Mastitis pathogens are conventionally divided into contagious organisms that spread from cow to cow at milking, chiefly Staphylococcus aureus and Streptococcus agalactiae, and environmental organisms such as Streptococcus uberis, Escherichia coli and Klebsiella species. The non-aureus staphylococci form a third and very frequent group, a heterogeneous collection of species with differing resistance traits (Becker et al., 2014). Resistance statements are only meaningful at species or group level. Penicillin resistance in S. aureus lowers the chance of cure (Barkema et al., 2006), whereas a Gram-negative isolate is intrinsically resistant to penicillin and reporting it as such tells nobody anything. Reviews of the field have repeatedly noted that pooled figures across unlike organisms mislead (Oliver and Murinda, 2012).

Disk diffusion, standardised by Bauer et al. (1966), remains the practical method for a laboratory without automated equipment. Its results depend on the medium, the inoculum density, the disk content and the incubation conditions, and they can be interpreted only against zone diameter criteria derived for that exact procedure. The Clinical and Laboratory Standards Institute publishes the procedure for bacteria isolated from animals as document VET01, with veterinary breakpoints in a companion supplement (CLSI, 2018). Schwarz et al. (2010) list the faults that recur in veterinary susceptibility papers: unnamed methods, missing quality control and human breakpoints applied to animal isolates. Clinical breakpoints also differ in purpose from epidemiological cut-off values, which separate wild-type from non-wild-type populations (Kahlmeter et al., 2003).

The survey needs a written definition of a smallholder herd, for example by number of milking cows, and a list from which herds can be drawn. Cooperative membership lists, milk collection centre registers and vaccination records each cover a different part of the population, and the report should say which was used and whom it leaves out. Herds are selected at random from the list within each administrative area. Selection by willingness, or from the clients of one clinic, produces a sample enriched for herds with treatment failures. The number of herds is set with a standard formula for estimating a proportion, inflated by a design effect because isolates from one herd resemble each other (Dohoo et al., 2009).

Within each herd all lactating cows are examined. Quarters with visible changes in milk or udder are classed as clinical. The remaining quarters are screened with the California mastitis test, a cow-side reagent test that estimates the somatic cell content of milk (Schalm and Noorlander, 1957), and those above a chosen score are classed as subclinical. The two classes are kept apart throughout, since their pathogen profiles and their treatment histories differ. A short questionnaire records, for each sampled cow, the products given in the current lactation and the source of advice. Recall of drug names is poor, so the enumerator asks to see the packaging.

A resistance survey built on contaminated samples describes the skin and the bedding. Collection follows the procedure in the National Mastitis Council handbook: clean and dry the teat, discard the first streams, scrub the teat end with alcohol, and collect a small volume into a sterile tube held nearly horizontal (National Mastitis Council, 2017). Samples are chilled at once and reach the laboratory within a day, or are frozen if that is impossible, with the fact recorded. One sample per quarter is the practical choice for a survey. The diagnostic consequences of relying on a single sample were worked out by Dohoo et al. (2011), and their definitions of intramammary infection are adopted so that a plate with a few colonies of a minor organism is not counted as a case.

A measured loop of milk is streaked on blood agar and on MacConkey agar and read after one and two days of aerobic incubation. Plates with three or more colony types are recorded as contaminated and excluded. Staphylococci are separated into S. aureus and non-aureus species by coagulase testing, streptococci by the CAMP reaction and aesculin hydrolysis, and Gram-negative rods by the standard biochemical tests, as the handbook describes (National Mastitis Council, 2017). Where a reference laboratory offers mass spectrometry or sequencing, a random subset is sent for confirmation, and the agreement is reported. Every isolate that enters susceptibility testing is stored frozen in glycerol broth under a code that links it to herd, cow and quarter.

Disk diffusion is performed on Mueller-Hinton agar, with blood added for streptococci, from an inoculum adjusted to a 0.5 McFarland standard, exactly as VET01 prescribes (CLSI, 2018). The panel is fixed before the survey starts. It contains the drug classes sold locally for mastitis, taken from the Stage 2 questionnaire and a visit to suppliers, together with agents of concern for human health, guided by the World Health Organization list of critically important antimicrobials (WHO, 2019). Cefoxitin is included for staphylococci as the screening agent for methicillin resistance, which matters because livestock-associated strains, including those carrying the mecC gene first found in bovine milk, can pass to people (García-Álvarez et al., 2011). The reference strains named in the standard are tested with every new batch of medium or disks and at least weekly. A run in which a control zone falls outside its permitted range is discarded.

Zone diameters are recorded in millimetres, never only as a category, so that the data can be reinterpreted when criteria are revised. Each diameter is then classified with the bovine mastitis breakpoint where one exists. Few do. For the remaining drug and organism pairs the report must state whether a breakpoint for another animal species, a human breakpoint or an epidemiological cut-off was borrowed, and such results are presented in a separate table with a caution. Isolates are described as multidrug resistant only under the published definition of non-susceptibility to at least one agent in three or more antimicrobial categories (Magiorakos et al., 2012), and intrinsic resistances are left out of the count.

Results are tabulated by organism, by drug and by clinical or subclinical origin. The proportion of non-susceptible isolates is given with a confidence interval that allows for clustering within herds, which a survey procedure or a mixed model with herd as a random effect will provide (Dohoo et al., 2009). Only one isolate of a given species per cow is counted. Organisms represented by a handful of isolates are listed without percentages. Associations between reported drug use and resistance may be explored, but a cross-sectional design cannot show which came first, and the text should not suggest otherwise.

Each farmer receives the culture and susceptibility result for the cows sampled, in plain language and through the local veterinarian or extension worker, since a survey that takes milk and returns nothing will not be welcomed twice. The survey report names the method, the standard and its edition, the control strains and the breakpoint source, and it publishes the zone diameter distributions. Isolates are kept so that later work can look for resistance genes.