Mastitis

Definition

Mastitis is inflammation of the mammary gland. In nearly all commercially relevant cases it is initiated by bacteria ascending the teat canal and establishing an intramammary infection.

Prevalence and incidence

Mastitis is the largest addressable problem in dairy. Across 37 Wisconsin herds, clinical mastitis occurred at 24.4 cases per 100 cow-lactations. This is 1.7x the next most common condition, foot disorders at 14.5, and far ahead of metritis at 11.2 and ketosis at 8.6 14. USDA NAHMS found that 24.8% of all US dairy cows experienced clinical mastitis in a single survey year, on 99.7% of operations 15.

These figures count only the cases a human saw. The subclinical form, which produces no visible signs and is treated in Section 1.4, occurs at 15 - 40x the rate of the clinical form 16, so the true burden carried by any given herd is substantially larger than its case records suggest, and almost entirely uncounted.

Pathogenesis

The bovine teat canal is lined with keratin containing antimicrobial fatty acids, which decrease susceptibility to infection and act as an initial line of defense to bacterial intrusion 45. If bacteria breaches the keratin layer, recognition is immunological: receptors on mammary epithelial and immune cells detect conserved bacterial components, with TLR4 responding to gram-negative lipopolysaccharide (present in E. coli and Klebsiella) and TLR2 to gram-positive cell wall constituents (present in S. aureus and streps) 43.

Recognition triggers an inflammatory cytokine cascade, principally TNF-α, IL-1β, IL-6 and IL-8, all of which recruit polymorphonuclear neutrophils from blood into the gland 43,46. These neutrophils are the dominant component of what is measured as somatic cell count (SCC), which is why SCC rises during infection and why it is used as the conventional indicator of udder health.

The response, counterintuitively, is also the source of much of the damage. As Sordillo and colleagues put it, “the inflammatory cascade results not only in the escalation of local antimicrobial factors, but also in the increased movement of leukocytes and plasma components from the blood that may cause damage to host tissues” 43. Tight junctions between mammary epithelial cells loosen and the blood-milk barrier fails. This facilitates an ion exchange by which secretory tissue is damaged, which in turn causes a decrease in milk yield during the episode and for the remainder of the lactation period.

Whether an episode resolves or becomes chronic depends on the balance of that response. “A precarious balance between pro-inflammatory and pro-resolving mechanisms is needed to ensure optimal bacterial clearance” 43. When resolution fails, the infection persists subclinically, the gland continues to lose productive capacity, and the animal becomes a reservoir for transmission to herdmates.

Clinical and subclinical presentation

Clinical mastitis presents itself through observable abnormality, characterized by clots, flakes, or discoloration in the milk. In more severe cases, swelling and heat in the quarter, or more systemic signs like fever and inappetence are observed in the cow. These indications are identified by a human at milking, and it is the form around which treatment protocols and case records are built.

Subclinical mastitis, by contrast, produces no visible signs. While the milk appears normal and the animal well, the gland is infected, resulting in elevated SCC and depressed yield. Cases of subclinical mastitis are detectable only by instrumentation, such as somatic cell counters, conductivity measurements, or bacterial culture. Subclinical mastitis is the more challenging case, both because of the difficulty to diagnose and its prevalence, occurring at 15 - 40x the rate of the clinical form 16.

Etiology

The organisms responsible for inducing mastitis are conventionally split into two classes. Contagious pathogens live on and in infected udders and spread from cow to cow. The principal species are Staphylococcus aureus, Streptococcus agalactiae, Mycoplasma spp., and Corynebacterium 39. Transmission occurs “especially during milking” 39, carried on milkers’ hands, on shared towels, and on teat cup liners passed between animals. The alternative class, environmental pathogens live in bedding, manure and standing water, and infect opportunistically. The principal species are Escherichia coli, Streptococcus uberis, Enterococcus spp., and coagulase-negative staphylococci 39. These organisms also frequently exploit the milking event, entering the gland “during milking owing to the liner slippage, or when cow’s natural immunity is weak” 39.

The two classes behave differently under treatment: E. coli infections are largely self-limiting, while S. aureus infections are often not cleared at all.

Risk factors

There is variance in a cow’s susceptibility to infection. Multiparous cows are more vulnerable to intramammary infection than primiparous cows, and infections cluster heavily around parturition and the first month of lactation 39. Housing contributes as well, with “high stocking density, contaminated floor, wet bedding, poor ventilation, and hot and humid climate” all promoting environmental infection 39.

Machine and routine factors act directly at the teat. Liner slip admits environmental organisms during milking 39. Overmilking and inappropriate vacuum damage the teat end, degrading the keratin barrier described in Section 1.3. Inadequate stimulation before unit attachment produces incomplete let-down and prolonged machine-on time.

The human contribution

It has been argued that the greatest perpetrator of mastitis is poor bovine husbandry. The clearest evidence for this precedent over bovine biology comes from genetics. The heritability of clinical mastitis, estimated from producer-recorded data, is 3.1%. Somatic cell score, the correlated indicator trait used as a selection proxy, sits at 12% 42. In other words, roughly 97% of the variation in whether a cow develops clinical mastitis is not attributable to her genes.

The between-farm data concur. Across 37 Wisconsin herds, incidence ranged from 1.7 to 46.8 cases per 100 cow-lactations 14. There is a 27.5-fold spread among farms in the same state and in the same year. These cows are exposed to the same pathogens and environmental factors, so this variance cannot be intrinsic to the cow.

This variance, then, lives in the parlor, and specifically in the execution of the milking routine. The problem is that execution is frequently poor, and almost nowhere is it measured. In a study of 112 workers across 16 commercial dairies in Michigan and Ohio, 69% of observed milkings had inadequate prep time before unit attachment 38. Insufficient teat coverage during post-milking disinfection was observed in 9.8% of milkings. 70% of the workers had less than one year of experience, and when tested on milk quality knowledge, they answered correctly only 49.3% of the time 38.

A separate survey of commercial Holstein-Friesian farms found that while all farms applied post-milking disinfection, only 83.7% used any form of pre-milking disinfection, and only 65.1% used a pre-dip 37. On the chemical side, four of seven farm managers did not know the concentration of the caustic solution used to clean their milking machine, five of seven did not know the acid concentration, and none knew the disinfectant concentration, “but they all stated that the disinfectants were used according to the product manuals” 37.

The labor context makes this harder. A study commissioned by the National Milk Producers Federation found that roughly one-third of US dairy operations employ foreign-born workers, and that those farms produce 80% of the nation’s milk supply 40. Combined with the 70% of milking technicians who had less than one year of experience 38, the picture is of a critical infection-control procedure executed hundreds of times per shift by a workforce that is frequently new, frequently under throughput pressure, and, although procedure-dependent, often unsupervised.

Mastitis is therefore possibly best characterized as a disease of husbandry that presents as a disease of the cow.

Economic burden

The most rigorous global estimate, adjusted for disease comorbidity, places total dairy-cattle disease losses at $65 billion USD annually, of which clinical mastitis accounts for $13 billion and subclinical mastitis $9 billion, roughly a third of the entire burden 17.

The cost of mastitis can be broken down onto a per-cow basis. A widely used model of a clinical case in the first thirty days of lactation totals $439 per case, of which 73% is indirect: premature culling at $173 (39%) and future milk production loss at $132 (30%). Diagnostics account for $10, and therapeutics account for $36, which combined is just 10.4% of the total 18. Independent models developed in Canada, Spain, and Indonesia also roughly converge on the same structure 19,20. In the Canadian analysis, subclinical mastitis accounted for 48% of total mastitis-associated cost, with 72% of that subclinical cost arising from reduced milk yield in cows nobody flagged 19.

Interestingly, a significant financial burden also lies in the lack of resolution of flagged cows. A simulation of mastitis economics on an automatic-milking dairy found a median total cost of €230 per case, of which €118 was attributable to chronic cases persisting beyond 28 days 27. These are animals the existing sensors have already flagged, over and over.

Prevention

The five-point plan

Mastitis prevention is, in principle, resolved. A control program formalized in the 1960s, including many of today’s standard practices post-milking teat disinfection, dry cow therapy, prompt treatment of clinical cases, culling of chronic cows, and milking machine maintenance, drove Streptococcus agalactiae to near-eradication in well-managed herds, and is the reason contagious mastitis is a smaller share of the modern disease burden than it once was.

The evidence for the individual components is irrefutable. According to the National Mastitis Council, “the rate of new intramammary infection can be 50% lower when disinfecting teats with an effective product immediately after every milking compared to no disinfection,” an effect that is “especially effective against the contagious pathogens Staphylococcus aureus and Streptococcus agalactiae” 44. The requirement is specific: application immediately after cluster removal, at every milking, with all four teats completely covered 44.

The economics concur. A cost-efficiency analysis of five mastitis control strategies found that “postmilking teat disinfection emerged as the only strategy with a positive net economic benefit, significantly reducing mastitis incidence and associated economic losses” 20.

The measurement gap

Given the aforementioned, the gap in mastitis prevention is therefore not knowledge. It is the execution of the known practices. Training demonstrably moves compliance. After a structured training intervention, correct answers on knowledge assessment rose from 49.3% to 67.6%, pre-milking disinfectant contact time increased by nine seconds per cow, and the proportion of milkings with inadequate prep time fell from 69% to 48% 38. Insufficient post-dip coverage fell from 9.8% to 5.9%, and the upward trend in bulk tank somatic cell count on the study farms was halted 38.

The success of training is encouraging. Yet, the more important finding is that compliance remained at 48% immediately after a supervised intervention. Compliance decays after the trainer leaves, and nothing in the standard toolkit measures the decay.

Monitoring the operator

Parlor performance software has existed for decades, but it measures the machine rather than the person: vacuum level, pulsation, unit-on time, bulk milk flow, automatic cluster removal settings. These are useful, but entirely blind to bovine husbandry practices.

Computer vision is the modality best suited to closing that gap, and it is already commercially established in the milking parlor. CattleEye, distributed through GEA, mounts an inexpensive 2D surveillance camera above the passageway exiting the parlour and scores every animal that walks beneath it, identifying individuals by coat pattern and head shape 47,48. Validated against human mobility scorers across three commercial farms and 6,040 mobility scores, the system achieved 86.9% agreement, and in detecting cows with painful foot lesions it returned higher sensitivity than a trained human assessor (52% versus 29%) at equivalent overall accuracy 47. This is a genuine proof for the modality: commodity hardware, no per-animal device, autonomous scoring at every milking, and performance at or above the human it substitutes for.

What CattleEye does not do is watch the person. Both the published validation and the product documentation describe animal traits exclusively, namely mobility and body condition 47,48. It is, in other words, positioned to observe the consequence of poor husbandry rather than the husbandry itself, and a camera above the exit race is in any case placed to observe gait, whereas teat preparation happens at the stall and its sub-steps must be resolved in seconds rather than scored once per animal.

Computer vision systems are commercialized which focus on the latter issue. Cattle Care installs its own cameras in the milking parlor, captures video from every shift, and reports protocol deviations directly. The output is framed explicitly as personnel management, with bilingual employee performance reports intended to let producers “hold individual employees responsible” 49. The modality has also now been examined in the peer-reviewed literature, with a diagnostic accuracy study of computer vision monitoring of milking routines published in 2025 50.

Both halves of the problem are therefore addressable with commodity cameras, and neither product spans them. This matters more than it first appears. CattleEye identifies individual animals and enrolls them automatically by RFID 48, but has no view of the operator. Cattle Care observes the operator in detail, but its documentation describes deviations at the level of the event and the employee, with no stated linkage to the identity of the cow that received the deficient handling 49. Neither system, as documented, can therefore trace a specific protocol failure forward to a specific animal and then to that animal’s udder health outcome.

Automated mastitis detection

Signal classes

Blood remains the accepted gold standard for systemic health status, but it cannot be sampled at milking frequency. Milk, by contrast, is produced on a fixed schedule by the organ of interest, and hereafter we discuss what can be reasonably recorded from milk.

Electrical conductivity is the oldest and most widely deployed, measuring the ion exchange described in Section 1.3 directly. It requires no reagent and no optics, which is why nearly every automatic milking system carries it, and why Section 3.5 examines its underperformance at length. Somatic cell count, measured in-line by optical or fluorescence methods, is the closest available proxy to the reference standard used by laboratories and regulators, and in-line units are deployed commercially by several manufacturers. Milk colour and blood detection, using optical absorbance, capture the visually abnormal milk that defines a clinical case in most on-farm case definitions.

Beyond milk chemistry, milk yield and flow-curve features, including production rate, quarter-level yield asymmetry, and incomplete milkings, are available at no marginal hardware cost on any parlor with milk metering, but the signal is weak. Quarter-level indicators of this class produce odds ratios in the range of 1.18 to 2.26 and correlations with somatic cell count of 0.26 to 0.44 33, which is screening-grade rather than diagnostic-grade. Behavioral and physiological proxies, like rumination time, activity, lying time, and body temperature from ear tags or rumen boluses, detect the systemic response rather than the gland itself, and inherit the placement problems examined in Section 3.4.

Two further classes are promising but constrained. Biochemical markers including lactate dehydrogenase, N-acetyl-β-D-glucosaminidase, haptoglobin and cathelicidin have all demonstrated diagnostic value in the literature, but require reagent chemistry that is difficult to sustain in a durable on-farm device. Spectroscopic methods, principally mid-infrared analysis, are powerful for some traits and weak for mastitis specifically: a head-to-head comparison of four algorithms on national milk-recording data found mid-infrared prediction of ketosis reached an area under the curve of 0.877, while mastitis reached only 0.641 36.

Difficulty

It is non-trivial to meet the industry bar for automated clinical mastitis detection. Set by ISO/FDIS 20966, a sensitivity of at least 80% and specificity of at least 99% are required.

Hogeveen and colleagues reviewed sixteen peer-reviewed detection models published since 1992 and found sensitivities ranging from 47% to 100% and specificities from 69% to 99.8%. Importantly, none satisfied both criteria, so the researchers concluded that “None of the described studies satisfied the demands for CM [clinical mastitis] detection models” 3.

Twelve years later, an independent field study across 114 farms and 7,411 cows evaluated four commercial automatic milking systems, and “None of the evaluated AMSs [automatic milking systems] achieved the minimum SP [specificity] limit of 99%” 4.

Some structural problems explain why this bar has proven so resistant. The first is that sensitivity and specificity are not independent, and the industry’s stated priority is the harder of the two. Raising a detection threshold reduces false positives and specificity improves, but true cases are missed and sensitivity falls. The two must be traded against each other, and ISO demands an extreme value on the specificity side specifically.

The second, and most consequential, is base rate. The daily probability that any individual cow suffers from mastitis is roughly 0.5%. Specificity is applied to the healthy majority and sensitivity to the sick minority, so a small specificity error is multiplied by a large number while a large sensitivity gain is multiplied by a small one. At that prevalence, even with a respectable sensitivity and specificity, Post et al. found the positive predictive value to collapse to 0.07, a 93% false positive rate among alerts. By restricting farmer notifications to high-risk alerts, the positive predictive value rises to only 0.20 5. The same analysis cites a deployed system operating at 70% sensitivity and 98% specificity that produced 52 true positives and 3,636 false positives.

The third is that clinical mastitis is not defined by a clear threshold. As a result, case definitions for clinical mastitis vary between studies, as do the time windows within which an alert is scored as a true positive, and the reference standard against which it is compared. Reported performance figures are therefore not directly comparable, which is why Hogeveen et al. proposed a standardized false alert rate, rather than relying on specificity alone 3. A 2026 review reached the same conclusion, attributing the “wide variability in performance” across systems to “differences in validation protocols, sensor placement, and data windows,” and judging that current systems “remain insufficiently accurate for autonomous decision-making” 2.

Consequences

These benchmarks paling in comparison to the standard seem to have materialized into behavioral implications. Producers used sensor alerts when there were fewer than roughly twenty of them, and “the longer the system was in place, the less likely producers were to utilize alerts” 6. In a survey of 229 dairy producers, 36% identified “too much information provided without knowing what to do with it” as a barrier to adoption 7.

Another possible explanation for farmers’ low conviction in sensor technology is the stark lack of validation quality associated with many dairy sensors. A systematic review identified 129 commercially available dairy sensor technologies and found only 18 with peer-reviewed external validation, comprising 14% of the market. Rumen boluses were the least-validated class at 7% 9.

The economic consequences are likely downstream of this. A study of farm accounting data from 217 dairies over six years, using a within-farm before-and-after design, concluded: “Productivity did not change after investment in sensor systems on dairy farms. Furthermore, no technical change was found after investment in sensor systems, suggesting that the potential technological improvement claimed by producers of sensor systems does not materialize on dairy farms.” 8.

The case for in-line sensing

Most commercial monitoring hardware, such as ear tags, collars, ankle monitors, and rumen boluses measure the animal rather than the gland, and inherits characteristic weaknesses from their respective anatomical positions. Surface-mounted sensors are dominated by ambient conditions. In a controlled comparison of body surface sites against rectal temperature, ear surface temperature correlated with rectal temperature at r = 0.58 but with ambient temperature at r = 0.68-0.76, leading the authors to conclude that “body surface temperatures are primarily related to ambient temperature […] rather than rectal temperature” and that “the predictive value of body surface temperatures for heat stress or rectal temperature is low” 10. Ear-tag accelerometer accuracy has separately been shown to degrade under heat stress, with concordance for eating time falling from 0.53-0.54 under thermoneutral conditions to 0.17-0.34 at an elevated temperature-humidity index 11.

Rumen boluses avoid ambient interference but introduce a different confound. Drinking water depresses reticulorumen temperature by 2.3 °C on average, with recovery documented up to 103 minutes, and a recent review reports depressions of up to 8.5 °C requiring up to two hours to resolve 12,13. The physiological signal these devices are meant to detect is only 0.3-0.8 °C above baseline, so the confound is roughly an order of magnitude larger than the signal it must be separated from. Independent health-alert performance for boluses has been reported at 24-64% sensitivity with a positive predictive value of 22% 13. Boluses are also neither retrievable nor rechargeable, and remain in the body cavity for the life of the animal.

In-line sensing occupies a structurally different position. A dairy cow is milked two to three times per day, every day, for roughly 305 days per lactation. This is then 600-900 opportunities per year to measure a fluid produced by the exact organ of interest, at zero incremental animal handling and zero labor cost. No other diagnostic access point in animal agriculture has that property.

Struggles with in-line conductivity

Milk conductivity is the oldest in-line mastitis signal and it has a poor reputation, with published field sensitivities clustering in the range of 25-54%. This seems however, to be an artifact of poor measurement and not an invalid biological mechanism. The underlying physiology, described in Section 1.3, is sound: healthy quarters sit near 5.3 mS/cm, subclinically infected quarters near 5.75, and clinically infected quarters near 6.73 32. There are largely four main engineering failures that account for the disparity between a valid biological pathway and accurate sensing.

First, carryover. Incumbent in-line designs route milk through a chamber that retains fluid between measurements. Retained milk from the previous animal or the previous phase of the milking contaminates the foremilk, which is the fraction that carries the earliest and strongest disease signal. Second, electrode drift. A two-electrode conductivity measurement passes its sense current through the electrode–milk interface, where protein and fat films accumulate. The measurement therefore reports the state of the electrode surface as much as the state of the milk, and it degrades continuously over a device’s service life. That deployed in-line measurement noise is large is directly demonstrated: evaluated against laboratory reference, one commercial in-line sensor showed 42% of within-episode variance attributable to measurement noise rather than to the cow 35. Third, no fat correction. Milk fat globules are electrically insulating; current is forced to route around them, and measured conductivity falls as a predictable function of fat fraction. Fat rises from roughly 1–2% in foremilk to 8–10% in strippings, producing an artifact within a single milking that is comparable in magnitude to the disease signal itself. An uncorrected conductivity trace is therefore substantially a fat trace. Fourth, fixed absolute thresholds. Conductivity “values can vary significantly between different animals, compromising the definition of thresholds for both healthy and non-healthy conditions” 31. A threshold that must span the whole population cannot resolve a difference smaller than the population’s spread. This is the same problem the inter-quarter evidence in Section 3.6 addresses.

Per-quarter sensing

Mastitis is a quarter-level disease. A cow has four functionally independent mammary glands, and an infection in one is diluted roughly fourfold by the time milk from all four combines. Nearly every in-line system in commercial use measures at the cow level or at best reports a composite, and the dilution is a direct loss of signal. The published evidence that quarter-level measurement is where the diagnostic power lives is strong.

The single most striking result comes from combining a cell-count measure with an inter-quarter conductivity ratio (comparing each quarter against the animal’s own other quarters rather than against a threshold). That combination reaches an area under the curve of 0.85, against 0.47–0.58 for conductivity measured conventionally 32. The comparison, not the analyte, produces most of the gain.

A quarter-level study in automatic milking herds found the same structure across several indicators: quarters whose conductivity differed from the udder average by no more than 2.5% carried 55% lower risk of elevated somatic cell count, and a milk production rate below 1.5 kg/h carried 2.26 times higher risk 33. And an intensive longitudinal study sampling ten cows across 42 consecutive milkings concluded directly that within-cow, quarter-to-quarter comparison detects deviation better than fixed population thresholds 34.

This is the statistical escape from the base-rate problem described in Section 3.2. A fixed threshold applied across a herd must accommodate the full between-animal variance of breed, parity, stage of lactation, diet and yield, all of which contribute to variance that dwarfs the disease signal. A quarter compared against its three siblings in the same animal at the same milking cancels all of it.

Mastitis treatment

The antimicrobial standard of care

Intramammary antibiotics remain the default treatment for clinical mastitis, usually administered empirically at the moment a case is identified and before any pathogen information is available. Approximately 60–70% of all antimicrobial doses administered on dairy farms are for mastitis 16,21, and in the most recent national US survey 50.7% of treated cows received third-generation cephalosporins 15, the class the European Union’s Regulation 2019/6 designates Category B, “Restrict.” Yet across 23 treatment trials, mean bacteriological cure was 69%, with no significant difference between treatments in sixteen of the twenty-three 22.

The following delineate pathogen classes:

  1. Escherichia coli: 97% bacteriological cure untreated versus 99% treated: effectively no attributable benefit.
  2. Klebsiella: 18% untreated versus 74% treated: very large benefit.
  3. Staphylococcus aureus: 33–43% cure, the worst of any pathogen; in a recent US and Canadian trial, “bacteriological cure did not vary between the treated and controls” 22,23.

Meanwhile, between 15% and 44% of clinical cases yield no bacterial growth at all, and one major study found that 32% of intramammary antibiotics were administered to cases that were bacteriologically negative before treatment 22.

Selective treatment

Culture-guided selective treatment halves antibiotic use with no difference in clinical cure, bacteriological cure, recurrence, somatic cell count, milk yield, or culling 24, a result confirmed by a thirteen-study systematic review and meta-analysis “with a high or moderate certainty of evidence” 25. Modeling of that protocol across a range of scenarios found net fiscal impact of €8.70 to €48.10 per case, and identified the cow-level inputs that determine whether it pays: clinical mastitis incidence, milking frequency, days out of the tank, and treatment history 26. It is unfortunate that culture takes 18-24 hours and requires trained handling of samples, which is likely why its adoption has not been widespread.

Anti-inflammatory therapy

As established in Section 1.3, a substantial share of the damage in mastitis is produced by the host inflammatory response rather than directly by the bacteria. Non-steroidal anti-inflammatory drugs are therefore used as adjunct therapy, and the evidence for them is stronger than the evidence distinguishing one antibiotic from another. A stochastic bio-economic model parameterized on trial data found a net economic benefit of €42 per case from adding meloxicam to standard clinical mastitis therapy. The modeled outcomes driving that result were substantial: conception rate 31% with meloxicam versus 21% without, calving-to-conception interval 132 versus 143 days, inseminations per conception 2.9 versus 3.7, and culling 12% versus 25% 41. Lactational milk production was essentially unchanged, at 8,441 versus 8,517 kg.

That is notably a larger and more consistent effect than the antibiotic comparisons in the same disease produce.

Non-antibiotic alternatives

The pressures pushing dairy away from antimicrobial therapy are structural and compounding. Regulatory restriction is tightening, with the EU’s Regulation 2019/6 prohibiting prophylactic group treatment and reserving key classes, and the US having moved remaining over-the-counter medically important antimicrobials to prescription status. Milk discard during withholding is the dominant direct cost of treating a case. Organic producers, a growing segment, cannot use antibiotics at all without permanently removing the animal from organic production. And the pathogen data in Section 4.1 show that a large fraction of treated cases derive no benefit from the drug in the first place.

The non-antibiotic alternatives that have been explored to date, including vaccines, bacteriophage, bacteriocins, immunomodulators, botanicals, ozone, and various intramammary preparations have produced a long literature and few durable commercial successes. The labor constraint deserves particular emphasis, because it is the failure mode most often overlooked. A treatment that works but requires a stockperson to locate, restrain, and manually treat individual animals on a fixed schedule is simply impractical for a dairyfarmer to implement. Producers who have trialled drug-free approaches frequently report exactly this pattern.

Open-loop solutions

In 2013, Rutten et al. published a review of dairy sensor technology, comprising 126 publications describing 139 sensor systems across four levels of maturity: (I) a technique that measures something about the cow; (II) an interpretation that converts that measurement into a status; (III) integration of that status with other information to produce advice; and (IV) a decision actually being made. Their finding was unambiguous: “Many studies presented sensor systems at levels I and II, but none did so at levels III and IV… No systems with integrated decision support models have been found.” For mastitis in particular, 92% of the reviewed work sat at Level II 1.

In 2026, this review was refreshed in light of emerging bovine technology. Researchers, applying the same framework, conducted a fresh review of 132 articles and 151 sensor systems, and produced a not so dissimilar distribution: Level I, 7.95%; Level II, 90.73%; Level III, 1.32%; Level IV, 0% 2. After thirteen years and substantial capital allocation, no dairytech systems exist that take action on diagnostic intelligence! The authors write that most systems “lack direct links to actionable decision support for farmers,” and “Information alone does not have much value; we need to combine this information with tailormade interventions.”

The consequence is measurable in the decisions farms actually make. Culling is the largest single cost bucket in clinical mastitis at 39-48% of per-case cost 18,19, and it remains largely unaided. As Lehenbauer and Oltjen observed, “culling decisions have an important influence on the economic performance of the dairy but are often made in a nonprogrammed fashion and based partly on the intuition of the decision maker” 29. In Canadian herds, 73.6% of culls are involuntary, with mastitis the second-largest named cause at 10.6% 28, meaning most culling is a reaction to a failure rather than a plan.

Perspective

The literature reviewed here suggests that the individual components of mastitis management are each better understood than their outcomes suggest.

Prevention is scientifically settled and operationally unmeasured. The single intervention with a clear positive economic return is post-milking teat disinfection, worth a 50% reduction in new intramammary infection 20,44, but is performed by hand, hundreds of times per shift, by a workforce with high turnover and short tenure, and its execution is not instrumented anywhere. Compliance was 48% immediately after supervised training 38. No system currently tells a farm what its compliance is on any given day.

Detection is limited less by transduction than by architecture. The ISO bar has stood unmet for three decades 3,4, but the reason is base rate and between-animal variance rather than any inability to sense the disease, and the published evidence indicates that within-cow, quarter-level referencing recovers most of the lost performance 32,33,34. The unit of measurement, not the analyte, is the constraint.

Treatment is effective for a minority of cases and applied to nearly all of them. Selective, information-guided treatment is proven superior and halves antimicrobial use 24,25, but it is gated on diagnostic information that arrives a day late. Anti-inflammatory adjunct therapy produces the most consistent outcome improvements in the disease 41, which points toward inflammation as a tractable therapeutic target and toward a drug-free, automated intervention as the modality the field is missing.

Decision support does not exist. Thirteen years and two reviews found zero systems operating at the level where a decision is made 1,2, while half of all mastitis cost sits in chronic cows that were flagged repeatedly and never resolved 27.

What the evidence supports, therefore, is not a better sensor, a better drug, or better training taken separately. Each of those has been tried and each has run into the same wall. The outputs are not closed in a loop. What is needed is a system in which prevention is continuously measured; in which detection operates per quarter and per milking, referenced within the animal; in which treatment is drug-free and automated enough to be applied consistently to everything detection finds; and in which the cases where all of that fails generate a specific instruction, rather than another alert.

Whether such a system outperforms its components is an empirical question, and answering it will require field data of a kind the industry has largely not published.

Conclusion

Mastitis is an inflammatory disease of the mammary gland whose damage is substantially self-inflicted by the host response, whose transmission occurs largely at milking, and whose incidence is only 3.1% heritable 42 and varies 27.5-fold between farms in the same state 14. It is, to a first approximation, a disease made by people.

The precision dairy industry spent a decade proving that cows can be measured. Two independent reviews thirteen years apart found that essentially none of that measurement reaches a decision 1,2. A within-farm study of 217 dairies found that productivity did not change after sensor investment 8. Farmers stop reading the alerts 6.

Meanwhile the disease itself is one in which therapeutics account for 8% of the cost 18, a third of treatments are administered to animals with no bacterial infection 22, the worst pathogen cures at 33–43% 23, and half the total cost sits in chronic cows that were detected repeatedly and never resolved 27.

References

  1. Rutten CJ, Velthuis AGJ, Steeneveld W, Hogeveen H (2013). Invited review: Sensors to support health management on dairy farms. Journal of Dairy Science 96(4):1928–1952.
  2. Sani MI, van der Voort M, Tekinerdogan B, Hogeveen H (2026). Invited Review — Application of precision livestock farming: challenges and opportunities. Animal Bioscience 39(4):250895.
  3. Hogeveen H, Kamphuis C, Steeneveld W, Mollenhorst H (2010). Sensors and Clinical Mastitis — The Quest for the Perfect Alert. Sensors 10(9):7991–8009.
  4. Bausewein M, Mansfeld R, Doherr MG, Harms J, Sorge US (2022). Sensitivity and Specificity for the Detection of Clinical Mastitis by Automatic Milking Systems in Bavarian Dairy Herds. Animals 12(16):2131.
  5. Post C, Rietz C, Büscher W, Müller U (2021). The Importance of Low Daily Risk for the Prediction of Treatment Events of Individual Dairy Cows with Sensor Systems. Sensors 21(4):1389.
  6. Eckelkamp EA (2018). On-Farm Utilization of Precision Dairy Monitoring: Usefulness, Accuracy, and Affordability. PhD dissertation, University of Kentucky.
  7. Bewley J (2017). Exploring the Potential of Precision Dairy Tools. WCDS Advances in Dairy Technology 29:285–304.
  8. Steeneveld W, Hogeveen H, Oude Lansink AGJM (2015). Economic consequences of investing in sensor systems on dairy farms. Computers and Electronics in Agriculture 119:33–39.
  9. Stygar AH, Gómez Y, Berteselli GV, Dalla Costa E, Canali E, Niemi JK, Llonch P, Pastell M (2021). A Systematic Review on Commercially Available and Validated Sensor Technologies for Welfare Assessment of Dairy Cattle. Frontiers in Veterinary Science 8:634338.
  10. Bakony M, Kovács L, Kézér LF, Jurkovich V (2023). The use of body surface temperatures in assessing thermal status of hutch-reared dairy calves in shaded and unshaded conditions. Frontiers in Veterinary Science 10:1162708.
  11. Weinert-Nelson JR, Werner J, Jacobs AA, Anderson L, Williams CA, Davis BE (2025). Effects of heat stress on the accuracy of an ear-tag accelerometer for monitoring rumination and eating behavior in dairy-beef cross cattle using an automated gold standard. Journal of Dairy Science 108(1):735–749.
  12. Cantor MC, Costa JHC, Bewley JM (2018). Impact of Observed and Controlled Water Intake on Reticulorumen Temperature in Lactating Dairy Cattle. Animals 8(11):194.
  13. Kovács L, Simon ÁG, Czirok M, Póti P, Jurkovich V (2026). Heat Stress Detection in Dairy Cattle Using Intraruminal Bolus Sensors. Animals 16:2516. Bolus health-alert performance figures also drawn from Hajnal É (2026), AgriEngineering 8(5):167.
  14. Gonçalves JL, de Campos JL, Steinberger AJ, Safdar N, Kates A, Sethi A, Shutske J, Suen G, Goldberg T, Cue RI, Ruegg PL (2022). Incidence and Treatments of Bovine Mastitis and Other Diseases on 37 Dairy Farms in Wisconsin. Pathogens 11(11):1282.
  15. USDA APHIS (2016). Milk Quality, Milking Procedures, and Mastitis on U.S. Dairies, 2014. NAHMS Dairy 2014, Report 2, #704.0916.
  16. Tommasoni C, Fiore E, Lisuzzo A, Gianesella M (2023). Mastitis in Dairy Cattle: On-Farm Diagnostics and Future Perspectives. Animals 13(15):2538.
  17. Rasmussen P, Barkema HW, Osei PP, Taylor J, Shaw AP, Conrady B, Chaters G, Muñoz V, Hall DC, Apenteng OO, Rushton J, Torgerson PR (2024). Global losses due to dairy cattle diseases: A comorbidity-adjusted economic analysis. Journal of Dairy Science 107(9):6945–6970.
  18. Rollin E, Dhuyvetter KC, Overton MW (2015). Cost of clinical mastitis in the first 30 days of lactation. AABP Conference Proceedings 48:236.
  19. Aghamohammadi M, Haine D, Kelton DF, Barkema HW, Hogeveen H, Keefe GP, Dufour S (2018). Herd-Level Mastitis-Associated Costs on Canadian Dairy Farms. Frontiers in Veterinary Science 5:100.
  20. Fadillah A, van den Borne BHP, Schukken YH, Poetri ON, Hogeveen H (2025). Cost-efficiency of mastitis control strategies on smallholder dairy farms. Journal of Dairy Science 108(8):8726–8741.
  21. Dhital B, Pudasaini R, Hsieh J-C, Pudasaini R, Chen Y-T, Chao D-Y, Chiang H-I (2026). One Health Perspective on Antimicrobial Resistance in Bovine Mastitis Pathogens — A Narrative Review. Antibiotics 15(1):84.
  22. Ruegg PL (2021). What Is Success? A Narrative Review of Research Evaluating Outcomes of Antibiotics Used for Treatment of Clinical Mastitis. Frontiers in Veterinary Science 8:639641.
  23. Kolar QK, Godden SM, Erskine RJ, Ruegg PL (2024). Confirmed bacteriological diagnosis and cure of nonsevere gram-positive clinical mastitis cases enrolled in a randomized clinical trial based on results of on-farm culture. JDS Communications 5(6):628–633.
  24. Lago A, Godden SM, Bey R, Ruegg PL, Leslie K (2011). The selective treatment of clinical mastitis based on on-farm culture results: I and II. Journal of Dairy Science 94(9):4441–4456 and 4457–4467.
  25. de Jong E, Creytens L, De Vliegher S, et al. (2023). Selective treatment of nonsevere clinical mastitis does not adversely affect cure, somatic cell count, milk yield, recurrence, or culling: A systematic review and meta-analysis. Journal of Dairy Science 106(2):1267–1286.
  26. Rico A, Armstrong D, Brady N, León L, Hancock A, Lago A (2024). Selective treatment of clinical mastitis: Assessment of the net cash impact on dairy farms under diverse scenarios — A European perspective. Journal of Dairy Science 107(12):10964–10979.
  27. Bonestroo J, Fall N, Hogeveen H, Emanuelson U, Klaas IC, van der Voort M (2023). The costs of chronic mastitis: A simulation study of an automatic milking system farm. Preventive Veterinary Medicine 210:105799.
  28. Dallago GM, Wade KM, Cue RI, McClure JT, Lacroix R, Pellerin D, Vasseur E (2021). Keeping dairy cows for longer: A critical literature review on dairy cow longevity in high milk-producing countries. Animals 11(3):808.
  29. Lehenbauer TW, Oltjen JW (1998). Dairy cow culling strategies: making economical culling decisions. Journal of Dairy Science 81(1):264–271.
  30. Gross JJ, Bruckmaier RM (2019). Review: Metabolic challenges in lactating dairy cows and their assessment via established and novel indicators in milk. Animal 13(S1):s75–s81.
  31. Martins SAM, Martins VC, Cardoso FA, Germano J, Rodrigues M, Duarte C, Bexiga R, Cardoso S, Freitas PP (2019). Biosensors for on-farm diagnosis of mastitis. Frontiers in Bioengineering and Biotechnology 7:186.
  32. Stanek P, Żółkiewski P, Januś E (2024). A review on mastitis in dairy cows research: current status and future perspectives. Agriculture 14(8):1292. Inter-quarter ratio and absolute conductivity figures as tabulated therein from Lukas et al. and Norberg et al. (2004), Journal of Dairy Science 87(4):1099–1107.
  33. Zucali M, Bava L, Tamburini A, Gislon G, Sandrucci A (2021). Association between Udder and Quarter Level Indicators and Milk Somatic Cell Count in Automatic Milking Systems. Animals 11(12):3485.
  34. Åkerstedt M, Forsbäck L, Larsen T, Svennersten-Sjaunja K (2011). Natural variation in biomarkers indicating mastitis in healthy cows. Journal of Dairy Research 78(1):88–96.
  35. Deng Z, Hogeveen H, Lam TJGM, van der Tol R, Koop G (2020). Performance of Online Somatic Cell Count Estimation in Automatic Milking Systems. Frontiers in Veterinary Science 7:221.
  36. Gruber S, Rienesl L, Köck A, Egger-Danner C, Sölkner J (2023). Importance of Mid-Infrared Spectra Regions for the Prediction of Mastitis and Ketosis in Dairy Cows. Animals 13(7):1193.
  37. Ózsvári L, Ivanyos D (2022). The use of teat disinfectants and milking machine cleaning products in commercial Holstein-Friesian farms. Frontiers in Veterinary Science 9:956843. doi:10.3389/fvets.2022.956843
  38. Rodriguez Z, Lopez-Benavides M, Gentilini MB, Ruegg PL (2025). Impact of training dairy farm personnel on milking routine compliance, udder health, and milk quality. Journal of Dairy Science 108(2):1615–1624.
  39. Cheng WN, Han SG (2020). Bovine mastitis: risk factors, therapeutic strategies, and alternative treatments — A review. Asian-Australasian Journal of Animal Sciences 33(11):1699–1713. doi:10.5713/ajas.20.0156
  40. Texas A&M University / National Milk Producers Federation (2015). Economic Impacts of Immigrant Labor on U.S. Dairy Farms. Commissioned report.
  41. van Soest FJS, Abbeloos E, McDougall S, Hogeveen H (2018). Addition of meloxicam to the treatment of bovine clinical mastitis results in a net economic benefit to the dairy farmer. Journal of Dairy Science 101(4):3387–3397. doi:10.3168/jds.2017-12869
  42. Gaddis KP (2021). Using Genomic Selection to Improve Mastitis Resistance. Council on Dairy Cattle Breeding / National Mastitis Council.
  43. Sordillo LM, et al. (2011). Immunopathology of Mastitis: Insights into Disease Recognition and Resolution. Journal of Mammary Gland Biology and Neoplasia 16:291–304. doi:10.1007/s10911-011-9230-4
  44. National Mastitis Council (2007). Postmilking Teat Disinfection. NMC Factsheet.
  45. Capuco AV, Bright SA, Pankey JW, Wood DL, Miller RH, Bitman J (1992). Increased susceptibility to intramammary infection following removal of teat canal keratin. Journal of Dairy Science 75(8):2126–2130.
  46. Paape MJ, Bannerman DD, Zhao X, Lee J-W (2002). Defense of the bovine mammary gland by polymorphonuclear neutrophil leukocytes. Journal of Mammary Gland Biology and Neoplasia 7(2):109–121.
  47. Anagnostopoulos A, Griffiths BE, Siachos N, Neary J, Smith RF, Oikonomou G (2023). Initial validation of an intelligent video surveillance system for automatic detection of dairy cattle lameness. Frontiers in Veterinary Science 10:1111057. doi:10.3389/fvets.2023.1111057
  48. CattleEye Ltd. Product documentation and GEA Farm Technologies product listing, accessed August 2026.
  49. Cattle Care Inc. Product documentation, cattle-care.com, accessed August 2026.
  50. Wieland M, Heuwieser W (2025). Monitoring of milking routines for dairy cows using a computer vision system: A diagnostic accuracy study. Journal of Dairy Science 108(9):9728–9740. doi:10.3168/jds.2025-26571