- Open Access
Temporal and spatial dynamics of Cryptosporidium parvum infection on dairy farms in the New York City Watershed: a cluster analysis based on crude and Bayesian risk estimates
© Szonyi et al; licensee BioMed Central Ltd. 2010
- Received: 12 April 2010
- Accepted: 17 June 2010
- Published: 17 June 2010
Cryptosporidium parvum is one of the most important biological contaminants in drinking water that produces life threatening infection in people with compromised immune systems. Dairy calves are thought to be the primary source of C. parvum contamination in watersheds. Understanding the spatial and temporal variation in the risk of C. parvum infection in dairy cattle is essential for designing cost-effective watershed management strategies to protect drinking water sources. Crude and Bayesian seasonal risk estimates for Cryptosporidium in dairy calves were used to investigate the spatio-temporal dynamics of C. parvum infection on dairy farms in the New York City watershed.
Both global (Global Moran's I) and specific (SaTScan) cluster analysis methods revealed a significant (p < 0.05) elliptical spatial cluster in the winter with a relative risk of 5.8, but not in other seasons. There was a two-fold increase in the risk of C. parvum infection in all herds in the summer (p = 0.002), compared to the rest of the year. Bayesian estimates did not show significant spatial autocorrelation in any season.
Although we were not able to identify seasonal clusters using Bayesian approach, crude estimates highlighted both temporal and spatial clusters of C. parvum infection in dairy herds in a major watershed. We recommend that further studies focus on the factors that may lead to the presence of C. parvum clusters within the watershed, so that monitoring and prevention practices such as stream monitoring, riparian buffers, fencing and manure management can be prioritized and improved, to protect drinking water supplies and public health.
- Dairy Farm
- Spatial Cluster
- Dairy Herd
- Protect Water Quality
Cryptosporidium is a protozoan parasite that is recognized as one of the most important biological contaminants in drinking water . Cryptosporidiosis is associated with gastrointestinal infection which can be life threatening in immuno-compromised individuals. The infection is transmitted by the fecal-oral route either by direct contact or through contamination of food and water . An experimental study of healthy adult volunteers revealed that the ingestion of as few as 30 Cryptosporidium oocysts can initiate infection . Water-borne transmission is facilitated by the long-lasting infectivity of the oocyst in the environment and its resistance to conventional water treatment technologies such as chlorination .
The New York City Watershed is currently the focus of a long-term project investigating the public health risk of waterborne cryptosporidiosis. Active surveillance in the city which began in 1994 has identified over 100 cases of cryptosporidiosis annually among NYC residents . A quantitative risk assessment model for cryptosporidiosis in NYC predicted that the mean annual risk estimates for infection for all ages and persons with or without HIV/AIDS exceed the proposed acceptable annual risk level of 1 case of infection per 10,000 .
The New York City water supply system provides drinking water to almost half the population of New York State, which includes over 8 million people in the City and one million in Upstate counties, plus millions of commuters and tourists. The water is supplied from a network of 19 reservoirs and three controlled lakes that contain a total storage capacity of approximately 2 billion cubic meters. The total watershed area for the system is approximately 5,100 km2 extending over 200 km north and west of NYC. The system is dependent on precipitation and subsequent runoff via streams and rivers to supply the reservoirs. The water is then moved via a series of gravity-fed aqueducts to the distributions system, where it is chlorinated before it reaches the consumers . Pathogens such as Cryptosporidium pose a significant threat to public health in the City's unfiltered water supply, because the oocysts are very resistant to chlorination, and they are regularly detected in reservoir effluents [3, 6].
Dairy calves are thought to be a primary source of zoonotic Cryptosporidium parvum contamination in watershed ecosystems . In the NYC Watershed, the Catskill/Delaware drainage system is home to approximately 200 dairy farms. To avoid building a huge filtration plant that could cost about $8 billion and the associated $300 million per year for operating costs, NYC implements extensive watershed management measures, including water quality monitoring and best management practices (BMP) on agricultural land, with the goal to protect water quality while maintaining economic viability on these farms . Watershed management requires a network design that demands distinct spatial and temporal monitoring and protection efforts . However to date, the spatial and temporal variation in the risk of C. parvum infection in dairy herds in watersheds has not been investigated. Understanding the spatial and temporal pattern of C. parvum infection on dairy farms would be useful in designing or modifying watershed management strategies to monitor and mitigate the risk of C. parvum contamination in watersheds.
Bayesian approach has been used increasingly in geographical epidemiologic studies, because it stabilizes crude risk estimates by reducing variance heterogeneity. Thus, risk maps based on Bayesian rather than crude risk estimates are preferred because they are more accurate and visually appealing . In a Bayesian approach, a prior probability distribution for the values of a parameter (based on previous studies) is converted (under the influence of current observations) to a posterior distribution of that parameter. This posterior distribution is used to provide an estimate for the parameter .
The objectives of the study were to 1) explore and map the temporal and spatial dynamics of the risk of C. parvum infection in dairy cattle in the NYC Watershed, and to 2) identify high-risk clusters in space and time. The study utilized both crude-, and Bayesian prevalence estimates to accurately describe the spatial epidemiology of this important zoonotic parasite among dairy herds in a large watershed ecosystem.
Description of data and study area
Characteristics of the data used in the study
Range of CPd
Range of BPf
The Bayesian prevalence estimates are summarized in Table 1.
Potential clustering of zoonotic strains
where N is the number of farms, is the average prevalence on the farms, X i and X j are the prevalence on farm i and j, respectively, and W ij is the spatial weight between farms i and j, determined by the distance between farms i and j. The Z Score associated with the index is based on the Randomization Null Hypothesis stating that "there is no spatial clustering". Thus Z-scores greater than 1.96 or smaller than -1.96 indicate significant spatial autocorrelation at the 5% level .
The spatial relationship among the farms was conceptualized with the inverse distance model (the impact of one feature on another decreases with distance). The global spatial autocorrelation test was performed on both the crud and the Bayesian prevalence estimates by season, using the geographical information system (GIS) software ArcView 9.2 (ESRI, CA, USA).
Where N is the total number of cases, c and E[c] represent the observed and expected number of cases in a window, while N - c and N - E[c] indicate the observed and expected number of cases outside the window. The indicator function I() is equal to 1 when the window has more cases than expected under the null hypothesis, and zero otherwise. The window with the highest likelihood ratio is the most likely cluster and is assigned a p value through 999 Monte Carlo simulations . A Poisson model was fitted to the raw data for each season to examine the presence of purely spatial clusters. A Poisson model was also applied to the entire dataset to determine whether purely temporal (i.e. seasonal) or space-time clusters existed during the course of a year. At each farm location, cases were defined as the number of calves that tested positive for Cryptosporidium by the flotation method, while the population size was the total number of calves that were tested. The maximum cluster size was set at the recommended value (50% of the total population at risk). Both circular and oval cluster shapes were evaluated.
Mapping the risk of C. parvum in the watershed
Geospatial coordinates for each farm were collected with a Garmin eTrex Summit handheld global positioning system (GPS) device (Garmin International Inc, Olathe, Kansas, USA) and imported into the GIS software Manifold System 8.0 Ultimate Edition (Manifold, Carson City, NV, USA). The geographical coordinates were re-projected into the Universal Transverse Mercator coordinate system, Zone 18(N), North American Datum 1983, and overlaid with the shapefile of Delaware County, NY obtained from the New York State Geographic Information System Clearinghouse http://www.nysgis.state.ny.us. Dot maps indicating the seasonal prevalence estimates on the study farms were created to examine the spatial dynamics of C. parvum in the watershed during the annual cycle, and to explore the differences between the crude and the Bayesian estimates.
Potential clustering of zoonotic strains
Results for the test of global spatial autocorrelation using Moran's I statistics based on crude-, and Bayesian risk estimates
The most likely temporal, spatial, and space-time clusters identified by the SaTScan statistics using the Poisson probability model, based on crude prevalence estimates
no significant cluster
no significant cluster
We carried out this study to evaluate potential clustering of dairy herds that are infected with C. parvum in the NYC Watershed. This was the first study to evaluate the spatial and temporal variation in the risk of C. parvum infection in dairy cattle in an important watershed.
The decision to include only pre-weaned calves in this study was based on the results of a quantitative risk assessment (QRA) of Cryptosporidium in dairy cattle in the NYC Watershed which revealed, that despite representing only a small proportion of the population and producing a small fraction of manure, pre-weaned calves produced the vast majority of all zoonotic C. parvum oocysts shed within the dairy cattle population. Specifically, it was estimated that pre-weaned calves produced 99.5% of the total C. parvum oocyst burden with a calculated mean log oocyst shedding of 4.02 × 1010 daily. Thus it was estimated that pre-weaned calves produce nearly all the C. parvum oocysts that contaminate the watershed .
One of the assumptions we made in the models was that all C. parvum-like oocysts shed by pre-weaned calves were zoonotic. We made this assumption because molecular typing is required to determine the zoonotic potential of C. parvum-like oocysts, and this necessity would have further amplified the problem of small numbers. While over-estimating the zoonotic risk, we felt this assumption was reasonable, because recent studies that applied molecular typing revealed that the majority of Cryptosporidium infections in pre-weaned calves were indeed zoonotic [11, 18, 22].
Both global (Moran's I) and specific (SaTScan) cluster detection methods identified a significant spatial cluster of C. parvum infection in calves in the winter, with a relative risk (RR) of 5.8, based on crude risk estimates. No other purely spatial clusters were identified with either method. Thus, there was complete agreement between the results of the two cluster detection methods. In addition, the scan statistics detected a significant space-time cluster in the summer with both circular (RR = 2.3) and elliptical (RR = 3) window settings. Further investigation revealed the presence of a significant temporal cluster (RR = 2) but the lack of a purely spatial cluster in the summer, which suggests that the space-time clusters identified in the summer were due to a temporal rather than a spatial increase in risk. The large sizes of the space-time clusters including nearly 50% of the population at risk (maximum allowed under the conditions specified) also supports the notion of a spatially diffuse increase in the risk of C. parvum infection in the summer.
It has been suggested that farms downstream of other farms may be contaminated with Cryptosporidium via runoff from farms upstream, although evidence for this epidemiologic link is lacking [23–25]. The rationale for considering elliptical spatial clusters in this study was that farm-to-farm transmission via runoff would be expected to produce an elliptical rather than a circular cluster.
The term disease cluster is defined as an increase in the expected number of cases within a population bounded in space and time . We used two different cluster detection methods to ensure comparability and robustness of results. We elected to use the scan statistics for the investigation of temporal and space-time clusters, because recent studies identified SatScan as the most developed and robust space-time surveillance software package that takes multiple testing problems into account, and is considered the most powerful for detecting localized clusters .
The major limitation of the study was the low number of pre-weaned calves on the farms, resulting in unstable risk estimates as small populations have large variability in rates . For example, the small number of cases and population at risk may have accounted for the high relative risk estimate (RR = 5.8) for the winter spatial cluster. This limitation was corrected with the use of a Bayesian approach that has the ability to stabilize the raw estimates derived from a small number of individuals . The Bayesian approach also allowed us to incorporate prior knowledge about the risk of C. parvum infection in the target population, and to account for the imperfections of the diagnostic test. However, the quality of this prior information might influence the quality of the estimate and hence could be a source of bias and limitations . The degree of smoothing provided by the Bayesian approach is a trade-off between high sensitivity (truly high risk areas correctly identified) and high specificity (areas without excessive risk correctly identified) such that sensitive but non-specific measure will generate many false positive findings, whereas a specific but not sensitive measure will miss areas with high risk . In our study we encountered contradictory results in the analysis of purely spatial clusters using Global Moran's I statistics. While crude estimates revealed significant spatial autocorrelation in the winter, Bayesian estimates indicated the lack of a spatial clustering in any season. Considering the limitations of using crude vs. Bayesian estimates in spatial analysis, the discrepancy in our results may be due either to the instability of the crude estimates leading to spuriously high values, or to the low sensitivity of the Bayesian model to detect areas with truly high risk.
Cryptosporidium is considered a non-point source pollutant in watersheds that is carried off the land surface during rain events [30, 31]. Monitoring of stream sites in the study area revealed that event based (e.g. after storms) Cryptosporidium concentrations were consistently higher than baseline results (up to 11.7 oocysts 50L-1), implicating runoff as contamination source . The close relationship between activities in the drainage basin and the quality of its water resources forms the underlying premise for all watershed management programs . Best management practices that protect water supply on farms such as fencing, filter strips, stream crossings, animal trails and walkway, manure composting facility, and runoff management systems would ideally and ultimately be implemented on every farm in the watershed. However, until that goal is achieved, prioritization methodologies to address non-point source pollutants need to be developed, and the identification of "hot spots" is an integral part of this process.
The occurrence of spatial or temporal clusters may be due to rapid spread between locations in the case of a highly contagious disease, or the presence of common environmental risk factors . The higher risk of C. parvum infection on dairy farms in the summer throughout the study area may be due to climatic or management factors that affect the entire area. This finding suggests that spreading manure in the summer (compared to other seasons) in any area of the watershed is associated with an increased risk of C. parvum contamination of the water supply. This finding is important because farms in the study area regularly spread untreated calf manure in the fields. The current recommendation is to avoid spreading manure in the spring and during frozen conditions, while summer is considered a lower risk period . If further studies confirm an increased risk of Cryptosporidium contamination in the summer, this knowledge will be useful to improve Nutrient Management Plans, which give recommendations about the most environmentally safe time and place to spread manure.
With the City's population expected to rise to 9.1 million by 2039 from 8.3 million in 2005, watershed management will continue to have an important part to play in protecting water quality . Over time, systematic and careful monitoring of disease-causing organisms and pollutants will determine the effectiveness of New York City's protection strategies and the continued success of its filtration avoidance plan. The identification of spatial or temporal "hot-spots" of C. parvum contamination within the watershed will have important implications for watershed monitoring and management, and need to be the focus of future investigations.
The identification of C. parvum clusters is a priority in designing cost-effective and targeted watershed management practices to ensure safety of the water supplies for public health. This study identified high risk clusters of C. parvum infection in dairy herds in both space and time in a large and important watershed, suggesting that further studies are needed to determine whether the presence of clusters are persistent and predictable. We recommend that future studies focus on the causes of these "hot spots" so that watershed monitoring and management strategies may be implemented and targeted to effectively decrease C. parvum contamination of the water supply.
The project was partially supported financially by USDA grant # 2006-51290 and from the EPA/NRC. We thank all the farmers who volunteered to participate in this project.
- Graczyk TK, Fayer R, Cranfield MR: Zoonotic transmission of Cryptosporidium parvum: Implications for water-borne cryptosporidiosis. Parasitol Today. 1997, 13: 348-351. 10.1016/S0169-4758(97)01076-4.PubMedView ArticleGoogle Scholar
- Smith HV, Caccio SM, Cook N, Nichols RA, Tait A: Cryptosporidium and Giardia as foodborne zoonoses. Vet Parasitol. 2007, 149: 29-40. 10.1016/j.vetpar.2007.07.015.PubMedView ArticleGoogle Scholar
- Betancourt WQ, Rose JB: Drinking water treatment processes for removal of Cryptosporidium and Giardia. Vet Parasitol. 2004, 126: 219-234. 10.1016/j.vetpar.2004.09.002.PubMedView ArticleGoogle Scholar
- New York City Department of Health and Mental Hygiene, Bureau of Communicable Diseases and New York City Department of Environmental Protection, Bureau of Water Supply: Waterborne Disease Risk Assessment Program: 2008 Annual Report. New York, NY. 2009Google Scholar
- Makri A, Modarres R, Parkin R: Cryptosporidiosis susceptibility and risk: a case study. Risk Anal. 2004, 24: 209-220. 10.1111/j.0272-4332.2004.00424.x.PubMedView ArticleGoogle Scholar
- New York City Department of Environmental Protection, Bureau of Water Supply: 2008 Watershed Water Quality Annual Report. New York, NY. 2009Google Scholar
- Xiao L, Feng Y: Zoonotic cryptosporidiosis. FEMS Immunol Med Microbiol. 2008, 52: 309-323. 10.1111/j.1574-695X.2008.00377.x.PubMedView ArticleGoogle Scholar
- New York City Department of Environmental Protection: Filtration Avoidance Annual Report, 2009. New York, NY. 2010Google Scholar
- Berke O: Exploratory spatial relative risk mapping. Prev Vet Med. 2005, 71: 173-182. 10.1016/j.prevetmed.2005.07.003.PubMedView ArticleGoogle Scholar
- Rezaeian M, Dunn G, St Leger S, Appleby L: Geographical epidemiology, spatial analysis and geographical information systems: a multidisciplinary glossary. J Epidemiol Community Health. 2007, 61: 98-102. 10.1136/jech.2005.043117.PubMedPubMed CentralView ArticleGoogle Scholar
- Szonyi B, Bordonaro R, Wade SE, Mohammed HO: Seasonal variation in the prevalence and molecular epidemiology of Cryptosporidium infection in dairy cattle in the New York City Watershed. Parasitol Res. doi 10.1007/s00436-010-1864-4,Google Scholar
- Georgi JR, Georgi ME: Parasitology for veterinarians. 1990, Philadelphia, PA: W.B. Saunders Co, 5Google Scholar
- Spiegelhalter D, Thomas A, Best N, Gilks W: BUGS: Bayesian inference using Gibbs sampling, version 0.50. 1996, Cambridge: MRC Biostatistics UnitGoogle Scholar
- Starkey SR, Kimber KR, Wade SE, Schaaf SL, White ME, Mohammed HO: Risk factors associated with Cryptosporidium infection on dairy farms in a New York State watershed. J Dairy Sci. 2006, 89: 4229-4236. 10.3168/jds.S0022-0302(06)72468-7.PubMedView ArticleGoogle Scholar
- Starkey SR, White ME, Mohammed HO: Cryptosporidium and dairy cattle in the Catskill/Delaware watershed: a quantitative risk assessment. Risk Anal. 2007, 27: 1469-1485. 10.1111/j.1539-6924.2007.00982.x.PubMedView ArticleGoogle Scholar
- Branscum AJ, Gardner IA, Johnson WO: Bayesian modeling of animal-and herd-level prevalences. Prev Vet Med. 2004, 66: 101-112.PubMedView ArticleGoogle Scholar
- Wade SE, Mohammed HO, Schaaf SL: Prevalence of Giardia sp. Cryptosporidium parvum and Cryptosporidium andersoni (syn. C. muris) [correction of Cryptosporidium parvum and Cryptosporidium muris (C. andersoni)] in 109 dairy herds in five counties of southeastern New York. Vet Parasitol. 2000, 93: 1-11. 10.1016/S0304-4017(00)00337-X.PubMedView ArticleGoogle Scholar
- Starkey SR, Zeigler PE, Wade SE, Schaaf SL, Mohammed HO: Factors associated with shedding of Cryptosporidium parvum versus Cryptosporidium bovis among dairy cattle in New York State. J Am Vet Med Assoc. 2006, 229: 1623-1626. 10.2460/javma.229.10.1623.PubMedView ArticleGoogle Scholar
- Moran PAP: Notes on continuous stochastic phenomena. Biometrika. 1950, 37: 17-33.PubMedView ArticleGoogle Scholar
- SaTScanTM v8.0: Software for the spatial and space-time scan statistics.http://www.satscan.org
- Kulldorff M: A spatial scan statistic. Communications in Statistics: Theory and Methods. 1997, 26: 1481-1496. 10.1080/03610929708831995.View ArticleGoogle Scholar
- Santin M, Trout JM, Xiao L, Zhou L, Greiner E, Fayer R: Prevalence and age-related variation of Cryptosporidium species and genotypes in dairy calves. Vet Parasitol. 2004, 122: 103-117. 10.1016/j.vetpar.2004.03.020.PubMedView ArticleGoogle Scholar
- Sischo WM, Atwill ER, Lanyon LE, George J: Cryptosporidia on dairy farms and the role these farms may have in contaminating surface water supplies in the northeastern United States. Prev Vet Med. 2000, 43: 253-267. 10.1016/S0167-5877(99)00107-5.PubMedView ArticleGoogle Scholar
- Ong C, Moorehead W, Ross A, Isaac-Renton J: Studies of Giardia spp. and Cryptosporidium spp. in two adjacent watersheds. Appl Environ Microbiol. 1996, 62: 2798-2805.PubMedPubMed CentralGoogle Scholar
- Hansen JS, Ongerth JE: Effects of time and watershed characteristics on the concentration of Cryptosporidium oocysts in river water. Appl Environ Microbiol. 1991, 57: 2790-2795.PubMedPubMed CentralGoogle Scholar
- Elliott P, Wartenberg D: Spatial epidemiology: current approaches and future challenges. Environ Health Perspect. 2004, 112: 998-1006. 10.1289/ehp.6735.PubMedPubMed CentralView ArticleGoogle Scholar
- Robertson C, Nelson TA: Review of software for space-time disease surveillance. Int J Health Geogr. 9: 16-10.1186/1476-072X-9-16.Google Scholar
- Olsen SF, Martuzzi M, Elliott P: Cluster analysis and disease mapping--why, when, and how? A step by step guide. Bmj. 1996, 313: 863-866.PubMedPubMed CentralView ArticleGoogle Scholar
- Berke O: Exploratory disease mapping: kriging the spatial risk function from regional count data. Int J Health Geogr. 2004, 3: 18-10.1186/1476-072X-3-18.PubMedPubMed CentralView ArticleGoogle Scholar
- Atwill ER, Pereira MD, Alonso LH, Elmi C, Epperson WB, Smith R, Riggs W, Carpenter LV, Dargatz DA, Hoar B: Environmental load of Cryptosporidium parvum oocysts from cattle manure in feedlots from the central and western United States. J Environ Qual. 2006, 35: 200-206. 10.2134/jeq2005.0099.PubMedView ArticleGoogle Scholar
- Graczyk TK, Evans BM, Shiff CJ, Karreman HJ, Patz JA: Environmental and geographical factors contributing to watershed contamination with Cryptosporidium parvum oocysts. Environ Res. 2000, 82: 263-271. 10.1006/enrs.1999.4022.PubMedView ArticleGoogle Scholar
- Stevens KB, Del Rio Vilas VJ, Guitian J: Classical sheep scrapie in Great Britain: spatial analysis and identification of environmental and farm-related risk factors. BMC Vet Res. 2009, 5: 33-10.1186/1746-6148-5-33.PubMedPubMed CentralView ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.