Geographical Perspectives on Epidemic Transmission of Cholera in Haiti, October 2010 Through March 2013

The current epidemic of El Tor cholera in the Caribbean republic of Haiti is one of the largest single outbreaks of the disease ever recorded. The prospects are that the epidemic will continue to present challenges to workers in public health medicine, epidemiology, and allied fields in the social sciences for years to come. This article introduces geographers to the environmental context of the Haiti cholera epidemic, the principal data sources available to analyze the occurrence of the epidemic, and evidence regarding its geographical origins and dispersal during the first thirty months of the epidemic, October 2010 through March 2013. Using weekly case data collated by the Haitian Ministère de la Santé Publique et de la Population (MSPP), techniques of time series analysis are used to examine inter- and intradepartmental patterns of cholera activity. Our analysis demonstrates a pronounced lag structure to the spatial development of the epidemic (Artibonite and northern departments → Ouest and metropolitan Port-au-Prince → southern departments). Observed variations in levels of epidemiological integration, both within and between departments, provide new perspectives on the spatiotemporal evolution of the epidemic to its March 2013 pattern.

T he ongoing epidemic of El Tor cholera in the Caribbean republic of Haiti is one of the largest recorded outbreaks of cholera in the modern era (Figure 1). Beginning with a putative introduction by Nepalese peacekeepers in mid-October 2010, the cumulative number of reported cholera cases had reached 704,371 by the week ending 19 July 2014, of whom 394,896 had been hospitalized and 8,580 had died (Pan American Health Organization [PAHO] 2014b). More than 6 percent of the country's population of 10.32 million had been infected in this period, with a case-fatality rate for clinical cases of 1.2 percent (PAHO 2014a(PAHO , 2014b. From Haiti, the disease has spilled into the Dominican Republic, the United States, Cuba, and, most recently, Mexico (Centers for Disease Control and Prevention [CDC] 2010d; Moore et al. 2014). The current plan of the Minist ere de la Sant e Publique et de la Population (MSPP) to eliminate cholera in Haiti by the year 2022 indicates that the disease will remain a problem in the country for years to come (MSPP 2013).
This article presents a medical geographical perspective on the evolution of the Haiti cholera epidemic to its spatial pattern in March 2013. We apply analytical concepts and modeling techniques that provide intrinsically geographical insights into transmission patterns and processes that, in the context of the Haiti epidemic, have not previously been explored in the literature. Based on a systematic analysis of the records of some 600,000 clinical cases of cholera, we demonstrate that the evolution of the epidemic was underpinned by a fundamental geographical divide between the northern and southern departments of the country. This north-south divide manifested as a spatial lead-lag structure in epidemic development, with the capital department (Ouest) serving as a pivot point in the southward spread of the disease. Importantly from the perspective of disease intervention, our analysis also identifies pronounced differences in the degree of epidemiological integration of geographical units. This information should inform the spatial planning and implementation of ongoing cholera elimination strategies in the country.

Epidemic Context
The cholera epidemic in Haiti has spread against a backdrop of poor health, water, and sanitation Figure 1. Location map of Haiti. The map shows the major towns, departments, communes, and rivers referred to in the text. Haiti is divided into ten departments, which, in turn, are subdivided into communes. For each department, the pie charts show the cumulative number of reported cholera cases and the proportion of cases hospitalized in the thirty-month interval to be analyzed in this article (October 2010-March 2013 infrastructure. In 2008, only 12 percent of Haitians received piped treated drinking water and just 17 percent had access to adequate sanitation (A. Ali et al. 2011;Ceccarelli et al. 2011). The epidemic followed nine months after a moment magnitude (Mw) 7.0 earthquake on 12 January 2010 with an epicenter just 25 km west of the capital, Port-au-Prince. The net outflow of people from the Port-au-Prince area in response to the earthquake is estimated to have been 20 percent of the pre-earthquake population (Bengtsson et al. 2011). Many internally displaced persons (IDPs) were housed in temporary camps that were then flooded on 5 November 2010 when Hurricane Tomas hit the country. Despite the importance of these factors in catalyzing the epidemic, public and political controversy has centered on the role of United Nations (UN) peacekeepers in the initial introduction of cholera (Cravioto et al. 2011).
As measured by morbidity, the Haitian epidemic is the largest reported outbreak of the disease since the nineteenth century, and it has occurred in a country with no previous recorded history of the disease. As a consequence, the disease has spread in an immunologically na€ ıve population (Dowell and Braden 2011). The more than 700,000 cases that had occurred by July 2014 put the magnitude of the epidemic on a different scale than other recent outbreaks (M. Ali et al. 2012). In 2010, Haiti accounted for 57 percent of the world's total cholera cases reported to the World Health Organization (WHO) and 53 percent of all cholera deaths. In 2011 the situation was little changed: 58 percent of the world's cholera cases and 37 percent of deaths reported to WHO were in Haiti (Barzilay et al. 2013). The outbreak continues, albeit at a much lower level. The weekly average for new cases in the first five months of 2014 was 291, with one death. This can be compared with the weekly averages of 993 cases and eight deaths in 2013, 1,498 cases and eleven deaths in 2012, and 7,697 cases and sixty-two deaths in 2011 (PAHO 2014b).
The Haitian epidemic is related to the country's poverty in several ways. First, in the absence of human and financial resources to deal with the consequences of the January 2010 earthquake, international peacekeepers were provided by the United Nations to help those displaced by the earthquake. This act appears to have introduced cholera into the country (A. Ali et al. 2011;Ceccarelli et al. 2011;Chin et al. 2011;Cravioto et al. 2011;Hendriksen et al. 2011;Mutreja et al. 2011). Second, the earthquake damage to already poor sewerage and water systems increased population vulnerability to cholera (A. Ali et al. 2011;Ceccarelli et al. 2011;Chery, Dodard, and Fournier 2012). Third, control of the epidemic, once it was under way, was handicapped by Haiti's weak, mostly privatized, national health care system, which is inaccessible to much of the population (Farmer 2011). Finally, the ongoing population displacement and associated interdepartmental movements since the earthquake have maintained the nationwide spread of the disease (Bengtsson et al. 2011;Chery, Dodard, and Fournier 2012).

The Study and Layout of the Article
Medical geographers have a long-standing interest in the spatial transmission of cholera (May 1951) and a rich vein of geographical scholarship has developed in relation to the spread of the disease. In addition to atlas-based treatments of pandemic wave transmission (May 1951;Cliff, Haggett, and Smallman-Raynor 2004), geographers have examined the spatial dimensions of nineteenth-and early twentieth-century cholera epidemics in, among other countries, Japan (Kuo and Fukui 2007), Russia (Patterson 1994), the United Kingdom (Cliff and Haggett 1988), and the United States (Pyle 1969). Other geographical studies have investigated the spread of cholera in relation to past conflicts (Smallman-Raynor and Cliff 1998a, 1998b, and social constructions of the disease in the nineteenth century have also been explored (Jackson 2013). More particularly, this study is underpinned by an ongoing geographical concern with the spatial transmission of cholera as illustrated by the recent series of modeling studies in relation to Matlab, Bangladesh (Giebultowicz et al. 2011;Emch et al. 2012).
Our examination of the Haiti cholera epidemic is framed by a broader geographical concern with the impacts and consequences of the Haiti earthquake of January 2010, including environmental health risks (Curtis et al. 2013;Widmer et al. 2014), postdisaster mobility, humanitarian logistics and aid (Sheller 2013;Versluis 2014), environmental sustainability (Abrahams 2014), cultural legacies (Cruse 2013), and representations of disaster vulnerability (Taylor 2013). As a contribution to this broader literature, we present a medical geographical perspective on the national spread dynamics and spatial structure of the postearthquake cholera epidemic from October 2010 through March 2013. Geographers have much to contribute to an improved understanding of the spatial and temporal dimensions of the outbreak. As such, this article is intended as both an introduction and reference point for future geographical investigations of the epidemic.
The article begins with a reconstruction of the initial diffusion of the cholera epidemic from its putative origin at the United Nations Stabilization Mission in Haiti (MINUSTAH) camp complex, Centre Department, to the rest of the country in the period between October and December 2010. This was the first, and the largest, of several major waves of cholera to spread through Haiti during the first thirty months of the epidemic ( Figure 2). In subsequent sections of the article, we use methods of time series and coherence analysis to examine the spatial structure of this ongoing epidemic. Our results shed fresh light on the geographical evolution of the epidemic and provide important insights for the spatial elimination of cholera in Haiti.

Background and Historical Context
The Haiti Earthquake (12 January 2010) The earthquake that struck Haiti at 16:53 hours (local time) on Tuesday 12 January 2010 was one of the most destructive in modern history (Centre for Research on the Epidemiology of Disasters 2014). With an epicenter near the town of L eogâne, approximately 25 km west of Port-au-Prince, the earthquake caused US$8,000 million in damage to the physical infrastructure of the capital city and elsewhere in the departments of Ouest, Nippes, and Sud-Est ( Figure 3A). Government officials estimated that the earthquake resulted in 230,000 immediate deaths, with an additional 300,000 people injured and some 2 million in need of temporary shelter. Among the latter, 1.5 million people settled in the many hundreds of overcrowded IDP camps that had appeared The earthquake caused severe damage to the health care infrastructure of the Ouest, Nippes, and Sud-Est departments and compromised the operational capabilities of the Haitian MSPP (Santa-Olalla et al. 2013). The postearthquake health situation was compounded by the overcrowded and unsanitary conditions that prevailed in the IDP camps, with many lacking potable water, toilets, and bathing facilities (International Organization for Migration in Haiti 2013; see Figure 5). Emergency health monitoring of IDPs in the three to four months that followed the earthquake identified acute respiratory infections, watery diarrhea, suspected malaria, and fever of unknown origin as the most commonly encountered conditions. Cases of suspected typhoid, bloody diarrhea, acute hemorrhagic fever syndrome, and acute febrile illness with jaundice were counted in the hundreds, and small numbers of actual or suspected cases of diphtheria, measles, meningococcal meningitis, rabies, tetanus, and whooping cough were also recorded (CDC 2010a(CDC , 2010bPolonsky et al. 2013).
Although disease monitoring activities were maintained throughout the summer of 2010, the available records suggest that cholera-actual or suspectedwas absent from Haiti at this time (CDC 2011). When the disease did finally arrive in October of the same year, it found a set of pre-and postdisaster circumstances that were highly conducive to its rapid transmission. Earthquake damage to the country's already poor sewage and water systems was compounded by overcrowded and unsanitary living conditions and deficiencies in public health infrastructure to treat and prevent the disease (Gelting et al. 2013;Santa-Olalla et al. 2013).

Hurricane Tomas (5 November 2010)
On the morning of Friday 5 November 2010, Hurricane Tomas brushed the westernmost tip of Haiti and tracked northeast through the Windward Passage that separates the islands of Hispaniola and Cuba. Although the hurricane did not make landfall in Haiti, heavy rains resulted in some flooding in the five southern departments (Grande-Anse, Nippes, Ouest, Sud, and Sud-Est), Artibonite, Centre, and Nord-Ouest; see Figure 1 for locations. To the west of Port-au-Prince, for example, IDP camps in the town of L eogâne were inundated by overflowing river water. Other communes in the vicinity of L eogâne, including Petit Goâve and Jacmel, were also reported to have been seriously affected by the storm. Elsewhere, in mountainous areas of the country, shelters were swept away by the heavy winds and roads were blocked by mudslides and rock falls. All told, media and government reports attributed twenty-one deaths to the hurricane (Center for Excellence in Disaster Management and Humanitarian Assistance [CFE-DMHA] 2010; Centre for Research on the Epidemiology of Disasters 2014).
Although the damage caused by Hurricane Tomas was much less severe than had been anticipated, and the UN had scaled down the membership of its Disaster Assessment and Coordination (UNDAC) team within a few days of the event, concerns were expressed that the flooding would serve to promote the spread of the nascent cholera epidemic and other water-borne diseases (CFE-DMHA 2010). The concerns seem to have been justified. In one of the largest IDP camps, Parc Jean-Marie Vincent near Port-au-Prince, cases of cholera began to appear on 8 November (Walton and Ivers 2011); elsewhere in the metropolitan area of Port-au-Prince, the epidemic took a hold in the slums of Cit e-Soleil (Piarroux et al. 2011). More generally, the flooding and disruption caused by Hurricane Tomas appears to have had a pronounced effect on the national epidemic curve, with a 121 percent increase in the weekly count of new cholera cases at this time (PAHO 2014a).

Cholera
Cholera is a potentially severe, sometimes rapidly fatal, diarrheal disease due to infection with the bacterium Vibrio cholerae. Transmission of the bacterium usually occurs via the ingestion of fecally contaminated water and, less commonly, food. The majority of infections with V. cholerae are asymptomatic. Clinical illness arises from the ability of the bacterium to produce an enterotoxin that causes the characteristic watery diarrhea ("rice water" stools) seen in cholera patients. Approximately 80 to 90 percent of clinical cases are of mild or moderate severity, with the balance following a severe clinical course. In the latter cases, an incubation period of two to five days is usually followed by the sudden onset of diarrhea and vomiting, giving rise to massive fluid loss and dehydration. Consequent symptoms include cramps, a reduction in body temperature and blood pressure leading to shock, and, ultimately, death within a few hours or days of symptom onset. Mortality is typically witnessed in 40 to 60 percent of untreated cases of severe disease (Tauxe 1998;Heymann 2008).
In terms of the epidemiology of cholera, contaminated watercourses are generally regarded as a prerequisite for a large cholera outbreak in a newly infected (nonendemic) setting. As regards the social environment, cholera epidemics are commonly associated with large population clusters, overcrowding, and unsanitary conditions. Inadequate systems of sewage disposal (leading to fecal contamination of watercourses) and lack of alternative supplies of potable water are integral to the transmission cycle. During the 1990s, for example, extensive waterborne transmission of cholera in Latin America was linked to faulty municipal water systems, contaminated surface waters, and unsafe methods of water storage in domestic settings (Heymann 2008).
Two serogroups of V. cholerae are associated with cholera outbreaks: O1 and O139. V. cholerae O1 has two biotypes (classical and El Tor) that have spread over the last two centuries as seven global pandemic events. The first six pandemics occurred in the nineteenth century and each is believed to have been associated with the classical biotype. The El Tor biotype was first identified in Indonesian pilgrims at the El Tor quarantine station, Egypt, in 1905 and is the causative agent of the current (seventh) global pandemic wave. In contrast, V. cholerae O139 was first identified in 1992 and, although it has been recognized as the cause of epidemics in several countries in South and Southeast Asia, it has yet to establish itself as a pandemic infection (Kaper, Morris and Levine 1995;Heymann 2008).

The Pandemic Sequence of El Tor Cholera
As far as the historical record allows, cholera was wholly unknown in Haiti prior to mid-October 2010 (Jenson, Szabo, and the Duke FHI Haiti Humanities Laboratory Student Research Team 2011). The current epidemic of the disease is an extension of the global pandemic of V. cholerae serogroup O1 (biotype El Tor) that can be traced to a likely onset in the Indonesian island of Sulawesi. Epidemics of El Tor were first observed in southern Sulawesi in the 1930s and, again, in the 1940s and 1950s. Beginning in 1960 and 1961, outbreak activity began to extend to the north of the island and into other parts of the Indonesian archipelago (Mukerjee 1963;Hermann 1973). From there on, the El Tor pandemic spread to mainland Asia (1960s), Africa (1970s), Europe (1970s), and the Americas (1990s; Barua and Greenough 1992). This pandemic sequence has since been characterized by Mutreja et al. (2011) as the product of at least three independent, but overlapping, global transmission waves centered on South Asia that spread outward in the 1960s to the 1980s (Wave 1), 1990s to the 2000s (Wave 2), and 1980s to the 2000s (Wave 3). The spread of each wave has been facilitated by a range of factors, including (1) the enhanced capacity for El Tor vibrios to survive in environmental niches; (2) the relatively mild nature of El Tor cholera and the high frequency of asymptomatic excretors; and (3) the heightened opportunities for disease dispersal with air passenger traffic (Kaper, Morris, and Levine 1995). Available evidence indicates that the Haitian outbreak is aligned with Wave 3 of the pandemic sequence (Mutreja et al. 2011).
The severity of the cholera epidemic in Haiti has been heightened by the variant nature of the epidemic strain of the El Tor biotype in the country and, in particular, its tendency to produce increased levels of enterotoxin and more severe disease outcomes (Piarroux and Faucher 2012). Moreover, the epidemic strain is also resistant to a range of antibacterial agents including co-trimoxazole, furazolidone, sulfafurazole, and streptomycin (Ceccarelli et al. 2011).

Vibrio Cholerae Non-O1/0139 in Haiti
Although one dominant (El Tor) cholera strain has been identified as the root cause of the Haitian outbreak, other serogroups of V. cholerae have contributed to the clinical picture. For example, the normally milder pathogen V. cholerae non-O1/O139 found in U.S. coastal waters could have reached Haiti as a result of the 2010 earthquake, in conjunction with a hot summer, hurricane-related flooding, and poor sanitation (Kupferschmidt 2012). Evidence of serotype switching has also been found. This phenomenon is commonly driven by growing herd immunity to the dominant serotype, rendering people susceptible to reinfection with a new serotype (CDC 2012; Barzilay et al. 2013). Consequently, the bacteriological picture of cholera in Haiti is now more complex than early reports suggested.
The First Cholera Wave: Initial Spread Reconstructions (October-December 2010) The first formal indication of cholera in Haiti can be traced to Tuesday 19 October 2010 when the Haitian MSPP was notified of unusually large numbers of patients from the Artibonite and Centre Departments with acute watery diarrhea and dehydration, in some instances resulting in death (CDC 2010c). These initially affected departments largely escaped earthquake damage ( Figure 3A) and had become the place of refuge for many thousands of earthquake-displaced persons from the Port-au-Prince area (Figure 4). Within four days of these reports, V. cholerae serogroup O1 (serotype Ogawa, biotype El Tor) had been isolated from patients in Artibonite. From there on, cholera spread rapidly out of the initial disease focus to reach all other departments of Haiti by the end of the year (Table 1).

Tracing the Source of the Epidemic
The source of cholera in Haiti is one of the most controversial and politically sensitive aspects of the epidemic. Early hypotheses focused on the possibility of a local origin. It was suggested that the epidemic resulted from the transmission of an environmental strain of V. cholerae to humans and was connected, in some way, to recent tectonic activity in the Gulf of Mexico or prevailing climatic conditions that had promoted the growth of the etiological agent in its environmental reservoir (Cravioto et al. 2011;Piarroux et al. 2011). Toxigenic strains of V. cholerae O1 are known to be present along the Gulf Coast of the United States and cases of cholera have periodically been identified there (A. Ali et al. 2011)-notably in the aftermath of Hurricanes Katrina andRita in 2005 (CDC 2006). As evidence regarding the nature of the epidemic strain of V. cholerae began to mount, though, attention increasingly turned to the possibility of an exogenous origin and the inadvertent introduction of the cholera agent by a human host from a distant geographical source. In particular, recognition that the epidemic was underpinned by the El Tor biotype of V. cholerae and that the Haitian strains were closely related to variant El Tor O1 strains that were circulating in Bangladesh (Chin et al. 2011) andNepal (Hendriksen et al. 2011) pointed to a South Asianpossibly Nepalese-source of the disease ("As cholera returns to Haiti" 2010; Hendriksen et al. 2011). Such a source was consistent with rumors that cholera had been introduced to Haiti by a battalion of Nepalese peacekeepers who were serving with MINUSTAH (Hendriksen et al. 2011).
Informed by the possibility that cholera was introduced from South Asia, separate investigations by a UN-appointed Independent Panel of Experts on the Cholera Outbreak in Haiti (Cravioto et al. 2011) and a joint French-Haitian team of epidemiologists (Piarroux et al. 2011) focused attention on the MINUSTAH camp complex at Mirebalais, Centre Department. The location of the camp, where a new contingent of Nepalese soldiers had arrived from Kathmandu in the period from 8 to 24 October 2010, is shown in Figure 6A. Both investigations pinpointed the camp as the likely source of the epidemic, with the adjacent Meye (or Meille) River (a tributary of the Artibonite River) having possibly been contaminated with fecal waste both directly from the camp and from a nearby septic pit that was used by the camp's local waste contractors. The inferred direction of spread of V. cholerae from these sources, along the Meye River for the town of Mirebalais and the Artibonite River, is shown by the vectors in Figure 6A.

Epidemic Diffusion Routes
It is convenient to conceptualize the subsequent diffusion of the cholera epidemic as a four-stage process.
1. An initial focus in the town of Mirebalais ( Figure 6A). 2. Rapid spread down the Artibonite River to the coastal plain ( Figure 6B). 3. Diffusion out of the Artibonite basin to other parts of the country ( Figure 6C). 4. International transmission to proximal countries.
We consider each stage in turn. To assist the discussion, Table 1 is based on the reports of the MSPP and gives the number of hospital admissions in the departments of Haiti by weekly periods between October and December 2010.  earlier cases is more speculative. Ivers and Walton (2012), for example, noted the case of a young male from Mirebalais who developed cholera-like symptoms on 12 October. The patient had a habit of bathing in the Latem River (fed by the Meye River) and died within twenty-four hours of symptom onset. Shortly thereafter, two attendants who prepared his body for the funeral wake also developed cholera-like symptoms.
Stage 2: Spread Along the Artibonite River Valley ( Figure 6B) From Mirebalais, the Artibonite River served as an efficient vector in the rapid carriage of V. cholerae downstream to the coastal plain ( Figure 6B). Multiple cases of severe diarrhea, vomiting, and dehydration presented more or less simultaneously in hospitals at Deschapelles and Saint-Marc on 20 October (Cravioto et al. 2011), with an explosive spread of the disease in the Artibonite River delta area during the next two days (Piarroux et al. 2011).
Several hydrological features contributed to the rapid spread of cholera in the Artibonite basin. The salinity gradient of the Artibonite River provides optimal environmental conditions for the proliferation of cholera vibrios. Surrounding the river, there was widespread use of river water for washing, bathing, drinking, and recreation, and a number of the early cholera cases were agricultural workers who were exposed to contaminated water from the river (CDC 2010c; A. Ali et al. 2011;Cravioto et al. 2011;Chery, Dodard, and Fournier 2012). Piarroux et al. (2011) noted that, as a consequence of the interaction of irrigation and canals with human use toward the river mouth, the start of the epidemic was explosive in Lower Artibonite. Upstream, between Mirebalais and the Artibonite delta, very few cases were initially reported. Whereas downstream and coastal plain locations were significant risk factors for the contraction of cholera, proximity to the town of Mirebelais was not (Barzilay et al. 2013).
Stage 3: Diffusion from the Artibonite Basin ( Figure 6C) From 22 October, cases of cholera began to be reported in areas that lay beyond the Artibonite basin, including Port-au-Prince and several, mainly mountainous, communes that bordered the Artibonite plain. Following Piarroux et al. (2011), the resulting spread of the epidemic to the end of November 2010 is characterized in Figure 6C as a series of five temporally and spatially staggered clusters of disease activity. Along with the initial foci of disease activity in Mirebalais and the Artibonite delta area (20-28 October), these clusters included the departments of (1) Nord-Ouest (11-29 November), (2) Ouest, including Portau-Prince (14-30 November), and (3) Nord and Nord-Est (21-30 November). The development of these clusters is reflected in the marked increase in reported hospital admissions from Week 4 of the epidemic (Table 1). In contrast, Table 1 shows that the more southerly departments (Grande-Anse, Nippes, Sud, and Sud-Est) remained relatively free of cholera at this time.

Stage 4: International Transmission
The cholera epidemic did not remain confined to Haiti. By mid-November 2010, cases of cholera that were directly linked to the Haiti outbreak had been confirmed in both the Dominican Republic and the United States (Florida). All of the Florida cases and several in the Dominican Republic were among travelers from Haiti. Additional transmission routes into the Dominican Republic were also apparent; a number of early cases were reported among communities on the banks of the Artibonite River close to the Haitian border, with consumption of untreated river water as the most likely source of infection (CDC 2010d).

Summary
By the beginning of 2011, the cumulative count of reported cholera cases in Haiti had exceeded 110,800. Of these, almost 94,800 had been hospitalized and 3,651 had died (PAHO 2014a). As judged by hospitalized cases, Table 1 shows that the major centers of reported disease activity were Nord (23,019 cumulative cases), Ouest (21,806),and Artibonite (20,111) departments; all other departments had reported fewer than 10,000 hospitalized cases. This period marked the zenith of the first-and the largest-of the several waves of cholera activity that spread through Haiti in the months and years leading up to March 2013 ( Figure 2). It is to the spatial structure of this ongoing epidemic activity that we now turn.

Spatial Diffusion Structures (October 2010-March 2013)
To examine the spatial structure of the cholera epidemic in Haiti, the cholera data collated by the Haitian MSPP over a thirty-calendar-month period from October 2010 through March 2013 are analyzed. The analysis is undertaken at two spatial levels: departments (n D 10) and communes (n D 140). We begin by outlining the nature of the disease matrices that can be constructed from the cholera surveillance reports at each spatial level. These matrices are then analyzed using cross-correlation analysis (departments) and coherence analysis (communes).

Cholera Surveillance Data
Prior to the 2010 earthquake, no systematic reporting of diarrheal disease was undertaken in Haiti at either national or regional levels (Cravioto et al. 2011). After the earthquake, two disease surveillance systems were quickly established by the MSPP in conjunction with the PAHO and the U.S. CDC. These were the National Sentinel Surveillance System (NSSS) and the Internally Displaced Persons Surveillance System (IDPSS). The systems were designed to gather data and guide earthquake relief efforts. Although the systems undertook surveillance for acute watery diarrhea, they were not designed to handle an outbreak of the size of the cholera epidemic that began in October 2010. To meet this need, a cholera-specific surveillance system, the National Cholera Surveillance System (NCSS), was established by the MSPP within days of the first detection of cholera. Using a modified WHO clinical case definition for cholera ("acute watery diarrhea, with or without vomiting"), the NCSS collects data at the department and commune levels (Barzilay et al. 2013, 600).
Standardized reports of hospital-and communitybased cholera cases and deaths are available on a weekly basis at (1) the department level from the week ending 23 October 2010 and (2) the commune level from the week ending 23 May 2011. All reports are accessible via the MSPP (http:// www.mspp.gouv.ht) and PAHO (http://new.paho. org/hq) web sites, where additional information is also available for the Dominican Republic.
There are several limitations to the cholera data collected through the NCSS and included in the published reports of the MSPP. First, the surveillance system is facility-based and coverage is partial in remote areas with limited access to health services. Second, the enumeration of community-based deaths is not performed by trained medical personnel. Anecdotal evidence suggests that community deaths due to cholera are likely to be underreported. Third, reporting completeness is sometimes sacrificed for timeliness by communes and, when necessary, antecedent data are updated during the preparation of reports. Finally, the majority of cholera cases in the NCSS are not laboratory confirmed and, inevitably, some cases of watery diarrhea caused by pathogens other than V. cholerae appear in the statistics (CDC 2010c, 2010e; Barzilay et al. 2013).

Cholera Data Matrices
Data abstracted from the MSPP reports were used to create space-time matrices of the weekly incidence of all cholera cases and hospitalized cholera cases. Matrices were built at the department level (n D 10) for a 127-week period (epidemic weeks ending 23 October 2010-23 March 2013) and at the commune level (n D 140) for a 98-week period (epidemic weeks ending 14 May 2011-23 March 2013. The matrices included records for a total of 593,340 cholera cases, of whom 360,748 were hospitalized. The pie charts in Figure 1 show the distribution of these cases by department. Associated epidemic curves, aggregated to the national level, are plotted on a weekly basis in Figures 2A and 2D.

Methods and Results I: Department-Level Analysis
In this section, we examine the manner in which departments have interacted with each other as the cholera epidemic has diffused through them. For this purpose, we follow Smallman-Raynor and Cliff (1998bCliff ( , 1999 in adopting one powerful technique for characterizing such spatial interactions in an epidemiological framework, namely, cross-correlation analysis. Full details of the statistical approach are provided in Appendix A but, in essence, the weekly counts of cholera incidence (new cholera cases and new cholera admissions to hospital) for each of the ten departments in the interval to March 2013 were treated as a time series. Cross-correlation analysis then proceeds by computing the degree of association (r k ) between any two cholera time series that are k time lags (here, weeks) apart. The time lag at which the maximum association occurs is taken as the lead or lag of the cholera epidemic in one department with respect to another.

Results
The principal results of the cross-correlation analysis are summarized in Table 2 and Figure 7A. Table 2 gives the average value of the maximum association (r k ) and the average weekly lead (positive values of k) or lag (negative values of k) of the named department with all other departments. To assist in the interpretation of the table, the maps in Figure 7A link each department to the department with which it shared the highest association. The information is shown for all cholera cases (maps i and ii) and hospitalized cholera cases (maps iii and iv). Additional facets of Figure 7A are summarized in the figure caption and Appendix A. Four interlinked features of the spatial structure of the cholera epidemic in Haiti emerge from Table 2 and Figure 7: (1) the tendency for Artibonite and the northern departments to lead the development of the epidemic; (2) the tendency for the southern departments to lag the development of the epidemic; (3) the implied north-south divide in epidemic transmission that follows from features 1 and 2; and (4) the role of hierarchical leaps in the spatial transmission structure. We consider each feature in turn. Table 2 shows positive values of k between northern departments (with the exception of Nord-Est) and other departments, whereas southern departments have negative or close to zero k with other departments. This implies that northern departments led the development of the epidemic. Artibonite has both the largest average association with all other departments and the longest lead. This is consistent with the likely role of Artibonite as an index location for the spread of cholera to adjacent (Nord, Nord-Ouest) and other departments. The lead-lag structure for hospitalized cholera cases in maps iii and iv of Figure 7A also reflects this.
In Figure 7A, the southernmost departments (Grand-Anse, Nippes, Sud, and Sud-Est) are The maps are based on the results of the cross-correlation analysis and link each department to the department with which it shared the highest association. The information is shown for all cholera cases (maps i and ii) and hospitalized cholera cases (maps iii and iv), with the map pairings based on analysis using raw (maps i and iii) and first differenced (maps ii and iv) time series. Vector widths define the strengths of the associations between departments. In-phase (reciprocal links; k D 0) and out-of-phase (unidirectional links; k 6 ¼ 0) are plotted. Vector colors define the lags, k. characterized by generally weak associations with the northern departments, consistent with their position in the lead (northern departments)-lag (other departments) structure of epidemic development. Two further features also emerge from Figure 7A: (1) the relatively high level of local epidemiological integration of the southernmost departments implied by the large numbers of associations with each other and, in turn, (2) their connections with the epidemic in Portau-Prince and Ouest. This is consistent with an epidemic that, in the southernmost districts, has been driven by the developments in the metropolitan area.
The evidence in Table 2 and Figure 7A suggests a north-south split dividing Haiti into two diffusion islands, with the south firing later than the north. There appears to be a watershed in the spread of cholera along the Chaîne des Matheux hills that run northwest-southeast between the Artibonite River and Port-au-Prince (see Figure 1 for location). Taking Artibonite Department as the reference series, its average lag with the set of southern districts (Grande Anse, Nippes, Ouest, Sud, and Sud-Est) in Table 2 implies that this watershed was crossed after four to six weeks.
Apparent long-distance "leaps" of cholera between northern and southern departments are implied by the prominent bonds in Figure 7A, including several that involve the index department (Artibonite) with the southernmost departments of Sud (map iii) and Grande-Anse (map iv). Similar associations are noted for Nord-Est with Sud-Est (map i), Grand-Anse (map iii), and Nippes (map iii) and for Centre with Nippes (map iv). Caution, however, should be exercised in the interpretation of these findings. Although it is tempting to assume that they are indicative of long-distance transmission episodes, possibly associated with the population movements engendered by the earthquake and the epidemic, some of the associations might reflect the confounding effects of delayed epidemic onset in peripheral departments of the country.
A schematic model of the diffusion process implied by Table 2 and Figure 7A is proposed in Figure 7B.

Methods and Results II: Commune-Level Analysis
To examine the geographical structure of cholera activity at the commune level, we draw on the concept of geographical coherence in epidemiological investigations (Cliff et al. 1992). As described in Appendix B, the weekly time series of reported cholera cases in each of the communes of Haiti can be correlated with each other to see how far the temporal behavior of one geographical unit corresponds with that of another. Pairs of communes that have very similar cholera time series can be described as having strong geographical coherence; those with dissimilar series have weak geographical coherence. For a given department, we take the average correlation (r) between all pairs of constituent communes as a measure of internal coherence for that department. Figure 8 summarizes the results of the coherence analysis for all cholera cases (A) and hospitalized cases (B). Both maps identify the northern departments of Artibonite and Nord (r 0:4) and the southern departments of Grande-Anse and Sud (r 0:3) as having the highest levels of internal coherence, indicative of groups of communes that display very similar patterns of reported disease activity. Lower levels of coherence are recorded for all other departments, although modest variations in the categorized values of r are evident between the two maps.

Results
The relatively high levels of internal coherence identified for Artibonite, Nord, Grand-Anse, and Sud are consistent with the operation of these departments as integrated epidemiological systems. Here, the primary vectors of V. cholerae (human carriers and watercourses) have resulted in an efficient epidemiological linkage of the constituent communes to yield very similar patterns of disease activity in time. In contrast, the relatively low levels of internal coherence identified for some other departments, including the major epidemic focus of Ouest, are consistent with the operation of weakly integrated epidemiological systems. In these systems, the primary vectors of V. cholerae have-for social, physical, technological, or other reasons-been limited in their capacity to link efficiently all the constituent communes so that the geographical extent or rate of disease transmission has been correspondingly curtailed. By way of a simple illustration of the phenomenon, the comparatively low coherence for Nord-Ouest could reflect (1) the relative isolation of more remote communes from the primary road network and (2) the role of the sea as a physical barrier in the transmission of cholera from the mainland communes to the island commune of La Tortue (Figure 1). It is reasonable to assume that the physical geography of Haiti ( Figure 7C) has played a more general role in the observed levels of epidemiological coherence, with physical barriers serving to delay or halt the spread of cholera to the more remote communes of some departments.

Conclusion
The current epidemic of cholera in Haiti is in urgent need of investigations that will lead to effective interventions for the control and, ultimately, the elimination of the disease. An improved understanding of the spatial and temporal dimensions of the epidemic is one approach toward achieving this goal (Andrews and Basu 2011;Bertuzzo et al. 2011;Chao, Halloran, and Longini 2011;Mukandavire, Smith, and Morris 2013). We suggest that geographers are ideally positioned to contribute to this pressing area of health research. As a reference point for future investigations, this article has attempted to introduce geographers to aspects of the origin and course of the epidemic and, importantly, to the substantial body of associated literature that has accrued since October 2010.
Our own analysis of the MSPP's morbidity data has added to a geographical understanding of the cholera epidemic in terms of both the spatial macrostructure of epidemic development (Figure 7 and Table 2) and the differential levels of epidemiological integration of the administrative units of Haiti (Figure 8). In an earlier study, Tuite et al. (2011) used a compartmental modeling strategy with a gravity model component to simulate the spatial transmission of cholera in the initial stages of the Haiti epidemic. Their analysis identified a clear tendency for the "gravitational effects" of population size and geographical proximity to reproduce the interdepartment dynamics of epidemic transmission. Although our use of cross-correlation analysis has underscored the contagious and hierarchical transmission elements implied by the gravity modeling approach of Tuite and colleagues (Figure 7), we have further demonstrated that the evolution of the epidemic to its pattern in March 2013 was underpinned by a fundamental geographical divide between the northern and southern departments of the country. This north-south divide manifested as a spatial leadlag structure (Artibonite and northern departments ! Ouest ! southern departments), with the capital department (Ouest) serving as a pivot point in the southward spread of the epidemic ( Figure 7B).
The National Plan of Action for the Elimination of Cholera in Haiti 2013-2022 foresees a ten-year project to eliminate cholera from Haiti and the Dominican Republic (MSPP 2013). Funds will be used to improve the supply of potable water and sanitation infrastructure, thus breaking cholera's fecal-oral transmission route. As control strategies move toward the elimination of the disease in localized areas of the country, the possibility of disease reimportations into cholerafree localities demands knowledge of those spatial channels through which such reimportations might occur. In this context, the results of our coherence analysis have pointed to the existence of pronounced differences in the degree of epidemiological integration of departments (Figure 8). An understanding of those factors that have governed these differential levels of epidemiological integration, and that serve to promote or impede the efficient spatial transmission of V. cholerae from one commune to another in a given department, might serve to guide the spatial planning and implementation of future intervention strategies (MSPP 2013).
The MSPP's plan to eliminate cholera from Haiti by the year 2022 indicates that the disease will remain a problem in the country in the medium term. The elimination program incorporates a strategy in which health services will be rolled out so that everyone lives within a "reasonable distance" of a health post (MSPP 2013, 4). If successful, this plan will equip Haitians with much-needed and long-awaited basic infrastructure that will have health and social benefits that go beyond the control of cholera. Geographers have an important role to play in the promotion of this broader development process.
where x i and x j are the means of the time series. In the notation of Equation A.1, the {x it } are termed the reference series and the {x jt } the comparison series. If k D 0, the reference and the comparison series are said to be in phase; Ck (k > 0) signifies that the reference series leads the comparison series, and -k (k < 0) that the reference series lags the comparison series. Finally, a plot of the correlation coefficient, r k , against the lag k yields the cross-correlation function (CCF).
The weekly series of cholera incidence in each of the ten departments were systematically treated as the reference series in Equation A.1 against which the remaining nine series were compared, thereby yielding a total of (10 £ 9 D) 90 CCFs for (1) all cholera cases and (2) hospitalized cholera cases. A basic assumption of crosscorrelation analysis is that the series to be analyzed are time-stationary (Chatfield 2003). Accordingly, prior to analysis, all series were detrended using techniques of first differencing (Gottman 1981). In this article, we report the results of the cross-correlation analysis as performed for both raw and detrended data.
The principal results of the cross-correlation analysis are summarized in Table 2, which gives the average value of the maximum correlation (r k ) and the average lead-lag in weeks (k) of a given department (reference series; x it in Equation A.1) with all other departments (comparison series; x jt in Equation A.1) over lags -12 k 12. Here, Ck indicates that, on average, the reference department leads the comparison departments by k weeks. Conversely, ¡k indicates that, on average, the reference department lags the comparison departments by k weeks. The maps in Figure 7A link each department to the department with which it shared the highest value of r k at lags 0 k 12 (i.e., the pattern of leads). Vector line widths define the strengths of the correlations. In-phase (reciprocal links; k D 0) and out-of-phase (unidirectional links; k 6 ¼ 0) are illustrated. Vector line colors define the lag, k.