J. Dairy Sci. 84:590–599 American Dairy Science Association, 2001.

Analysis of an Outbreak of Streptococcus uberis Mastitis

R. N. Zadoks,* H. G. Allore,†H. W. Barkema,‡ O. C. Sampimon,‡Y. T. Grohn,† and Y. H. Schukken*†*Department of Farm Animal Health, Faculty of Veterinary Medicine,Utrecht University, Yalelaan 7, 3548 CL Utrecht, The Netherlands†Department of Population Medicine and Diagnostic Sciences,College of Veterinary Medicine, Cornell University, Ithaca, NY 14853, USA‡Department of Ruminant Health, Animal Health Service,PO Box 361, 9200 AJ Drachten, The Netherlands

ABSTRACT

An outbreak of Streptococcus uberis mastitis was de-scribed to gain insight into the dynamics of Strep. uberisinfections at a herd level. Data were obtained from alongitudinal observational study on a commercialDutch dairy farm with good udder health management.Quarter milk samples for bacteriological culture wereroutinely collected at 3-wk intervals from all lactatinganimals (n = 95 ± 5). Additional samples were collectedat calving, clinical mastitis, dry-off, and culling. Duringthe 78-wk observation period, 54 Strep. uberis infec-tions were observed. The majority of infections occurredduring a 21-wk period that constituted the disease out-break. The incidence rate was higher in quarters thathad recovered from prior Strep. uberis infection thanin quarters that had not experienced Strep. uberis infec-tion before. The incidence rate of Strep. uberis infectiondid not differ between quarters that were infected withother pathogens compared with quarters that were notinfected with other pathogens. The expected numberof new Strep. uberis infections per 3-wk interval wasdescribed by means of a Poisson logistic regressionmodel. Significant predictor variables in the model werethe number of existing Strep. uberis infections in thepreceding time interval (shedders), phase of the study(early phase vs. postoutbreak phase), and prior infec-tion status of quarters with respect to Strep. uberis, butnot infection status with respect to other pathogens.Results suggest that contagious transmission may haveplayed a role in this outbreak of Strep. uberis mastitis.(Key words: Streptococcus uberis, mastitis, out-break, modeling)

Abbreviation key: β = transmission parameter, theprobability per unit of time that an infectious quarter

Received June 5, 2000.Accepted October 23, 2000.Corresponding author: R. N. Zadoks; e-mail: R.N.Zadoks@

vet.uu.nl.

590

will infect a noninfected quarter, IR = incidence rate,PMTD = postmilking teat disinfection.

INTRODUCTION

Streptococcus uberis is a widely occurring causativeagent of mastitis in modern dairy herds. It is responsi-ble for the majority of clinical and subclinical cases ofmastitis in New Zealand (McDougall, 1998) and theUnited Kingdom (Hillerton et al., 1993), and ranksamong the most prevalent causes of mastitis in theUnited States (Hogan et al., 1989a) and the Nether-lands (Barkema et al., 1998). Progress has been madein the control of Streptococcus agalactiae, Streptococcusdysgalactiae, and Staphylococcus aureus mastitis, butthere has been little reduction in the incidence of Strep.uberis mastitis over the past 30 yr (Leigh, 1999).

Streptococcus uberis, like Escherichia coli, is consid-ered to be an environmental pathogen (Radostits et al.,1994). The primary reservoir of environmental patho-gens is the dairy cow’s environment, and exposure ofuninfected quarters to environmental pathogens canoccur at any time during the life of a cow (Smith et al.,1993). In contrast, the primary reservoir for contagiousmastitis is the cow, and exposure of uninfected mam-mary quarters to contagious pathogens is restricted tothe milking process (Radostits et al., 1994; Smith etal., 1993).

Diseases that are transmitted through individual-to-individual contact, such as contagious mastitis, can bedescribed by a mathematical model called the Reed-Frost model. This model assumes that the probabilityof infection for a susceptible individual depends on thenumber of infectious individuals to which it is exposed(Becker, 1989). For other disease agents, the probabilityof infection depends on characteristics of the susceptibleindividual and its environment, as is assumed to bethe case for environmental mastitis pathogens. In thissituation, the probability of exposure does not dependon the number of infectious individuals. A Greenwood

ANALYSIS OF STREPTOCOCCUS UBERIS OUTBREAK 591

model describes this in mathematical terms (Becker,1989). Comparison of the output of mathematical mod-els with observational data can be used to understandthe mode of transmission of infectious agents (DeJong, 1995).

Mathematical models can also be used to comparethe susceptibility of groups with specific characteristics,such as the absence or presence of other pathogens ora history of prior infection (Halloran, 1998; Lam et al.,1996). Minor pathogens were shown to reduce the riskof IMI with Strep. uberis (Lam, 1996). However, otherstudies indicated an increased risk of infection withenvironmental pathogens in quarters infected with mi-nor pathogens (Hogan et al., 1988). Recovery fromStrep. uberis IMI conferred protection against subse-quent reinfection in experimental studies (Hill, 1988).

The purpose of this paper is to describe the dynamicsof an outbreak of Strep. uberis mastitis with mathemat-ical models. Susceptibility of quarters without a historyof Strep. uberis infection is compared with susceptibilityof quarters that recovered from Strep. uberis infection,and the susceptibility of uninfected quarters is com-pared with susceptibility of quarters infected with otherpathogens. Finally, fit of a model that assumes the num-ber of new infections to be a function of the numberof existing infections is compared with a model thatassumes the number of new infections to be indepen-dent of the number of existing infections.

MATERIALS AND METHODS

Data Collection

Data were obtained from a longitudinal observationalstudy (from June 1997 to December 1998) in a commer-cial Dutch dairy herd. Cows were mainly Holsteins,partly cross-bred with Dutch Friesians or Meuse-Rhine-Yssel cows. In winter, animals were housed in a free-stall barn with a concrete slatted floor and cubicles withwood shavings as bedding material. Lactating cowswere on pasture during summer. Cows were milkedtwice daily in a two × four open tandem parlor. Milkinghygiene included regular monitoring of milking ma-chine function, use of individual paper towels and, atthe start of the study, postmilking teat disinfection.Predipping was not practiced, as it is illegal in theNetherlands. Blanket dry cow treatment was used, ashad been herd practice for a number of years.

Quarter milk samples were collected every 3 wk fromall lactating animals (n = 95 ± 5), using aseptic tech-nique. Additional quarter milk samples were collectedby the farmer at calving (prior to first contact with themilking machine), dry-off, culling, and in the case ofclinical mastitis (any visual abnormality of milk or ud-der, with or without systemic signs of disease). Samples

Journal of Dairy Science Vol. 84, No. 3, 2001

were stored at −20°C until processing. Within 3 wk ofcollection, 0.01 ml of milk was cultured and bacterialspecies were identified according to National MastitisCouncil standards (Harmon et al., 1990). Colony countswere recorded for each bacterial species. Up to 10 cfuper plate, all colonies were counted, while higher countswere categorized as 11 to 50 cfu/plate, 51 to 200 cfu/plate, or >200 cfu/plate. Preliminary identification ofStrep. uberis was based on colony morphology and ae-sculin hydrolysis on Edward’s medium. Cultures wereconfirmed as Strep. uberis using the API 20 Strep sys-tem (Leigh, 1999). Shedding levels of Strep. uberis dur-ing the pre-outbreak and outbreak part of the observa-tion period (early phase) were compared with sheddinglevels during the postoutbreak period (late phase) bymeans of a chi-square test (Statistix, 1998).

Definition of Infection and Reinfection

A quarter was considered to have an IMI when ≥1000cfu/ml of a pathogen (major or minor) were culturedfrom a single sample, when ≥500 cfu/ml of a pathogenwere cultured from two out of three consecutive milksamples, when ≥ 100 cfu/ml were cultured from threeconsecutive milk samples, or when ≥ 100 cfu/ml werecultured from a clinical sample. Samples containingmore than three bacterial species were considered con-taminated and were not informative of IMI status. Apreviously infected quarter was considered recoveredfrom IMI for a species if none of the above definitionswere met and the sample was free of the pathogen(Barkema et al. 1998; Roberson et al., 1994; withslight modifications).

Based on IMI status, quarters were classified as be-longing to one of six compartments in a compartmentalmodel (Figure 1). Categorization was repeated eachtime samples were obtained. Quarters that were notinfected with any pathogen and that had not been in-fected with Strep. uberis at any stage during the studywere classified as “uninfected.” Quarters infected withStrep. uberis only were called “infected.” Quarters thatwere not currently infected with any pathogen but hadbeen infected with Strep. uberis earlier in the studywere considered “recovered-uninfected.” Quarters in-fected with any pathogen other than Strep. uberis werecalled an “other pathogen” infection if infection withStrep. uberis had not been observed at any point duringthe study. Quarters with mixed infection of Strep. ub-eris and other intramammary pathogens were termed“infected, including other pathogens.” Quarters thatwere infected with pathogens other than Strep. uberis,but had recovered from Strep. uberis IMI, were consid-ered “recovered and infected with other pathogens.” Theupper level of Figure 1 represents quarters that were

ZADOKS ET AL.592

Figure 1. Six compartment model describing Streptococcus uberisdynamics in a dairy herd. “Infected” represents infection with Strep.uberis. “Other pathogen” represents infection with any other intra-mammary pathogen. U1 denotes the compartments that were neverinfected with Strep. uberis (Uninfected, Other pathogen) and is con-trasted to R, which denotes the compartments that have recoveredfrom infection with Strep. uberis (Recovered-Uninfected, Recovered+ Other pathogen). U2 denotes the compartments that are not infectedwith Strep. uberis or any other pathogen (Uninfected, Recovered-Uninfected) and is contrasted to OP, which denotes the compartmentsthat are infected with other pathogens, but not with Strep. uberis(Other pathogen, Recovered + Other pathogen). I denotes the com-partments that are infected with Strep. uberis with or without otherpathogens (Infected, Infected + Other pathogens). Arrows indicatepossible transitions between compartments. Dotted arrows indicatetransitions with zero observations during an 18-month observationperiod in a 95-cow herd.

never infected with Strep. uberis, the middle level repre-sents quarters that are infected with Strep. uberis, andthe lower level represents quarters that recovered frominfection with Strep. uberis. The left side of Figure 1represents quarters that are not infected with otherpathogens, while the right side of the diagram showsquarters that are infected with other pathogens. It isimpossible to return to the upper level of the diagramfrom either the lower or the middle level.

Quarter-days at risk in a susceptible compartment(any compartment that represents quarters that arenot infected with Strep. uberis) or quarter-days infectedwith Strep. uberis were calculated based on startingpoint and endpoint of uninfected and infected periods.For IMI that were first detected at calving or as clinicalmastitis, sample date was assumed to be the date ofonset. For IMI that were last detected at dry-off or atculling, sample date was taken as endpoint of infection.For all other combinations, e.g., IMI starting duringlactation or ending between a clinical sample and theconsecutive routine survey sample, the midpoint of the

Journal of Dairy Science Vol. 84, No. 3, 2001

last negative and first positive sample was taken asstarting point, and the midpoint between last positiveand first negative sample was taken as endpoint of IMI.The terms “positive” and “negative” sample refer to IMIstatus of the sample.

Comparison of Incidence Rates

Incidence rates (IR) were calculated as the numberof new IMI per quarter-day at risk (Barkema et al.,1998, Hogan et al., 1988). When no IMI were observedin a group of quarters, the highest incidence rate thatis compatible with this observation was calculated as(Hanley et al., 1983):

1- maximum risk = α1/n [1]

with

α = probability of type 1 errorn = number of observations

To compare all quarters never infected with Strep. ub-eris to all recovered quarters, incidence rate in compart-ments denoted as U1 (top row in Figure 1), was con-trasted to incidence rate in compartments denoted asR (bottom row in Figure 1). To compare all quartersinfected with other pathogens to all quarters withoutother pathogens, the incidence rate in compartmentsdenoted as OP (right hand column in Figure 1) wascontrasted to the incidence rate in compartments de-noted as U2 (left hand column in Figure 1). Incidencerates were compared between compartments as de-scribed by Greenland and Rothman (1998) using a two-sided test. This statistical analysis of incidence ratesassumes an underlying Poisson probability model. Toassure validity of the Poisson model, the distributionof incidence data (number of new IMI) was examinedwith BestFit (1993). When the distribution of numberof new IMI was overdispersed, statistics for significancetesting were deflated by the square root of the over-dispersion factor (Cameron et al., 1998).

Regression Analysis

The probability of transmission of a disease can beexpressed in transmission parameter β, the probabilityper unit of time that an infectious quarter will infect anoninfected quarter. Values for β can be estimated fromDe Jong’s modification of a model originally describedby Becker (De Jong, 1995):

number of new IMIi per time interval [2]= βi * (I/N) * Si

ANALYSIS OF STREPTOCOCCUS UBERIS OUTBREAK 593

with

IMIi = IMI in susceptible quarter from compartment i,I = number of infected quarters,

Si = number of susceptible quarters in compartmenti, and

N = total number of quarters.

Because incidence data (number of new IMI) are countdata, a Poisson regression model was used for analysis(Cameron et al., 1998). Values for βi were estimated bya modification of the generalized linear model with log-link and Poisson error described for Staph. aureus byLam et al. (1996):

ε [ln(IMI)] = ln(β′) + ln (S/N) + θ1 * ln(I) [3]+ θ2 * y + θ3 * Um + θ4 * y * Um

with

ε = expected value,IMI = number of new infections in susceptible

compartment,β′ = transmission parameter for model with

ln(S/N) as offset,S = size of susceptible compartment expressed

in quarterdays at risk,N = size of total population expressed in quart-

erdays lactating,θi = regression parameters,I = exposure expressed in quarterdays in-

fected in preceding time interval,y = dummy variable for phase (y = 0 for early

phase, y = 1 for late phase),Um = dummy variable for compartment U1 vs.

R (m = 1) or U2 vs. OP (m = 2), andθ4 * y * Um = interaction term for compartment and

phase.

At first, ln(S/N) was used as model offset. If ln(I) wasfound to be a significant predictor variable, ln(S*I/N)was used as model offset, replacing ln(S/N) + θ1 * ln(I),and ln(β′) was replaced by ln(βi) in model [3]. In thismodel, infected quarters were assumed to be infectious,as the definition of infection was based on shedding ofthe bacteria of interest.

The mastitis outbreak reported by Lam et al. (1996)was best described by a model that stratified the analy-sis by start (“outbreak”) versus remainder (“steadystate”) of the observation period. Therefore, a variablefor phase of study was included in model [3]. To deter-mine the starting point of the late phase for the Strep.uberis outbreak described here, models were fitted witheach possible sampling interval as start of the postout-

Journal of Dairy Science Vol. 84, No. 3, 2001

Figure 2. Prevalence (infected proportion of lactating quarter-days, right vertical axis) and incidence (number of new infectionsand reinfections, left vertical axis) of Streptococcus uberis IMI per 3-wk sampling interval in a 95-cow herd, showing new infections atquarter level in compartments U1, R and at calving. U1 = neverinfected with Strep. uberis, with or without presence of other patho-gens; R = recovered from prior Strep. uberis infection, with or withoutpresence of other pathogens.

break period. Models with lowest deviance were consid-ered the best models.

The model was run to compare compartments U1 toR and U2 to OP, respectively. Compartment sizes wereexpressed in number of quarter-days at risk. Signifi-cance of predictor variables in the regression analysiswas declared at α = 0.01. A conservative level was cho-sen because the correlation between repeated observa-tions (quarters within cow) could not be corrected for.Analysis was performed using statistical software(Statistix, 1998).

RESULTS

Descriptive Results

A total of 11,932 quarter milk samples were obtained,out of which 686 were collected at calving, 263 fromcows with clinical mastitis, 526 at dry-off, 146 at culling,and 91 as extra samples before treatment of subclinicalIMI with antibiotics. Samples were missing for eightcows at calving, and for 17 cows at culling. Cows witha blind quarter did occur (n = 10). Eighty-eight sampleswere excluded from the analysis because of contamina-tion (0.2% of all samples).

During the 18-mo observation period, 39 new Strep.uberis infections and seven reinfections were observedin lactating animals, and four infections were observedin quarters at calving (Figure 2, Appendix 1). The ma-jority of cases occurred in a limited time interval, cov-ering sampling intervals 9 through 15 (from November1997 to April 1998). This 21-wk period is called the

ZADOKS ET AL.594

Table 1. Probability distribution of number of new Streptococcus uberis IMI within susceptible compartmentsper 3-wk interval, based on 26 observed intervals. U1 = never infected with Strep. uberis, with or withoutother pathogens at sampling prior to detection of Strep. uberis IMI; R = recovered from infection with Strep.uberis, with or without other pathogens at sampling prior to detection of Strep. uberis IMI; U2 = not infectedwith Strep. uberis or other pathogens at sampling prior to detection of Strep. uberis IMI, irrespective ofrecovery history; OP = infected with other pathogens at sampling prior to detection of Strep. uberis IMI,irrespective of recovery history.

OverdispersionCompartment Distribution Mean Variance factor

U1 NegBin1(1.00,0.40) 1.50 3.75 2.5R NegBin(5.00,0.95) 0.27 0.28 1.1

Poisson(0.27) 0.27 0.27U2 NegBin(1.00,0.68) 0.46 0.67 1.5

Poisson(0.46) 0.46 0.46OP NegBin(1.00,0.43) 1.31 3.00 2.3

1NegBin = Negative binomial distribution with shape parameters.

outbreak period. Six weeks after the outbreak hadended, at sampling 18, most infected animals (10 outof 14) were separated from uninfected animals, andinfected animals were treated with antibiotics. Treatedanimals were added to the main herd again at samplingnumber 20, irrespective of bacteriological cure.

Infections with Strep. uberis were observed in unin-fected quarters, in quarters infected with other patho-gens and in recovered quarters infected with otherpathogens (seven reinfections in six quarters from fivecows), but not in recovered-uninfected quarters (Appen-dix 1). With 1315 recovered-uninfected quarter-days atrisk, the 99% confidence interval for the maximum riskof infection per recovered-uninfected quarter-day atrisk (maximum incidence rate) is 0 ≤ IR ≤ 0.0035. Inci-dence rates for compartments U1, R, U2, and OP areshown in Table 2.

Infected quarters that were present at the start ofthe study (n = 4) did not contribute to incidence, butdid contribute to prevalence of infection. The prevalenceof infection, expressed in number of infected quarter-days (with or without other pathogens) as a proportionof the total number of lactating quarter-days per sam-pling interval, is included in Figure 2.

Shedding levels did not differ between early (interval1 to 15) and late phase (interval 16 to 26) of the observa-tion period (shedding levels ≤10, 11 to 50, 51 to 200,and >200 cfu/ml; χ2 = 2.72, P = 0.44).

Other Pathogens

Infections with other pathogens were predominantlycaused by corynebacteria (81.9% of IMI with otherpathogens) and coagulase-negative staphylococci(26.0%). Other minor pathogen infections (enterococci,Bacillus spp., non-dysgalactiae and non-uberis strepto-cocci, micrococci) and major pathogen infections (Strep-tococcus dysgalactiae, Escherichia coli, Staphylococcus

Journal of Dairy Science Vol. 84, No. 3, 2001

aureus, Arcanobacterium pyogenes) accounted for 1.7and 3.1% of other pathogen-infected samples, respec-tively. Total of percentages adds up to more than 100%because mixed infections occurred in 13.1% of samplesfrom quarters infected with other pathogens. Strepto-coccus agalactiae was never isolated.

Comparison of Incidence Rates

Incidence data (number of new IMI in each time inter-val) for compartments U1 and OP were best describedby a negative binomial distribution or a geometric dis-tribution. Incidence data for compartments R and U2were adequately described by a negative binomial, geo-metric, or Poisson distribution. Distribution parame-ters and overdispersion factors are shown in Table 1.Overdispersion factors for U1 and OP were used to de-flate statistics for significance testing.

Incidence rate (number of new IMI per quarter-day atrisk) in quarters that had recovered from Strep. uberisinfection was 7.5 times as high as incidence rate inquarters that had never experienced Strep. uberis infec-tion (P < 0.001, Table 2). Incidence rate in quartersinfected with other pathogens was 1.3 times as high asincidence rate in quarters that were not infected withother pathogens. The difference was not significant (P> 0.2, Table 2).

Regression Analysis

Model deviance was calculated with a range of start-ing points for the postoutbreak period in model [3].Model fit was best when the late phase started at sam-pling interval 16, both when contrasting U1 to R andwhen contrasting U2 to OP.

For comparison of compartments U1 and R, phase ofstudy was a significant predictor variable. Other sig-nificant predictor variables were compartment of origin

ANALYSIS OF STREPTOCOCCUS UBERIS OUTBREAK 595

Table 2. Number and rate of Streptococcus uberis IMI per susceptiblecompartment of origin. U1 = never infected with Strep. uberis, withor without other pathogens at sampling prior to detection of Strep.uberis IMI; R = recovered from infection with Strep. uberis, with orwithout other pathogens at sampling prior to detection of Strep. uberisIMI; U2 = not infected with Strep. uberis or other pathogens at sam-pling prior to detection of Strep. uberis IMI, irrespective of recoveryhistory; OP = infected with other pathogens at sampling prior todetection of Strep. uberis IMI, irrespective of recovery history.

Susceptible Number of new Rate of newcompartment infections1 infections2

U1 39 0.20R 7 1.50U2 12 0.19OP 34 0.25Nonlactating 4 —3

1Total number of new infections detected during an 18-month obser-vation period.

2(Number of new Strep. uberis infections:susceptible quarterdaysat risk) * 1000.

3Three out of four infections in nonlactating animals were detectedin heifers at calving. Number of quarter-days at risk for heifers wasunknown. Hence, infection rate was not calculated. Infections in non-lactating animals were not included in any of the mathematical mod-els described in this paper.

and number of quarter-days infected in the preceding3-wk interval (Table 3). The interaction between com-partment and phase of study was not significant andwas omitted from the model. Although raw data wereoverdispersed, modeled data could be described by aPoisson distribution (deviance 61.46, P = 0.06, df = 46).

For comparison of U2 and OP, phase of study andnumber of quarter-days infected during the preceding

Table 3. Estimates, standard errors, and P-values for ln(β′) and regression coefficients of explanatoryvariables in Poisson logistic regression model ε [ln(IMI)] = ln(β′) + ln (S/N) + θ1 * ln(I) + θ2 * y + θ3 * Umwhere ε = expected value, IMI = number of new infections with Streptococcus uberis in current time interval,β′ = transmission parameter for model with ln(S/N) as offset, S = number of quarter-days susceptible incurrent time interval, N = total number of quarter-days in current time interval, I = number of quarter-days infected in preceding time interval, y = dummy variable for phase (y = 0 for early phase, y = 1 for latephase), Um = dummy variable for compartment (Um = 0 for R or OP, Um = 1 for U1 or U2) and θi = regressioncoefficient.

Model Parameter1 Estimate Standard error P-value

U1 vs. R2 ln(β′) 0.11 0.78 0.8793ln(I) 0.68 0.15 <0.0001study phase −1.55 0.35 <0.0001compartment −2.06 0.42 <0.0001

U2 vs. OP3 ln(β′) −1.84 0.67 0.0067ln(I) 0.71 0.15 <0.0001study phase −1.47 0.35 <0.0001compartment −0.36 0.34 0.2846

1For ln(β′) the estimated value is shown, for other parameters estimated values of the regression coefficientsare shown. No value is given for ln(S/N) because this was used as the model offset.

2U1 = never infected with Strep. uberis, with or without other pathogens at sampling prior to detectionof Strep. uberis IMI; R = recovered from infection with Strep. uberis, with or without other pathogens atsampling prior to detection of Strep. uberis IMI.

3U2 = not infected with Strep. uberis or other pathogens at sampling prior to detection of Strep. uberisIMI, irrespective of recovery history; OP = infected with other pathogens at sampling prior to detection ofStrep. uberis IMI, irrespective of recovery history.

Journal of Dairy Science Vol. 84, No. 3, 2001

Table 4. Estimates of transmission parameters β (probability perunit of time that an infectious quarter will infect a noninfected quar-ter; based on linear regression model with ln(S*I/N) as offset) foruninfected quarters without infection history (U1), for quarters recov-ered from Streptococcus uberis IMI (R), for quarters without otherpathogens (U2) and for quarters with other pathogens (OP) duringearly (intervals 1–15) and late phase intervals (16–26) of the study.Within a model, values with different superscript are different (P <0.001).

Model Compartment βearly phase βlate phase

U1 vs. R U1 0.033a 0.005b

R 0.246c 0.040d

U2 vs. OP U2 0.029a 0.005f

OP 0.041e 0.007f

3-wk interval were significant predictor variables (Ta-ble 3). The transmission parameter estimate for com-partment U2 was not significantly different from thetransmission parameter estimate for compartment OP.Again, the interaction between compartment and phaseof study was not significant, and was omitted from themodel. Modeled data were slightly overdispersed (devi-ance 68.93, P = 0.02, df = 46).

Estimates for transmission parameters, based onmodel [3] with ln(S*I/N) as offset, are shown in Table 4.

DISCUSSION

In this paper we describe an outbreak of Strep. uberismastitis. Susceptibility of groups or compartments ofquarters at risk is compared with the use of mathemati-cal models. In addition, the hypothesis is tested that

ZADOKS ET AL.596

a model that includes number of infected quarters aspredictor variable (Reed-Frost model) describes the out-break better than a model that does not include thisfactor (Greenwood model).

Quarters of a cow were treated as independent units.Exposure to infected quarters within the same cow canbe considered of minor importance, because proportionsof new IMI resulting from cross-infections betweenquarters of one cow are small compared with new IMIresulting from transmission between cows (Baxter etal., 1992; McDougall, 1998). Quarters within a cow areclustered with respect to exposure to sources outside thecow itself and with respect to susceptibility to infection.Barkema et al. (1997) provide pathogen-specific esti-mates for correlation of quarters within cows, and giveexamples of how to correct for this correlation in clinicaltrials and cross-sectional prevalence studies. This cor-rection does not apply to regression analysis of inci-dence data in longitudinal studies. As an alternativeto analysis of data at quarter level, analysis at cow levelhas been considered. This alternative would not allowfor classification of cows as belonging in one compart-ment only, because infection status may differ per quar-ter. Therefore, we preferred analysis at quarter level.

Number of new IMI with Strep. uberis varied overtime, with zero cases during the first months of thestudy period, and high numbers during an outbreakthat lasted 21 wk. Although no IMI with Strep. uberiswere observed in recovered quarters that were not in-fected with other pathogens, the 99% confidence inter-val for incidence rate in this group covers the observedincidence rates for compartments U1, R, U2, and OP.Hence, it cannot be concluded that recovered uninfectedquarters are fully resistant to reinfection. Incidencerates of Strep. uberis IMI in our study (number of newIMI per quarter-day at risk) were in the same range asincidence rates reported by Todhunter et al. (1995) andHogan et al. (1988) for environmental streptococci andby Barkema et al. (1998) for clinical Strep. uberismastitis.

Based on comparison of incidence rates and on trans-mission parameters for transmission to uninfected (U1)and recovered quarters (R), it is concluded that suscep-tibility of quarters that recovered from Strep. uberisinfection is higher than susceptibility of quarters thatnever experienced Strep. uberis infection. Hill (1988)described reduced susceptibility to infection after previ-ous exposure to Strep. uberis in an experimental situa-tion. The same strain of Strep. uberis was used forprimary and subsequent infections. Finch et al. (1997)demonstrated that vaccination was less effectiveagainst strains other than the immunizing strain. Thedifference between Hill’s findings and our study maybe a result of differences in bacterial strains, in cow

Journal of Dairy Science Vol. 84, No. 3, 2001

susceptibility or in study type. Use of quarter-days atrisk as denominator in incidence calculations does nottake into account that repeated observations within aquarter over time are correlated. Some cows or quartersmay be more susceptible to IMI than others. This effectcan be more pronounced in compartment R than incompartment U1, because the number of quarter-daysat risk in compartment R is smaller. However, use ofperson-time at risk is routine practice in epidemiology(Greenland et al., 1998), and comparison of incidencerates for mastitis pathogens is usually based on timeat risk expressed in cow-days or quarter-days (Hoganet al., 1988; Lam et al., 1996; Todhunter et al., 1995).Despite its shortcomings, we therefore consider the cal-culation of incidence rates justified. Poor sensitivity ofbacterial culture is an issue when dealing with detec-tion of Staph. aureus. For isolation of Strep. uberis poorculture sensitivity is hardly ever reported as a problem.Because of that, and because all quarters yielded atleast two Strep. uberis negative samples before reinfec-tion was observed, it is unlikely that apparent reinfec-tions were a result of persistent infections that wentundetected in the period between Strep. uberis iso-lations.

The majority of infections in the study herd that werenot caused by Strep. uberis were attributable to cory-nebacteria and coagulase-negative staphylococci. Thosespecies are commonly grouped under the collective de-nominator “minor pathogens” (Radostits et al., 1994).In our analysis, minor pathogens and the small propor-tion of major pathogens other than Strep. uberis wereall treated as one group. When comparing quarters thatwere not infected with other pathogens (U2) to quartersinfected with other pathogens (OP), no difference inincidence rate or transmission parameters was detectedfor Strep. uberis. Lam (1996) found the same resultfor quarters infected with Staphylococcus species, butobserved a decreased risk of Strep. uberis IMI in quar-ters infected with Corynebacterium bovis. Hogan et al.(1988) found the rate of environmental streptococcalmastitis to be significantly higher in quarters infectedwith Corynebacterium bovis or Staph. spp. comparedwith uninfected quarters. An explanation for the dis-crepancy between results obtained from different stud-ies could be the fact that Lam used a matched case-control analysis, comparing quarters within cows tocorrect for possible confounding by cow effect. Resultspresented in this paper and results obtained by Hoganet al. (1988) are based on observational studies withoutcorrection for cow or quarter effects.

Phase of study is a significant predictor variable forthe number of new IMI with Strep. uberis, with lowertransmission parameters during the late phase of theobservation period. Observational data are best de-

ANALYSIS OF STREPTOCOCCUS UBERIS OUTBREAK 597

scribed by the model when a dummy for postoutbreakperiod is included, starting with interval 16 as the firstpostoutbreak interval. A similar result has been foundby Lam et al. (1996) for an outbreak of Staph. aureusmastitis. Significance of study phase can be interpretedas a change in transmission parameter β at the end ofthe disease outbreak. This change may be due to achange in infectiousness of infected quarters or achange in susceptibility of exposed individuals (Hal-loran, 1998). Theoretically, a change in infectiousnesscould be caused by a change in shedding levels, or achange in pathogen population at the subspecies level.In our study, shedding levels did not differ between theearly and late phase of the observation period. Lam etal. (1996) postulated that a change in susceptibilityof exposed individuals may be the result of immunitydeveloped after short duration IMI. In their study, shortduration IMI could go unnoticed as a result of the sam-pling scheme. In the present study, sampling was morefrequent, and samples that contained >1000 cfu/ml onlyonce were considered to be infected to improve detectionof short duration IMI. Still, short-duration IMI thatoccurred between samplings may have gone unnoticed.Development of immunity in response to short durationIMI would be in disagreement with the increased sus-ceptibility in recovered quarters observed in this study,unless short-duration IMI and long-term IMI have dif-ferent pathogenesis and cure mechanisms.

A change in transmission rates could also be the re-sult of a change in herd contact structure or a changein probability of transmission upon contact (Halloran,1998). Segregation of animals is an example of a man-agement strategy that reduces contact between individ-uals. Postmilking teat disinfection (PMTD) is an exam-ple of a control procedure that reduces the probabilityof infection after contact. In the study herd, segregationof animals was used temporarily. Most Strep. uberisinfected cows were housed separately and milked lastfor 6 wk (10 out of 14 animals). Separation of animalsviolates the assumption of random mixing underlyingthe Reed-Frost and Greenwood model. However, thischange in management was temporary, and took placeseveral weeks after the end of the outbreak period.Hence, it does not explain or affect a change in trans-mission at the end of the outbreak itself. PMTD wasused during intervals 1 to 6 and 16 to 19, but discon-tinued during intervals 7 to 15 and 20 to 26 (exactdates unknown). When added to model [3], PMTD wasa significant predictor variable (P < 0.01) and did notaffect significance or direction of effect of other predictorvariables. Use of PMTD reduced transmission parame-ter β. PMTD may have contributed to the end of theoutbreak.

Journal of Dairy Science Vol. 84, No. 3, 2001

In all models used for calculation of transmissionparameters, the number of infected quarter-days in thepreceding 3-wk interval was a significant predictor vari-able, with higher exposure to shedders predictinghigher numbers of new IMI. This implies that the Reed-Frost assumption is more appropriate than the Green-wood assumption and suggests that Strep. uberis canbehave as a contagious pathogen. Although this mayseem contrary to the prevailing classification of Strep.uberis as an environmental pathogen, several authorsinclude the option of contagious transmission when de-scribing environmental streptococci (Leigh, 1999;Neave et al., 1969).

The characteristic setting environmental pathogensapart from contagious pathogens is that in additionto the possibility of contagious transmission throughexposure during milking time, there can be nonconta-gious transmission at other times and through exposureto other sources than the milking process. The presentstudy and other studies provide several arguments infavor of contagious transmission, in addition to environ-mental transmission. Arguments in favor of contagioustransmission from this study include the significanceof infection prevalence as predictor for the number ofnew IMI, and the decrease in predicted number of newIMI during periods that PMTD was used. PMTD isthought to kill bacteria that are transmitted duringthe milking process, e.g., via teat cup liners, and thusreduce the incidence of IMI. For Staph. aureus, trans-mission via contaminated teat cup liners has been de-scribed (O’Shea, 1987). Using liner swabs, we detectedStrep. uberis in teat cup liners after milking of Strep.uberis shedding cows, and after milking of up to twononshedding cows following a shedding cow (data notshown). This indicates that transmission of Strep. ub-eris via teat cup liners is possible, which may explainpart of the effect of PMTD. An outbreak of Strep. uberismastitis described by Cattell (1996) also illustrates therole of infected quarters as a source of infection.

Other observations in our study are in agreementwith an environmental mode of transmission. FourStrep. uberis infections were observed in nonlactatinganimals, out of which three were heifers. Heifer masti-tis is a well-known phenomenon (Oliver et al., 1983).Because heifers have not been exposed to the milkingmachine, contagious transmission during the milkingprocess cannot explain infections in preparturientmammary glands. One possible source of exposure tomastitis pathogens in the environment is bedding mate-rial. Cows in this study were housed in a free stall withwood shavings for bedding. In wood-based materialsGram-negative bacteria tend to predominate overGram-positive bacteria (Rendos et al., 1975), but num-bers of Strep. uberis in bedding can increase with in-

ZADOKS ET AL.598

creased organic contamination (Hogan et al., 1989b).Bedding management may have played a role in thisoutbreak of Strep. uberis mastitis, but attempts to cul-ture Strep. uberis from wood shavings were unsuccess-ful (results not shown) and the role of bedding materialcannot be proven.

The middle ground between the so-called contagiousand environmental modes of transmission exists. Thefact that a model including the number of infected indi-viduals describes incidence data better than a modelwithout the number of existing IMI does not prove acausative role of infected individuals, nor a specificmechanism of transmission. This is an inherent limita-tion of mathematical models (De Jong, 1995). It is possi-ble that the number of infected quarters merely reflectsunobserved changes in environmental conditions or ex-posure to an increasing and then decreasing environ-mental Strep. uberis load. It is also possible that in-fected quarters themselves are indeed the main sourceof exposure, but that transmission takes place via theenvironment, e.g., when infected cows contaminatestalls through milk leaking. Another transmissionmechanism could be via flies, as shown for Staph.aureus by Owens et al. (1998). If cows are the majorsource of exposure, but transmission is not via the milk-ing process, the dynamics of Strep. uberis infectionsat herd level can not be classified in the traditional“contagious” versus “environmental” dichotomy.

In principle, strain typing of Strep. uberis isolatescould support or contradict the notion of conta-giousness. Thus, mathematical and molecular tech-niques could supplement each other in the approachof an epidemiological problem. Many techniques havebeen used for typing of strains within the species Strep.uberis (Jayarao et al., 1992; Leigh, 1999) and a widevariety of strains has been shown to cause intramam-mary infections. In the case of contagious transmission,all IMI that are part of an outbreak should be attribut-able to a limited number of strains. Not all isolatesfrom this outbreak were available for strain typing.However, if strain typing in future outbreaks wouldshow that isolates belong predominantly to one or afew strains, we can agree with Cattell’s suggestion(1996) that Strep. uberis, like Strep. dysgalactiae(Smith et al., 1993), can be characterized as intermedi-ate between contagious and environmental.

CONCLUSION

Udder quarters that recovered from prior Strep. ub-eris IMI had a higher incidence rate of Strep. uberisIMI than quarters that had not experienced Strep. ub-eris IMI before. The incidence rate of Strep. uberis IMIdid not differ between quarters that were infected with

Journal of Dairy Science Vol. 84, No. 3, 2001

other pathogens and quarters that were not infectedwith other pathogens. A prediction model for the num-ber of new Strep. uberis infections that included thenumber of existing infections fit the observed data bet-ter than a model that did not include exposure to ex-isting infections. Contagious spread of the pathogenmay have played a role in this outbreak of Strep. ub-eris mastitis.

ACKNOWLEDGMENTS

The authors thank the farmer and his family for theircooperation in this study, and Patrick van Valkengoedfor invaluable technical support. Elizabeth Halloranand Carlos Castillo-Chavez are acknowlegded for theirsuggestions regarding mathematical analyses. Thisproject was partially funded by Intervet InternationalBV (The Netherlands).

REFERENCES

Barkema, H. W., Y. H. Schukken, T.J.G.M. Lam, D. T. Galligan, M.L. Beiboer, and A. Brand. 1997. Estimation of interdependenceamong quarters of the bovine udder with subclinical mastitis andimplications for analysis. J. Dairy Sci. 80:1592–1599.

Barkema, H. W., Y. H. Schukken, T.J.G.M. Lam, M. L. Beiboer, H.Wilmink, G. Benedictus, and A. Brand. 1998. Incidence of clinicalmastitis in dairy herds grouped in three categories by bulk milksomatic cell count. J. Dairy Sci. 81:411–419.

Baxter, J. D., G. W. Rogers, S. B. Spencer, and R. J. Eberhart. 1992.The effect of milking machine liner slip on new intramammaryinfections. J. Dairy Sci. 75:1015–1018.

Becker, N. G. 1989. Analysis of infectious disease data. Chapmanand Hall, London, United Kingdom.

BestFit for Windows. 1993. Version 2.0d. Palisade Corporation, New-fields, NY.

Cameron, A. C. and P. K. Trivedi. 1998. Regression analysis of countdata. Cambridge University Press, Cambridge, United Kingdom.

Cattell, M. B. 1996. An outbreak of Streptococcus uberis as a conse-quence of adopting a protocol of no antibiotic therapy for clinicalmastitis. Pages 123–127 in Proc. 35th Nat. Mastitis Council Ann.Meeting, Nashville, TN. Natl. Mastitis Counc., Inc., Arlington,VA.

De Jong, M.C.M. 1995. Mathematical modelling in veterinary epide-miology: why model building is important. Prev. Vet. Med.25:183–193.

Finch, J. M., Winter, A., Walton, A., and J. A. Leigh. 1997. Furtherstudies on the efficacy of a live vaccine against mastitis causedby Streptococcus uberis. Vaccine 15:295–304.

Greenland, S. and K. J. Rothman. 1998. Introduction to categoricalstatistics. Pages 231–252 in Modern Epidemiology. K. J. Rothmanand S. Greenland, eds. 2nd edition. Lippincott-Raven Publishers,Philadelphia, PA.

Halloran, M. E. 1998. Concepts of Infectious Disease Epidemiology.Pages 529–554 in Modern Epidemiology. K. J. Rothman and S.Greenland, eds. 2nd ed. Lippincot-Raven Publishers, Philadel-phia, PA.

Hanley, J. A., and A. Lippman-Hand. 1983. If nothing goes wrong,is everything all right? Interpreting zero numerators. J. Am. Med.Assoc. 249:1743–1745.

Harmon, R. J., Eberhart, R. J., Jasper, D. E., Langlois, B. E. andWilson, R. A. 1990. Microbiological procedures for the diagnosis ofbovine udder infection. Natl. Mastitis Council, Inc. Arlington, VA.

Hill, A. W. 1988. Protective effect of previous intramammary infectionwith Streptococcus uberis against subsequent clinical mastitis inthe cow. Res. Vet. Med. 44:386–387.

ANALYSIS OF STREPTOCOCCUS UBERIS OUTBREAK 599

Hillerton, J. E., M.F.H. Shearn, R. M. Teverson, S. Langridge, andJ. M. Booth. 1993. Effect of pre-milking teat dipping on clinicalmastitis on dairy farms in England. J. Dairy Res. 60:31–41.

Hogan, J. S., K. L. Smith, D. A. Todhunter, and P. S. Schoenberger.1988. Rate of environmental mastitis in quarters infected withCorynebacterium bovis and Staphylococcus species. J. Dairy Sci.71:2520–2525.

Hogan, J. S., K. L. Smith, K. H. Hoblet, P. S. Schoenberger, D. A.Todhunter, W. D. Hueston, D. E. Pritchard, G. L. Bowman, L. E.Heider, B. L. Brockett, and H. R. Conrad. 1989a. Field survey ofclinical mastitis in low somatic cell count herds. J. Dairy Sci.72:1547.

Hogan, J. S., K. L. Smith, D. H. Hoblet, D. A. Todhunter, P. S.Schoenberger, W. D. Hueston, D. E. Pritchard, G. L. Bowman,L. E. Heider, B. L. Brokett, and H. R. Conrad. 1989b. Bacterialcounts in bedding materials used on nine commercial dairies. J.Dairy Sci. 72:250–258.

Jayarao, B. M., B. J. Bassam, G. Caetano-Anolles, P. M. Gresshoff,and S. P. Oliver. 1992. Subtyping of Streptococcus uberis by DNAamplification fingerprinting. J. Clin. Microbiol. 30:1347–1350.

Lam, T.J.G.M. 1996. The effect of natural infections with minor patho-gens on susceptibility to natural infections with major pathogensin the bovine mammary gland. Pages 131–145 in Dynamics ofbovine mastitis—a field study in low somatic cell count herds.Ph.D. thesis, Utrecht University, The Netherlands.

Lam, T.J.G.M., M.C.M. de Jong, Y. H. Schukken, and A. Brand. 1996.Mathematical modeling to estimate the efficacy of postmilkingteat disinfection in split-udder trials of dairy cows. J. Dairy Sci.79:62–70.

Leigh, J. A. 1999. Streptococcus uberis: a permanent barrier to thecontrol of bovine mastitis? Vet. J. 157:225–238.

APPENDIX

Appendix 1. Data used for analysis in this paper. For each 3-wk sampling interval the number of quarter-days at risk or infected is shownfor each compartment, as well as the number of new Streptococcus uberis IMI observed in quarters originating from each of the foursusceptible compartments. op = other pathogens (IMI with bacteria other than Strep. uberis).

Quarter-days at risk or infected with Strep. uberis per compartment New Strep. uberis infections per susceptible compartment

Other Infected Recovered Other RecoveredInterval Uninfected pathogens Infected + op Recovered + op Total Uninfected pathogens Recovered + op Total

1 3103.5 4804.5 34.5 1.5 1.5 10.5 7956 0 0 0 0 02 2440 5110 21 0 0 21 7592 0 0 0 0 03 2376.5 5229.5 21 0 0 21 7648 0 0 0 0 04 2564 5412 21 0 0 21 7978 0 0 0 0 05 2565.5 5044.5 21 0 9 30 7670 0 0 0 0 06 2563 5069 21 0 31.5 31.5 7716 0 0 0 0 07 2528.5 4859.5 6 7 49 6 7456 0 0 0 0 08 2750.5 4570.5 0 0 63 0 7384 0 0 0 0 09 2661.5 4577.5 0.5 21 63 20.5 7344 0 2 0 0 2

10 2883.5 4333.5 0 31.5 84 52.5 7385 0 0 0 0 011 3027.5 4224.5 10.5 10.5 93 54 7420 0 1 0 0 112 3142 4517.5 52.5 21 70.5 76.5 7880 1 2 0 0 313 3277.5 4976.5 158.5 54 42 89.5 8598 2 6 0 2 1014 3099 4777 239.5 74 67.5 63 8320 3 4 0 0 715 2661 4905 284 87.5 94.5 84 8116 2 4 0 1 716 2609 5067.5 327.5 73.5 66 136.5 8280 0 1 0 0 117 2467.5 5333 239.5 97.5 72.5 194 8404 0 1 0 0 118 2172 5680.5 138.5 50.5 149 281.5 8472 0 1 0 1 219 1886 6000 155 39.5 68.5 352 8501 0 0 0 0 020 1719 5713 127 52.5 11.5 277 7900 1 1 0 1 321 1737.5 5951.5 104.5 73.5 24.5 272.5 8164 0 0 0 0 022 1868.5 6048.5 111.5 60.5 42 277 8408 0 2 0 0 223 1790.5 5947 94.5 42 10.5 283.5 8168 0 1 0 1 224 1833 5528 60.5 69 28.5 237 7756 2 0 0 0 225 2307 5523 59.5 73.5 83.5 241.5 8288 0 0 0 0 026 2504.5 5194.5 51.5 63 90 232.5 8136 1 1 0 1 3Total 64,498 134,397.5 2360.5 1003 1315 3366 206,940 12 27 0 7 46

Journal of Dairy Science Vol. 84, No. 3, 2001

McDougall, S. 1998. Efficacy of two antibiotic treatments in curingclinical and subclinical mastitis in lactating dairy cows. NZ Vet.J. 46:226–232.

Neave, F. K., F. H. Dodd, R. G. Kingwill, and D. R. Westgarth. 1969.Control of mastitis in the dairy herd by hygiene and management.J. Dairy Sci. 52:696–707.

Oliver, S. P., and B. A. Mitchell. 1983. Intramammary infections inprimigravid heifers near parturition. J. Dairy Sci. 66:1180–1183.

O’Shea, J. 1987. Machine milking factors affecting mastitis—A litera-ture review. Bull. Int. Dairy Fed. 215:5–32.

Owens, W. E., S. P. Oliver, B. E. Gillespie, C. H. Ray, S. C. Nickerson.1998. Role of horn flies (Haematobia irritans) in Staphylococcusaureus induced mastitis in dairy heifers. Am. J. Vet. Res.59:1122–1124.

Radostits, O. M., K. E. Leslie, and J. Fetrow. 1994. Herd Health: FoodAnimal Production Medicine. 2nd ed. W.B. Saunders Company,Philadelphia, PA.

Rendos, J. J., R. J. Eberhart, and E. M. Kessler. 1975. Microbialpopulations of teat ends and bedding materials. J. Dairy Sci.58:1492–1500.

Roberson, J. R., L. K. Fox, D. D. Hancock, and C. C. Gay. 1994.Coagulase-positive Staphylococcus intramammary infections inprimiparous dairy cows. J. Dairy Sci. 77:958–969.

Smith, K. L., and J. S. Hogan. 1993. Environmental mastitis. Vet.Clin. North Am.: Food Animal Pract. 9:489–498.

Statistix for Windows. 1998. Version 2.0. Analytical Software Co.,La Jolla, CA.

Todhunter, D. A., K. L. Smith, and J. S. Hogan. 1995. Environmentalstreptococcal intramammary infections of the bovine mammarygland. J. Dairy Sci. 78:2366–2374.