Receiver operating characteristic analysis of milk lactose for identification of mastitis in buffaloes

Transcription

1 Indian J. Anim. Res., 50 (6) 2016 : Print ISSN: / Online ISSN: AGRICULTURAL RESEARCH COMMUNICATION CENTRE Receiver operating characteristic analysis of milk lactose for identification of mastitis in buffaloes T.K. Patbandha*, K. Ravikala, B.R. Maharana, Rupal Pathak 1, S. Marandi, P.U. Gajbhiye 2, T.K. Mohanty 3 and R. Malhotra 3 College of Veterinary Science and A.H., Junagadh Agricultural University (JAU), Junagadh , India. Received: Accepted: DOI: /ijar ABSTRACT Receiver operating characteristic (ROC) analysis is a simple statistical tool used to classify a diagnostic indicator in terms of area under a ROC curve (AUC) and to develop potential threshold values of a diagnostic indicator. Milk lactose was analyzed by ROC analysis to see its accuracy to discriminate infected and healthy udder quarters, and to develope an optimum threshold value along with corresponding sensitivity (Se), specificity (Sp) and positive likelihood ratio (LR+) value. Data for the present study comprised of 1516 milk samples collected from Jaffrabadi buffaloes. Milk lactose was estimated by milk analyzer LACTOSCAN and further samples were checked for sub-clinical mastitis by California mastitis test (CMT). The threshold values of milk lactose for identification of moderate and severe infection were found to be 5.31g% (Se, 58.82%; Sp, 58.28%) and 5.23g% (Se, 70.97%; Sp, 64.41%), respectively by ROC analysis. Milk samples with lactose content below 5.31g% were 1.41 times more likely come from moderately infected quarters (LR+ = 1.41); whereas, below 5.23g% were 1.99 times more likely come from severely infected quarters (LR+ = 1.99). The overall accuracy of milk lactose for discrimination of normal quarters from moderately infected quarters was 64% (AUC=0.64) and from severely infected quarters was 72% (AUC=0.72) (P<0.001). Thus, the present study indicated that milk lactose classified mastitic and healthy udder quarters in Jaffrabadi buffaloes with moderate accuracy. Key words: Buffaloes, Milk lactose, ROC analysis, Mastitis. INTRODUCTION Receiver operating characteristic (ROC) analysis is an important statistical tool originally developed in 1950s (Zou et al., 2007). ROC analysis has been used in medical science for the first time to assess the imaging devices. Since, then it has been used for the evaluation of accuracy of clinical laboratory diagnostic tests as well as diagnostic indicators or markers for classification of diseased and non diseased cases (Hughes and Bhattacharya, 2013). Additionally, it is being used for development of optimum threshold value of diagnostic indicators in medical and veterinary sciences (Fan et al., 2006; Hughes and Bhattacharya, 2013; Patbandha et al., 2012, 2013). The ROC analysis is considered as a simple statistical tool used to characterize a diagnostic variable in terms of area under a ROC curve (AUC) and further provides range of potential threshold values along with their corresponding Sensitivity (Se), Specificity (Sp) and positive likelihood ratio (LR+) values (Patbandha et al., 2012, 2013). Milk constitutes major components such as fat, protein and lactose which are most important to the dairy farmers, producers and consumers (Reis et al., 2013). However, these traits are affected by environmental and nutritional factors as well as physiological and pathological condition of animals. Further, fat is considered as the most variable milk component, followed by protein and lactose is the least variable (Ravikala et al., 2014). On the other hand, during mastitis, alteration of milk lactose occurs markedly compared to other traits (Bansal et al., 2007; Sharif et al., 2007; Tripaldi et al., 2010; Hussain et al., 2012; Reis et al., 2013). Hortet and Seegers (1998) reviewed the alteration of milk fat and protein during mastitis and observed that the alterations were contradictory to several authors. Recently, Reis et al. (2013) also reported contradictory alteration of milk fat and protein during mastitis. Moreover, milk lactose showed little variation within lactation as well as from one lactation to the next, and consistently decreased during udder inflammation (Pyorala, 2003). Therefore, milk lactose content could be used as mastitis indicator in dairy animals. Though, few studies reported threshold values of milk lactose for mastitis identification in cows (Pyorala, 2003; Bansal et al., 2005; Tripaldi et al., 2010) and buffaloes (Bansal et al., 2007), the threshold values for identification of mastitis in Jaffrabadi buffalo is not available. Hence, the present study aimed to see the overall accuracy of milk lactose to classify *Corresponding author s patbandhavet@gmail.com and address Polytechnic in Animal Husbandry, College of Veterinary Science & Animal Husbandry, Junagadh Agricultural University (JAU), Junagadh , Gujarat. 1 College of Agriculture and Research Station, Janjgir-Champa , C.G., India. 2 Cattle Breeding Farm, Junagadh Agricultural University (JAU), Junagadh , Gujarat, India. 3 National Dairy Research Institute, Karnal , Haryana, India.

2 970 INDIAN JOURNAL OF ANIMAL RESEARCH mastitis and normal condition and to find a suitable optimum threshold value of milk lactose for mastitis identification along with Se, Sp and LR+ in Jaffrabadi buffaloes using ROC analysis. MATERIALS AND METHODS The study was carried out at Junagadh Agricultural University, Junagadh (Gujarat), India between March 2014 and February 2015 on Jaffrabadi buffaloes. The experimental buffaloes were reared in loose housing system of management on cemented flooring at Cattle Breeding Farm of the University. Green and dry fodders were given adlibitum to the animals according to seasonal availability, and buffaloes were fed an additional allowance of cottonseed cake, ground maize and Amul dan during milking (twice milking per day i.e am and 4.00 pm) at the milking parlour. Splashing was practiced during summer to reduce heat stress and to reduce environmental effect on the animals; all managerial practices were followed evenly throughout the year. A total of 1516 quarterly milk samples were taken at afternoon milking and milk lactose content was recorded on the same day by milk analyzer LACTOSCAN (New dairy engineering and trading company Pvt. Ltd., Delhi, India). The CMT test was used as gold standard cow side test to detect mastitis. Samples with CMT score 0, +, ++ and +++ were considered as negative, mild, moderate and severe infection, respectively. Statistical analysis: The comparison of milk lactose among normal (N=1371), mild (N=32), moderate (N=51) and severe infection (N=62) was done by one way analysis of variance (ANOVA) followed by Tukey post hoc test to compare all pair wise differences in means. Further, milk lactose was analyzed by ROC analysis using Sigmaplot 11 software package (Systat software, Inc, California, USA) to see the accuracy of the test as indicated by AUC and for the development of optimum threshold value along with their corresponding sensitivity (Se), specificity (Sp) and positive likelihood ratio (LR+) value. Though, ROC analysis produces range of potential threshold values of a diagnostic indicator, the value having highest combines Se and Sp is called as optimum threshold value of that indicator (Patbandha et al., 2012, 2013). The AUC of ROC analysis produces a 2-dimensional graph where Se and 1-Sp are plotted in Y-axis and X-axis, respectively; which measures the accuracy of diagnostic indicators. Accuracy of diagnostic indicator is said to be non-discriminative or non- informative if AUC is 0.5 (i.e., 50% Se and 50% Sp) and perfect if AUC is 1 (i.e., 100% Se and 100% Sp) (Fan et al., 2006); less accurate if AUC is , moderately accurate if AUC is and highly accurate if AUC is (Patbandha et al., 2013). The Se of a diagnostic indicator depends on number of true positive (TP) cases and false negative (FN) cases; whereas, Sp depends on number of true negative (TN) and false positive (FP) cases. Here, TP indicated the number of mastitic milk samples that had milk lactose below the optimum threshold value of lactose and FP indicates the number of normal milk samples that had milk lactose below the optimum threshold value. The TN indicates the number of normal milk samples that had milk lactose above threshold value and FN indicates the number of mastitic milk samples that had milk lactose above threshold value (Table 1). The Se was defined as the proportion of mastitic milk samples that had milk lactose below the threshold value [(TP)/(TP + FN)] and Sp as the proportion of normal milk samples that had milk lactose above the threshold value [(TN)/(TN + FP)]. Further, the Se and Sp together defined the LR+ value; where, LR+ was the number of times a sample that had milk lactose below the threshold value as likely come from infected quarters [LR+ = Se/(1- Sp)] (Peterson et al., 2008; Patbandha et al., 2013). RESULTS AND DISCUSSION The variation of mean milk lactose percent of quarterly milk samples in Jaffrabadi buffaloes during normal and infective conditions are presented in Figure 1. The mean lactose per cent during normal and mild, moderate and severe infection condition were 5.65±0.05, 5.52±0.14, 5.21±0.10 and 5.10±0.07g%, respectively. In the present study, milk lactose per cent differed significantly (P<0.05) between normal and infected quarters. The milk lactose per cent remained significantly higher in normal quarters and decreased significantly by 7.79% and by 9.73% during moderate and severe infection, respectively (P<0.05). Lower milk lactose observed during mastitis is in agreement with previous studies in dairy buffaloes (Bansal et al., 2007; Sharif Fig 1: Least square mean (±SE) of milk lactose between normal and infected quarters samples Table 1: Confusion matrix for binary outcomes Test outcome (based on Gold standard test outcome diagnostic indicator) Positive Negative Positive (exposed) a [TP] b [FP] Negative (un-exposed) c [FN] d [TN] TP, true positive; FP, false positive; FN, false negative; TN, true negative

3 et al., 2007; Tripaldi et al., 2010; Hussain et al., 2012). Lower lactose content may be due to impairment of synthetic activity of mammary tissues or decomposition by leucocytes or mastitis causing pathogens or leakage from milk into blood (Hussain et al., 2012; Reis et al., 2013). Hence, lactose could be used as an indicator due to consistent change during intramammary infections (Tripaldi et al., 2010; Reis et al., 2013). The threshold value having highest combining Se and Sp of a test is called an optimum threshold value indicates the point of coincidence of Se and Sp. The threshold values of milk lactose for identification of moderate and severe infection and their corresponding Se and Sp are presented in Figure 2 and 3, respectively. The threshold value of milk lactose for identification of moderate infection was 5.31g% (Se and Sp values were and 58.28%, respectively). Similarly, for identification of severe infection, the threshold values of milk lactose was 5.23g% (Se and Sp values were and 64.41%, respectively). The optimum threshold value suitable for identification of mastitis observed in Jaffrabadi buffaloes are more or less comparable with earlier studies in dairy bovines (Pyorala, 2003; Bansal et al., 2005; Bansal et al., 2007; Tripaldi et al., 2010). Several authors reported threshold value of milk lactose for mastitis identification to be 4.7g% (Pyorala, 2003; Tripaldi et al., 2010) and 4.8g% (Bansal et al., 2005) in cows and 5.5g% (Bansal et al., 2007) in buffaloes. However, more practical experience is required to determine the usefulness of the threshold values (Pyorala, 2003; Tripaldi et al., 2010). The Se and Sp reflects the proportion of TP and TN cases of a test, respectively. The Se of 58.82% indicated that optimum threshold value of milk lactose (5.31g%) could correctly identify 58.82% of moderately infected quarters (TP cases) but 41.18% cases of moderately infected quarters may go unidentified (FN cases). Similarly, Sp of 58.28% indicated that optimum threshold value of milk lactose could correctly identify approximately 58.28% normal udder quarters (TN cases), but 41.72% normal quarters may be incorrectly identified as moderately infected (FP cases). Further, while Volume 50 Issue 6 (2016) 971 considering identification of severely infected udder quarters, the Se and Sp values were comparatively higher than moderate infection, indicated that milk lactose was more accurate to identify severely infected udder quarters. We observed comparable Se but lower Sp compared to Pyorala, (2003), who reported 43.8 to 60.8% Se and 80.6 to 94.7% Sp for threshold value of milk lactose to identify sub-clinical mastitis in composite milk samples. The LR+ value indicated the number of times a sample would as likely come from infected quarters if milk lactose content remained below the threshold values. In the present study, if milk lactose test value remained below 5.31g%, then the sample was 1.41 times (LR+ = 1.41) as likely come from moderately infected udder quarter. Similarly, a sample was said to be 1.99 times (LR+ = 1.99) as likely come from severely infected udder quarter, if milk lactose test value remained below 5.23g%. Thus, LR+ increases with decrease in milk lactose content, indicates negative relationship of milk lactose with probability of mastitis. Further, decrease threshold value of milk lactose indicates the severity of infection, that is, lower the lactose threshold value, higher the severity of udder infection. The value of AUC of milk lactose for discrimination of normal and infected quarters was 0.64 (Figure 4) and 0.72 (Figure 5) during moderate and severe infections, respectively (P<0.001). The AUC in ROC analysis reflects the overall accuracy of diagnostic indicator (that is, milk lactose) for classification or discrimination of infected and non infected udder quarters. Further, as AUC does not depend on threshold value or cut off values or prevalence of mastitis, considered as a good measure of accuracy to classify healthy and infected quarters (Kumari et al., 2014). The overall accuracy of classification power of milk lactose was 64% during moderate infection and 72% during severe infection. Previous studies also reported that milk lactose is a good measure to distinguish infected and non infected quarters in dairy cows (Pyorala, 2003; Bansal et al., 2005; Bansal et Fig 2: Optimum threshold value of milk lactose (g%) for identification of moderate udder infection Fig 3: Optimum threshold value of milk lactose (g%) for identification of severe udder infection

4 972 INDIAN JOURNAL OF ANIMAL RESEARCH Fig 4: Area under the receiver operating characteristic curve (AUC) during moderate infection al., 2007) and buffaloes (Bansal et al., 2007). Milk lactose could classify infected and non-infected udder quarters correctly in cows with 81% accuracy (Bansal et al., 2005) and in buffaloes with 83.76% accuracy (Bansal et al., 2007). Further, Pyorala (2003) reported that milk lactose could be used to differentiate infected quarters from healthy quarters by 73.9 to 77.1% accuracy. These authors also reported more accuracy to classify infected and healthy udder quarters using milk lactose compared to milk SNF (55.08% accuracy), ph (59% accuracy) and EC (62.94 to 69.0% accuracy). The value of AUC (0.64 to 0.72) reflected that milk lactose is moderately accurate to classify infected and healthy udder quarters in Jaffrabadi buffaloes (Patbandha et al., 2013). Fig 5: Area under the receiver operating characteristic curve (AUC) during severe infection CONCLUSION Results of the present study indicated that milk lactose decreased with increase in severity of infection. Milk samples with lactose content below 5.31g% were more likely to come from moderately infected quarters; whereas, below 5.23g% were more likely come from severely infected quarters. However, the milk lactose is moderately accurate to classify mastitic and healthy udder quarters in Jaffrabadi buffaloes. ACKNOWLEDGEMENT The authors are greatly thankful to Principal and Dean, College of Veterinary Science and Animal Husbandry and Director of Research, Junagadh Agricultural University, Junagadh for providing all research facilities. REFERENCES Bansal, B.K., Hamann, J., Grabowski, N. and Singh, K.B. (2005). Variation in the composition of selected milk fraction samples from healthy and mastitis quarters, and its significance for mastitis diagnosis. J. Dairy Res., 72: Bansal, B.K., Hamann, J., Lind, O., Singh, S.T. and Dhaliwal, P.S. (2007). Somatic cell count and biochemical components of milk related to udder health in buffaloes. Ital. J. Anim. Sci., 6: Fan, J., Upadhye, S. and Worster, A. (2006). Understanding receiver operating characteristic (ROC) curves. Can J Emerg Med., 8: Hortet, P. and Seegers, H. (1998). Loss in milk yield and related composition changes resulting from clinical mastitis in dairy cows. Prev. Vet. Med., 37:1-20. Hughes, G. and Bhattacharya, B. (2013). Symmetry properties of bi-normal and bi-gamma receiver operating characteristic curves are described by Kullback-Leibler divergences. Entropy, 15: Hussain, R., Javed M.T. and Khan, A. (2012). Changes in some biochemical parameters and somatic cell counts in the milk of buffalo and cattle suffering from mastitis. Pak. Vet. J., 32: Reis, C.B.M., Barreiro, J.R., Mestieri, L., Porcionato, M.A. and dos Santos, M.V. (2013). Effect of somatic cell count and mastitis pathogens on milk composition in Gyr cows. BMC Vet. Res., 9:67. doi: / Patbandha, T.K., Mohanty, T.K., Layek, S.S., Kumaresan, A. and Behera, K. (2012). Application of pre-partum feeding and social behaviour in predicting risk of developing metritis in crossbred cows. Appl. Anim. Behav. Sci., 139:10-17.

5 Volume 50 Issue 6 (2016) 973 Patbandha, T.K., Mohanty, T.K., Layek, S.S., Kumaresan, A., Kantwa, S.C., Malhotra, R., Ruhil, A.P. and Prasad, S. (2013). ROC analysis of pre-partum feeding time can accurately predict post-partum metritis development in Holstein Friesian (HF) crossbred cows. J. Vet. Behav., 8: Peterson, A.T., Papes, M. and Soberon, J. (2008). Rethinking receiver operating characteristic analysis applications in ecological niche modelling. Ecol. Model., 213: Pyorala, S. (2003). Indicators of inflammation in the diagnosis of mastitis. Vet. Res., 34: Ravikala, K., Patbandha T. K. and Vataliya, P.H. (2014). Nutritional management of dairy animals through milk yield and its component evaluation. Proc. of 21st annual convention of Indian Society of Animal Production and Management, January 28-30, AAU, Anand, Gujarat, India. pp Sharif, A., Ahmad, T., Bilal1, M.Q., Yousaf, A. and Muhammad, G. (2007). Effect of severity of sub-clinical mastitis on somatic cell count and lactose contents of buffalo milk. Pak. Vet. J., 27: Kumari, S., Prasad, S., Patbandha, T.K., Pathak, R., Kumaresan, A., Boro, P., Manimaran A. and Mohanty, T.K. (2014). Metabolic indicators for retention of fetal membranes in Zebu and crossbred dairy cattle. Anim Prod. Sci., doi: /AN14941 Tripaldi, C., Palocci, G., Miarelli, M., Catta, M., Orlandini, S., Amatiste, S., Di Bernardini, R. and Catillo, G. (2010). Effects of mastitis on buffalo milk quality. Asian-Aust. J. Anim. Sci., 23: Zou, K.H., O Malley, A.J. and Mauri, L. (2007). Receiver Operating Characteristic analysis for evaluating diagnostic tests and predictive models. Circulation, 115: