Author affiliations: University of Bristol, Bristol, UK (C. Tamayo, E. Szilassy, F. Sánchez-Vizcaíno), University of Liverpool, Liverpool, UK (A.D. Radford), University of Cambridge, Cambridge, UK (J.R. Newton)*Corresponding author. Tel. +44(0)7444566819. Email address: [email protected]: The Small Animal Veterinary Surveillance Network (SAVSNET) has developed mathematical models to analyse veterinary practice and diagnostic laboratory data to detect genuine outbreaks of canine disease in the UK. There are, however, no validated methods available to establish the clinical relevance of these genuine statistical outbreaks before their formal investigation is conducted. The aim of this study was to gain actionable understanding of veterinary practitioner’s preferences regarding which outbreak scenarios have a substantial impact in veterinary practice for six priority canine diseases in the UK.Methodology: An intensity sampling approach was followed to recruit veterinary practitioners according to their years of experience and the size of their practice. In depth semi-structured and structured interviews were conducted to describe outbreak notification and outbreak response thresholds for six canine endemic diseases, exotic diseases and syndromes. These thresholds reflected participants’ preferred balance between levels of excess case incidence and predictive certainty of the detection system. Interviews were transcribed and a thematic analysis was performed using NVivo 12.Results: Seven interviews were completed. Findings indicate higher preferred levels of predictive certainty for endemic diseases than for exotic diseases, ranging from 95-99% and 80-90%, respectively. Excess case incidence levels were considered clinically relevant at values representing an increase of two to four times the normal case incidence expectancy for endemic agents like parvovirus, and where they indicated a single case in the practice’s catchment area for exotic diseases like leishmaniosis and babesiosis.Conclusion: This study’s innovative methodology uses veterinary practitioners’ opinion to inform the selection of a notification threshold value in real world applications of stochastic canine outbreak detection models. The clinically relevant thresholds derived from participants’ needs will be used by SAVSNET to inform its outbreak detection system and to improve its response to canine disease outbreaks in the UK.Keywords: Disease surveillance, canine diseases, qualitative research, outbreak detection, outbreak reporting.1 IntroductionOne of the main factors that determine the effectiveness of an epidemic response is the timeliness of detection and notification to those that are potentially affected (1). In the UK, surveillance systems in farm animals and public health run centrally by government departments and agencies to identify increasing disease trends and detect disease outbreaks in their early stage, facilitating the prevention and control of health threats nationally and regionally (2,3). The relevant information derived from these surveillance activities is shared with the public via weekly reports (4) and online dashboards (5). However, these surveillance protocols do not currently exist in small companion animals, for which there is no standardised system of disease reporting or routine collection of surveillance data at a national level. This leaves canine populations in the UK vulnerable to epidemic threats.To begin to bridge this gap, the SAVSNet-Agile initiative (6) is developing a nationwide system for the timely detection and response to canine disease outbreaks in the UK. However, before such a surveillance and control system can be set up and implemented, it is necessary to determine which notification thresholds of increased level in case incidence relative to a previously identified baseline of expected cases would warrant alerting relevant stakeholders of potential outbreak threats.There are several methods that have been described to determine statistical outbreak notification thresholds. These methods vary depending on the disease type and the quality of the data that is available for surveillance purposes. For diseases that are endemic to the country, systems rely on historical data to establish a baseline level of disease and then use different mathematical methods to determine notification thresholds based on increases in case incidence, relative to the previously identified baseline (7,9). Other commonly used methods to establish outbreak notification thresholds are multi-chart schemes, which combine the results of individual time series that enable the rapid detection of subtle changes in disease (9), or methods that involve setting a number of standard deviations above the baseline of expected cases (10). For exotic and rare diseases, with lacking baseline data to draw patterns from, notification thresholds are defined ad-hoc, and it is often common to accept a single case as a threat that warrants generating an alert (11).Whilst these statistical methods have proven to be powerful for detecting disease anomalies, they often signal outbreaks that are not clinically relevant for veterinarians in practice. Therefore, outbreak notification systems that rely on such statistical signals might overload practitioners with information that is not actionable. In the long term, this could lead to a lack of confidence and engagement with the surveillance and outbreak notification system. To address these limitations, the aim of this study was to explore what threshold values based on veterinary practitioners’ opinion correspond to outbreaks that should be notified when detected by statistical methods because of their significant impact in veterinary practice for six priority canine diseases in the UK (12). In addition, we gained an understanding of the reasons that drive veterinary practitioners in selecting such threshold values and of how their in-practice behaviour can be impacted by clinically relevant outbreaks. To achieve these aims, an innovative methodology was developed based on the combination of semi-structured and structured interviews with companion animal veterinarians.2 Materials and methodsEthical approval for this work was granted by the University of Bristol Faculty of Health Sciences Ethics Committee (FREC, reference code: 98843).2.1 Study populationThe population of interest for this study were small animal veterinary clinicians working in the UK at the time of its conduction. Study participants were selected from this population following an intensity sampling approach, a type of purposive sampling to select information-rich cases located at the end of a population’s distribution (13). To do so, relevant population characteristics, or descriptors, were defined. These descriptors were believed to influence participant perspectives and behaviour regarding canine epidemics, and therefore influence their responses during the interviews. The following descriptors and levels of interest were used in the sampling process:Years of experience in small animal practice: it was assumed that more senior veterinarians are more likely to have experienced canine outbreaks throughout their career and have spent more time in practice overall, and this could influence their opinions and decision-making. Cut-off points were established to differentiate newly graduated veterinarians from those with many years of in-practice experience.Recent graduates: with less than five years of experience.Senior veterinarians: with over 10 years of experience.Practice size: since smaller practices have fewer employed veterinarians and see a lower number of cases, compared to bigger veterinary centres, it was expected that an outbreak would affect them differently and could potentially overwhelm their ability to cope with the increase in case incidence. To accurately reflect the difference between small and big veterinary practices, a summary of the existing veterinary practices in the UK was requested to the Royal College of Veterinary Surgeons (RCVS). This database included the total number of registered practices in the UK, and a breakdown of the number of employed veterinarians per practice. The practice directory was analysed to understand what the average size of a practice is and inform the categorisation. A total of 4252 individual veterinary sites were listed on the database. Over half of these sites had four or fewer registered veterinary surgeons (2917 or 68%). A total of 23% (984) of the sites had between five and nine employed veterinarians, and only a small number (348 or 8%) had 10 or more registered veterinary surgeons.Small veterinary practice: with fewer than four employed veterinarians.Large veterinary practices: with more than 10 employed veterinarians.2.2 Participant recruitmentPotential study participants were contacted through different means. Veterinary clinicians who were part of a pre-established network of collaborators for SAVSNET-Agile were emailed directly by the corresponding author (CTC). Further, veterinary practices that contributed data to SAVSNET at the time of the conduction of the study were contacted via email and via their practice management software (PMS); these practices contain a SAVSNET plugin window that can be used by the latter to relay messages to attending veterinarians (14). A participant recruitment advert was posted on the SAVSNET website (15), and shared on social media, namely on Twitter and Facebook. Lastly, an interview to advertise the study was conducted by the corresponding author (CTC) with UK veterinary magazine Vet Times (16).2.3 Interviews with veterinary companion animalsRecruited veterinarians took part in an interview session, which was conducted online via Microsoft Teams (17) or Zoom (18). The overall aim of the interviews was to explore clinically relevant outbreak scenarios for notification for two canine endemic diseases (leptospirosis and parvovirus), two canine exotic diseases (leishmaniosis and babesiosis), and two canine syndromes (respiratory and gastrointestinal disease). The interviews consisted of two components, with different aims.2.3.1 Semi-structured interviewThe first part of the interview followed a semi-structured (19), in-depth format and aimed to gain an understanding of the reasons that drive veterinary practitioners in defining what constitutes a clinically relevant outbreak and to understand how their in-practice behaviour can be impacted by such outbreaks. To facilitate the discussion, the interviewer first provided an overview of the epidemiological characteristics of the disease under consideration. The topic guide developed for the semi-structured interview can be found on Supplementary Material 1.2.3.2 Structured interviewOnce participants had reflected upon the subject matter, the interview changed to a structured format, to understand which outbreak scenarios would be selected by participants to receive timely alerts, due to their potential impact in their practice. Outbreak scenarios were described using two parameters, which represented characteristics of an outbreak notification:Excess case incidence: increased incidence above the expected baseline of cases in your practice’s catchment area, that would be of practical significance to a) warrant a notification about a potential outbreak, and b) drive you to change your behaviour in practice in response to an outbreak. Where selected levels of excess case incidence were different for a) and b), the selected value for the former was used to define a notification threshold, and the value for the latter was used to define an outbreak response threshold for canine diseases.Predictive certainty: level of confidence of the alerts generated by statistical outbreak detection models, defined by their credible interval, which normally takes values that range from 90 to 99% (20).Questions included in the structured interview (Supplementary Material 2) aimed to introduce the concepts of excess case incidence and predictive certainty to study participants and use them to describe disease-specific outbreak scenarios in a way that resonated with participants and their experience in practice.2.4 Data analysisInterview data were audio recorded and transcribed verbatim. All the analyses were conducted on NVivo (version 12) qualitative data analysis software (21). A coding framework was iteratively developed by the corresponding author (CTC) based on expected and emergent themes using deductive and inductive approaches, respectively. To enhance the consistency and reliability of the analysis, two authors (CTC, FSV) independently coded the transcript data from one of the interviews. Codes generated deductively and inductively from interview transcripts were grouped together into themes by following a hybrid approach to thematic analysis (22,23) (Figure 1). To ensure reliability and transparency, themes were continuously compared to the interview transcripts, to ensure they were true to the original data (24).
