.79 were included as variables and log-odds ratios for model building. As a result, the C-statistic of the ROC curve was 0.96, and the selected cutoff value (0.46) gave a sensitivity and specificity of 0.96 and 0.89, respectively (Fig. 5B, Table 3). Evaluation of the predictive model by GV yielded a C-statistic of 0.83, with a sensitivity and specificity of 0.80 and 0.79 (Fig. 5E, Table 3). For FB, only age 2–5 years and 1.14 was selected as a variable and calculated log-odds ratio, respectively, with a C-statistic of 0.91, cutoff value of 0.38, and sensitivity and specificity of 0.75 and 0.92, respectively (Fig. 5C, Table 3). Evaluation of GV yielded a C-statistic of 0.58, and sensitivity and specificity of 0.2 and 0.97, respectively (Fig. 5F, Table 3).

Fig. 5.

Receiver Operating Characteristic curves for each group generated by LASSO logistic regression analysis. The C statistics were 0.96 (A), 0.96 (B), 0.91 (C), 0.89 (D), 0.83 (E), and 0.58 (F), respectively. (A) Neoplasia (NP) and sino-nasal aspergillosis (SNA) in Group M (GM). (B) Non-infectious rhinitis (NIR) and rhinitis secondary to dental disease (DD) in GM. (C) Rhinitis secondary to foreign body (FB) in GM. (D) NP and SNA in Group V (GV). (E) NIR and DD in GV. (F) FB in GV.

DISCUSSION

In this study, we created a diagnostic prediction model using less invasive tests such as signalment, clinical signs, and radiography and investigated whether it is possible to distinguish nasal diseases. Few studies on diagnostic prediction models have been reported in veterinary medicine so far. Some studies have demonstrated that artificial intelligence (AI) is useful for identifying lung and cardiovascular diseases from radiographic findings [12, 19, 25]. Other studies have reported prediction models for chronic kidney disease using signalment and blood tests [13]. However, there have been few studies on diagnostic prediction models that use a variety of tests rather than a single test, and there are no reports on their implementation in nasal disease. In this study, NP was most common in both the GM and GV, followed by NIR. This result is similar to those of previous reports, suggesting that these diseases are major nasal diseases in dogs [21, 27, 35]. However, SNA resulted in fewer cases than in previous reports, especially for GV [21]. This may be due to the large number of small dogs raised in Japan. Additionally, regional differences are believed to have an effect. In terms of age, the majority of patients were over 6 years old for all diseases except FB. Specifically for NP, 79 of the 82 dogs with combined GM and GV were over 6 years old. Nasal neoplasia is a major disease of older dogs [22], with a median age of 8–11 years in previous reports [8, 14, 31, 35]. However, in this study, NP was also observed in dogs aged 6–7 years, suggesting that NP should be considered as the main differential disease in dogs aged >6 years. In addition, our results are similar to those that reported that NIR, DD, and SNA are more likely to occur in middle-aged and elderly dogs [11, 20, 35, 38]. Unlike these diseases, FB was most common in 10 of 19 dogs under six years of age, confirming previous reports that it tends to occur in dogs under seven years of age [6]. Therefore, the difference in the age at which these diseases occur is an important factor in distinguishing nasal diseases.

Among clinical signs, nasal discharge and sneezing were the most common. In particular, nasal discharge was observed in 155 of 189 dogs (82.0%) included in this study, which is consistent with previous reports [23, 29]. Regarding the localization and characteristics of the nasal discharge, NIR and DD were bilateral and often purulent, NP and SNA were often unilateral and hemorrhagic, and FB was often unilateral, similar to previous reports [2, 6, 14, 29, 31, 38]. However, one study reported that more than half of the rhinitis cases were unilateral [29], and another study reported that hemorrhagic nasal discharge is also common in rhinitis [20, 37]. The localization and characteristics of nasal discharge are considered to be related not only to the type of nasal diseases but also to the severity and progression of inflammation. Therefore, we believe that the evaluation of nasal discharge should be carefully performed in conjunction with the duration of clinical signs and other tests. Other characteristic clinical signs were epistaxis in the NP and SNA [21, 35], and palate ptosis and facial deformity in the NP, supporting previous reports [14, 31]. Epistaxis and facial deformity have also been observed in other nasal diseases; however, in this study, the incidence was 4% (4/100) and 1% (1/100), respectively. Therefore, these signs were specific to the NP and SNA. In addition, because palate ptosis was observed only in NP, this sign is considered an important factor in the diagnosis of NP. Previous studies have reported a median duration of 3 months for NP, 4.5 months for NIR, 7 months for SNA, 2.5 months for DD, and several days for FB [35]. In this study, more than half of the NP and FB cases were less than 3 months, corroborating previous results. However, chronic rhinitis has been reported to last from several weeks to years [37], and cases of various durations were observed in this study. Neoplasia is a progressive disease, and FB is an acute disease; therefore, patients often visit hospitals relatively early. However, as rhinitis has various symptoms, progressions, and causes, it persists for a short period to a long period of time. Therefore, long-term findings of >1 year are strongly suggestive of chronic rhinitis; however, in other cases, it is difficult to rule out rhinitis based on duration alone.

Radiographic findings showed increased opacity in 160 (84.7%) of the 189 dogs included in the study, increased frontal sinuses in 140 (74.1%), and osteoclasia in 112 (59.3%) dogs, similar to previous studies [10, 23, 32]. Increased opacity was the most common bilateral result in all diseases. Non-infectious rhinitis is generally considered to be bilateral [10, 32], whereas NP and SNA are generally considered to be more unilateral. However, previous studies have shown that some results are more unilateral, while others are more bilateral in NP and SNA [10, 23, 32,33,34]. This is thought to be due to the severity and the degree of progression of the disease. In this study, there were many cases in which the condition had progressed before coming to the hospital; therefore, bilateral cases were believed to be more common. In addition, there are reports that NIR is often unilateral [10, 20]. Based on these findings, it is considered that increased opacity of the nasal cavity is unlikely to be a specific finding in both unilateral and bilateral cases. However, since there were many cases in which laterality was observed in the NP in this study, the presence or absence of laterality may assist in the diagnosis, even in bilateral cases. Osteoclasia was most common in the nasal turbinate, followed by the cribriform plate. This finding was most common in the NP and SNA. Although previous studies have reported that turbinate destruction is common in NP and SNA, it is also observed in NIR and FB [10, 23, 35]. In the present study, this was observed in half of the NIR and FB cases. In NIR, it was often observed in cases with a long clinical period; therefore, it is thought to be caused by chronic inflammation. In addition, many cases were included in which NIR revealed destruction of the nasal turbinate on radiography but not on CT and MRI. This may have caused the nasal discharge to resemble an osteoclasia. In addition, FB include cases of vomiting and regurgitation, which can cause destroy the nasal turbinate. In NP and SNA, the destruction of the nasal septum and bones surrounding the nasal cavity (nasal bone, maxilla, and frontal bone) is a characteristic finding [5, 10, 32,33,34]. In this study, it was frequently observed in these diseases, but it was also observed in small numbers in NIR, DD, and FB. Similar to the destruction of the nasal turbinate, it cannot be denied that chronic and strong inflammation may be involved, but it is also possible that problems such as the radiographic techniques, such as rolling and overlapping of surrounding bones, may affect the results.

In this study, a diagnostic prediction model was created using two patterns, manual and logistic regression analyses, based on characteristic items in less invasive examinations, such as clinical signs and radiographic findings. In veterinary clinical practice, many cases of nasal disease are suspected; however, diagnosis is often difficult because of similar clinical signs and radiographic findings [15, 21, 29, 30, 35]. Therefore, it is necessary to conduct evaluations using various items rather than just a single test. However, to date, no reports have evaluated these items in combination or created a diagnostic prediction model. In the diagnostic prediction models created in this study, the results of NP and SNA were also useful for evaluating the performance using the validation set. Neoplasia and SNA have similar clinical signs and radiographic findings, leading to the development of predictive models for both diseases. In addition, the items selected for NP and SNA were similar in both predictive models, suggesting that these items were specific to NP or SNA. In this prediction model, both patterns tended to have a higher sensitivity than specificity for the performance evaluation of GV. This prediction model, with its high sensitivity, can be useful for disease exclusion. Therefore, if a predictive model is not applied, the probabilities of NP and SNA are low. In addition, SNA is more likely to occur in large breeds, especially in retriever breeds [2, 27]; however, breeds could not be included in the conditions because of the small number of cases with SNA in our study. Therefore, there is a possibility that improvement in diagnostic accuracy can be expected by considering these breeds. In addition, predictive models were developed because NIR and DD as well as NP and SNA have similar clinical and radiographic findings. For the NIR and DD diagnostic prediction models, Model 1 had high specificity, whereas Model 2 had high sensitivity and yielded different results. In addition, both models had lower sensitivity and specificity in GV compared to GM. These factors may be due to fewer specific findings in NIR and DD and more conditions in GM to exclude other diseases. Therefore, it was more difficult to create predictive models for NIR and DD than for NP or SNA. Additionally, the creation of prediction models using AI is expected for these diseases. No clear or useful results were obtained for the performance evaluation of the predictive models for FB. One reason for this discrepancy may be the small sample size. In this study, there were 12 dogs in GM and 10 dogs in GV, which tended to be fewer than those with other diseases. In addition, in the FB cases included in this study, various factors were considered. Apart from common causes like grass and trees, instances involving the entry of gastric contents into the nasal cavity due to vomiting or regurgitation were also included. Age, clinical signs, duration, and radiographic findings varied due to these factors, and the absence of history items such as vomiting and regurgitation may have affected the prediction model [6, 17]. Therefore, in the future, a more useful prediction model can be created by accumulating more cases and adding investigation items including events that cause FB. Furthermore, in this study, in manual creation, it was difficult to evaluate the radiographic findings in brachycephalic breeds for anatomical reasons; therefore, we created it separately. Both the NP and rhinitis showed moderate results, but the number of cases was small, as with FB; therefore, it is necessary to accumulate more cases in the future.

A limitation of the diagnostic prediction model developed in this study was the difference in the radiography interpretation techniques. The radiographic images were evaluated by two veterinarians who specialize in respiratory; however, because the interpretation ability differed for each veterinarian, the difference in findings may have affected the prediction model. In addition, the presence or absence of intranasal opacity and osteoclasia is also affected by differences in the contrast of radiographic images and whether the radiographic images are evaluable; therefore, care should be taken during evaluation. One solution for this limitation is the use of AI. In recent years, the use of AI in radiographic findings has been reported in veterinary medicine, and its usefulness in detecting pleural effusion, cardiogenic pulmonary edema, and left atrial enlargement has been reported [12, 19, 25]. Therefore, we believe that the use of AI is a useful method for diagnostic prediction models. However, it should be noted that AI can resolve differences in interpretation techniques, and it is essential to obtain evaluable radiographic images. Another limitation of the predictive model in this study was that it did not include nonnasal diseases having nasal signs. In particular, epistaxis, when it is a single symptom, may indicate not only nasal disease, but also systemic diseases such as thrombocytopenia, coagulopathy, hypertension, and systemic infections that cause coagulopathy [3, 26]. In general, when NP or SNA is present, multiple symptoms such as nasal discharge are often present in addition to epistaxis; however, in the case of epistaxis alone, systemic diseases must be considered. Therefore, in cases presenting with epistaxis, other systemic diseases should be ruled out before using predictive models. Among systemic diseases, thrombocytopenia, coagulopathy, and hypertension are particularly common, and these diseases can be easily screened using blood tests and blood pressure measurements. Infectious diseases such as Ehrlichia have also been reported to often cause epistaxis [26], and testing may be necessary in areas of high incidence. Another limitation is that our university hospital does not provide emergency medical care, and all patients were consulted at a referral hospital for a second opinion. As a result, there are cases in which the period until hospital visits has been extended; therefore, there may be a difference in the period until hospital visits between referral hospital and university, which affects the usefulness of the diagnosis prediction model. In addition, because the facilities targeted in this study were located in a limited area, there may have been a bias in the distribution of diseases, such as a few infectious diseases.

In conclusion, in this study, we created a diagnostic prediction model that combines less invasive examinations such as signalment, clinical signs, and radiography in a complex manner for nasal diseases that are considered difficult to diagnose, and its usefulness for some diseases has been clarified. However, because the current predictive models focus on whether a disease can be roughly classified rather than on a definitive diagnosis, these prediction models can be used to present appropriate testing and treatment. We believe that these diagnostic prediction models can accurately determine the priority of patients who require examination under anesthesia. In the future, research into the effectiveness of AI utilization in the evaluation of intranasal radiography, and the addition and reexamination of items may lead to further improvements in diagnostic prediction models.

CONFLICT OF INTEREST

No conflicts of interest have been declared.