This longitudinal study was conducted from 2019 to 2022 in the periurban areas of the Alto Selva Alegre (ASA) district in Arequipa. These areas were established on the outskirts of the city of Arequipa over the past few decades (26, 27). Periurban areas are characterized by unplanned and rapid growth, often comprised by rural–urban migrants, low SES, very limited infrastructure, and high environmental vulnerability (e.g., poor solid waste management), and overall low socio-economic status (SES). The National Institute of Statistics and Informatics classifies city blocks into five SES levels based on household income and other census-derived indicators (28). In our study area, most city blocks fell within the two lowest SES categories. New settlers usually come together from the same rural town, occupy a discrete and continuous piece of land, and request formal land tenure from the district and city authorities. This discrete geographical area is called a locality, and each locality is engaged in a specific economic activity, such as rock extraction or the raising of domestic farm animals (29, 30). Moreover, each locality is characterized by a different level of backyard livestock keeping. The study population consisted of feral dog and their caves located in four periurban localities: the San Isidro Labrador Pig Farmers Association (APSIL), San Luis Gonzaga Zone A, San Luis Gonzaga Zone D, and El Roble (Figure 1). These caves, which vary in depth, are found in the ground or walls along hillside paths. Feral dogs use them for resting and reproduction (4). Some caves are dug by the dogs themselves, while others are natural formations or landscape features that dogs utilize (4).

Between March 2020 and March 2022, Peru implemented several phases of COVID-19 restrictions, which varied in intensity and duration, but affected all the country with some consistency in urban areas such as Arequipa city. The initial phase (March–June 2020) involved a strict nationwide lockdown, including curfews, border closures, suspension of public transportation, and the shutdown of schools, restaurants, and markets. Gradual reopening began in mid-2020, followed by alternating periods of partial restrictions through 2021, including nighttime curfews, domestic travel limits between cities, and reduced operating capacity for restaurants. However, some of these restrictions were extended for longer periods in some cities that experience higher COVID-19-associated mortality and one of those cities was Arequipa (31, 32).

In Arequipa, these measures affected the flow of organic waste from restaurants, markets, and informal food vendors, which we consider the key anthropogenic food sources for free-roaming dogs. While some reports indicate a temporary reduction in food waste entering municipal landfills (24), others describe local adaptations such as increased street vending and small-scale food production that may have partially offset these declines (25).

At each study locality, caves were identified by field teams following visible paths created by regular movement of feral dogs. These paths were readily apparent both in the field and on satellite maps. Monthly surveillance visits were conducted to monitor the formation of new caves as well as to collect longitudinal data on the use of caves by feral dogs. Detailed information on the monthly surveys is provided elsewhere (4). Briefly, between September and December 2019, we conducted monthly pedestrian surveys within a 1.6-km² area encompassing the study localities in northeastern ASA to document free-roaming dogs and identify caves used as shelters. A cave was defined as a naturally occurring or canine-modified structure at least 1 meter deep, or sufficiently large to permit entry and shelter, with direct or indirect evidence of dog use; shallower structures with limited evidence of use were mapped as potential sleeping sites. Initial reconnaissance revealed extensive dog-formed trail networks, prompting a pilot comparison of grid-based transects versus dog-trail–based searches. The dog-made trail method was superior to the grid-based method, leading us to adopt that approach for the full survey. We mapped visible dog footpaths using Google Earth (33) and subsequently conducted paired-observer surveys along these trails, recording dog sightings and cave characteristics.

Field teams collected data through a standardized mobile data collection form developed in the World Veterinary Service (WVS) application (34). The form in the WVS app was designed and piloted by the research team and included fields for cave registration and the presence or absence of various indicators of feral dog activity (Supplement 1).

For each monthly visit, data on feral dog evidence was recorded. Observed evidence was categorized as direct or indirect. Direct evidence of feral dog presence included: sighting of solitary dogs, dog packs, puppy litters, or dead dogs inside or around the cave opening (Figure 2). Indirect evidence of feral dog presence included canine feces, fresh or desiccated, canine tracks, canine scratch marks, and prey carcases (Figure 3). Additional data were recorded along trails—regardless of cave presence—on dog packs, solitary dogs, and carcasses. No animal handling (e.g., capture, restraint, or sample collection) was conducted in this study. Each cave visit was conducted by two observers with equivalent experience levels, working together to increase detection sensitivity for both caves and animals (4).

The primary outcome analyzed was the count of caves with general evidence of feral dogs per month. Secondary outcomes included the count of caves with direct and indirect evidence, and the number of litters, solitary live dogs, packs, and dead dogs per locality (i.e., all those associated with a cave and those observed not associated with a cave). We compared these primary and secondary outcomes before and after/during the COVID-19 pandemic restrictions. We used Student’s t-tests for mean comparisons and Wilcoxon rank-sum tests for median comparisons. Simple generalized linear models (GLMs) were constructed to explore potential associations of time period (pre- or post- pandemic restrictions) with the primary and secondary outcomes. We used GLMs with a negative binomial family due to overdispersion in the data. In addition, given that a higher number of caves visited increases the likelihood of detecting caves with evidence of feral dogs, we included the number of caves visited each month in each locality as an offset term in our models and evaluated its inclusion using Akaike Information Criterion (AIC; Supplement 2).

After establishing a potential association between the outcomes and the pandemic restrictions, we conducted an interrupted time series (ITS) analysis as our primary analytical framework. ITS is a method used in longitudinal studies to assess whether a response variable changes, and if that change is immediate or progressive, following an intervention. In this study, the onset of COVID-19 restrictions was treated as the intervention, and changes in the number of caves with feral dog evidence were evaluated across this temporal breakpoint. The ITS model tested whether the onset of pandemic restrictions was associated with changes in the number of caves showing evidence of feral dogs before and during the restrictions. ITS used three variables besides the response variable (number of caves with evidence): a continuous variable representing time since the beginning of the study, a continuous variable representing time since the intervention (0 before), and a dichotomous variable indicating whether the time point is before or after the intervention. ITS results indicate whether there was an immediate and/or progressive change in the number of caves with evidence at the start of the restrictions, and whether such a change was statistically significant. The ITS analysis provided information about pre-intervention trends and whether those trends changed following the intervention. For the regression model we used a negative binomial family based on our simple GLMs results. The final parsimonious model was selected by comparing versions of the model with and without the locality variable using log-likelihood estimation. All statistical analyses, tables, and figures were prepared using R version 4.2.0 (35).

In order to understand the interactions and conflicts between feral dogs and human populations, we conducted in-depth interviews with farmers from periurban communities. These interviews were aimed at capturing lived experiences, farming practices, and community-level responses to the presence of feral dogs. In particular, we sought to explore how changes in food systems, such as the availability of organic waste, fluctuations in livestock production, and informal feeding practices, could contribute to shaping the ecological dynamics of feral dog populations. In the interviews, we explored the effect of the pandemic restrictions on farming activities and the indirect impact on periurban food systems that sustain feral dog populations. By examining how human activities related to food production and waste disposal influence dog behavior and survival, we aimed to generate insights into the socio-ecological drivers of human–dog interactions in rapidly changing periurban environments.

Purposive sampling was used to select participants across all four study localities. All participants were residents and farmers from periurban neighborhoods. Members of the research team recruited participants through door-to-door visits, during which they obtained contact information (e.g., cell phone numbers). To minimize face-to-face interaction and reduce the risk of COVID-19 transmission, interviews were conducted a few days later by phone.

Interview guides were developed to explore four main topics: farm animal husbandry, changes in feed and animal prices, interactions with feral dogs, and the impact of the COVID-19 pandemic on farming practices. Two members of the research team conducted the interviews: a Peruvian disease ecology researcher (MDLP, DVM, PhD in Epidemiology) and a Peruvian veterinary epidemiologist (OCA, DVM, MSc).

All interviews were digitally audio-recorded and transcribed, with detailed notes taken throughout. Centering on the four main topics mentioned above, an inductive coding approach was employed. Initially, the research team immersed themselves in the data to allow key topics to emerge organically, without preconceptions (36). Codes were then developed based on the emerging themes. Data was imported into Dedoose (37) and double-coded by two members of the research team. Any discrepancies in coding were discussed thoroughly until consensus was reached, with new codes added as necessary. All transcripts were re-coded using the finalized coding scheme. To address the primary research question, all data coded under themes related to feral dogs were summarized in tables, and representative quotes were selected for presentation.

The study area covered 3.91 km² across the four periurban localities. From September 2019 to March 2022, field teams conducted 9,114 cave observations. During the study period, we identified 38 previously undocumented caves, while 61 of the caves initially recorded had collapsed or otherwise deteriorated and no longer appeared suitable for use as dog shelters. In nearly every monthly visit, at least one cave showed evidence of feral dog activity, except for 2 months in San Luis Gonzaga D-the largest locality but with the fewest farms. On average, 16.42 caves per month showed signs of dog presence, primarily indirect (e.g., paw prints, claw marks, feces), while direct evidence such as live dogs or litters was less frequent (mean = 2.1 caves/month). Paw prints were the most common indicator (99% of positive observations), followed by claw marks (91%).

Across the entire study area, we recorded 46 litters of puppies, 166 dead dogs, 42 solitary adult dogs, and 97 packs. Of the 46 litters, 8 were located within or immediately adjacent to caves, suggesting feral reproduction, while the remaining 38 were found in open areas or near human dwellings, where dogs likely had some level of human contact. In or near caves, we also 25 observations of adult dogs inside caves, and 176 observations of dogs near caves were entrances. The mean litter size was 7 puppies, and the average pack size was 11 dogs.

Following the onset of COVID-19 restrictions, a statistically significant and sharp decline was observed in the number of caves with general, indirect and direct evidence of feral dog presence, evaluated with simple GLMs. Median monthly caves count with general evidence fell from 31.5 to 10 (p < 0.001), while direct evidence declined from 2 to 0 (p < 0.001; Table 1). Additional indicators such as litters (p < 0.001), solitary dogs (p < 0.001), dead dogs (p < 0.001), and packs (p < 0.001) also showed statistically significant reductions. However, no significant differences were found in the number of puppies per litter (p = 0.542) or the number of dogs per pack (p = 0.142; Table 2). In addition to measures of central tendency, bar graphs show these trends clearly (Figure 4). A gradual decrease was observed in the number of litters (Figure 4A), dead dogs (Figure 4B), and packs (Figure 4D) at the onset of the pandemic restrictions, along with a substantial drop in the number of live dogs (Figure 4C).

The ITS models for presence of feral dogs showed a statistically significant decrease in the number of caves with general evidence, indirect evidence, and direct evidence at the onset of COVID-19 restrictions (p < 0.010, p = 0.012, and p = 0.002, respectively; Table 3). Compared to the pre-restriction period, at the onset of restrictions, caves with general evidence decreased by 42%, caves with indirect evidence decreased by 41%, and caves with direct evidence decreased by 76% (Figure 5). Temporal trend variables—“visit number since the start of the study” and “visit number since the onset of restrictions”—were significant for direct evidence. Prior to restrictions, caves with direct evidence increased by 25% per month. After restrictions were implemented, this trend reversed, with the number of caves decreasing by 25% per month. Temporal variables were not statistically significant for general and indirect evidence. In these full multivariable models, locality was also a significant predictor. For all types of evidence, each locality experienced a more pronounced decline following the onset of pandemic restrictions compared to APSIL, the most densely populated human settlement in the study area.

We recruited 64 farmers during the door-to-door visits; when called, only 41 accepted to participate and were interviewed. All the interviewees raised or had raised farm animals and 20 of them were women. The interviewed farmers mentioned that the pandemic affected the prices of their animals, forcing them to reduce the number of animals they were raising, or, in some cases, to stop raising them altogether. This decline in small-scale animal husbandry not only impacted household income and food security and may have had ecological consequences, such as reduced availability of organic waste that could previously have served as a food source for feral dogs.

Farmers noted that animal farming can serve as a food source for feral dogs, due to inadequate disposal of animal carcasses, which are easily accessible to roaming dogs.

The reduction in solid waste during the pandemic, a primary food source for dogs, was associated with increased starvation among feral dog populations. Unexpectedly, despite mostly unfavorable discussions regarding feral dogs, one woman reported that some people actually felt sorry for them:

Multiple dimensions of human–animal conflict emerged during the in-depth interviews with local farmers, mostly describing concerns about safety of themselves, their children, or their farm animals encountering hungry feral dogs. Although some farmers described a short-term rise in predation immediately following the onset of lockdowns, most reported fewer dog attacks later in the pandemic period, consistent with the observed reduction in feral dog evidence. The identified dimensions and their descriptions are presented in Table 4. Interestingly, fecal contamination was not mentioned, despite the widespread presence of feces observed in the study area.