## Abstract

Haiti is in the midst of a cholera epidemic. Surveillance data for formulating models of the epidemic are limited, but such models can aid understanding of epidemic processes and help define control strategies.

To predict, by using a mathematical model, the sequence and timing of regional cholera epidemics in Haiti and explore the potential effects of disease-control strategies.

Compartmental mathematical model allowing person-to-person and waterborne transmission of cholera. Within- and between-region epidemic spread was modeled, with the latter dependent on population sizes and distance between regional centroids (a “gravity” model).

Haiti, 2010 to 2011.

Haitian hospitalization data, 2009 census data, literature-derived parameter values, and model calibration.

Dates of epidemic onset and hospitalizations.

The plausible range for cholera's basic reproductive number (R_{0}, defined as the number of secondary cases per primary case in a susceptible population without intervention) was 2.06 to 2.78. The order and timing of regional cholera outbreaks predicted by the gravity model were closely correlated with empirical observations. Analysis of changes in disease dynamics over time suggests that public health interventions have substantially affected this epidemic. A limited vaccine supply provided late in the epidemic was projected to have a modest effect.

Assumptions were simplified, which was necessary for modeling. Projections are based on the initial dynamics of the epidemic, which may change.

Despite limited surveillance data from the cholera epidemic in Haiti, a model simulating between-region disease transmission according to population and distance closely reproduces reported disease patterns. This model is a tool that planners, policymakers, and medical personnel seeking to manage the epidemic could use immediately.

None.

### Editors' Notes

#### Context

Haiti is in the midst of a cholera epidemic. Surveillance data to inform public health decision making are limited.

#### Contribution

The authors constructed a mathematical model of epidemic dynamics that is based on both population and distance. The model's results closely match the contour of the epidemic to date. The model was used to project the probable effect of different approaches to allocation of vaccines and clean water on the course of the epidemic.

#### Caution

The model does not include the effect of antibiotic treatment on transmission of cholera.

#### Implication

A publicly available tool to assist in managing the cholera epidemic in Haiti has been developed and can be further modified and refined.

## Methods

### Data Sources

### Model

^{4}.

### Model Parameterization

_{0}) is an important index of epidemic potential for a communicable disease (10–12, 20). Existing estimates of R

_{0}for cholera vary widely (21–23). We obtained seed estimates from the published biomedical literature of the duration of infectiousness in patients with cholera and durability of

*Vibrio cholerae*in source water (18). We assumed that cholera could be transmitted through either contaminated water or close contact but that waterborne transmission (or consumption of food items contaminated with infective water) was a far more important method of transmission (24–26). Most documented cases of person-to-person transmission occur in households, usually because of infection in persons involved in food preparation (27, 28).

*V. cholerae*were derived from studies of bacterial survival in sediments (29). Population sizes were obtained from 2009 Haitian census data (30). We used centroids for each department's capital city to calculate straight-line distances between patches (Appendix Figure 2). To account for the shape of the country, we assumed that Haiti was divided into 2 zones when calculating distances: the lower zone included Grand'Anse, Sud, Nippes, and Sud-Est, and the upper zone included Nord-Ouest, Nord, Nord-Est, Artibonite, and Centre. Ouest was treated as a conduit between the 2 zones, such that the distance between 2 departments in different zones was calculated as the sum of the distance between the department in zone 1 and Ouest and between Ouest and the department in zone 2. For example, the distance between Grand'Anse and Centre was the sum of the distance between the capital of Grand'Anse and the capital of Ouest plus the distance between the capital of Ouest and the capital of Centre.

### Model Calibration

*d*) to best reproduce the ordering of case appearance by department (Appendix Table 1). We considered values of

^{−}^{n}*n*that ranged from 0.8 to 3.0. Goodness of ordering was evaluated by calculating Spearman correlation coefficients for observed versus expected order of first case appearance in the 9 non-Artibonite departments. Ordering was optimized for

*n*ranging from 2.0 to 2.4 (and was identical for

*n*in this range). For simplicity, we used an

*n*of 2.0. We subsequently optimized model fit through iterative adjustment of components of R

_{0}, as well as κ (the “gravitational constant”) and [xgr ] (the pathogen decay rate in water) to minimize the least-squares distance between overall model hospitalization estimates and those reported by MSPP, by using the optimize function of the Berkeley Madonna software package. We estimated credible intervals for model parameters by iteratively refitting our multipatch susceptible–infectious–water–recovered model, assuming that errors in case counts were Poisson-distributed (20).

### Optimization of Control Strategies

### Changing Dynamics of Infection Over Time

*f*) (34), such that the reproductive number (

*R*) at time

*t*was taken to be R

_{t}= R

_{0}/(1 +

*f*)

^{t}.

### Role of the Funding Source

## Results

### Model Parameterization and Calibration

_{0}for cholera was approximately 2.78 (plausible range, 2.06 to 2.78), which is within the range generated by our group and others (21–23). Using an inverse square-distance term in our gravity model combined with best-fit parameters, we found close correlation between model projections of ordering initial cases and initial case dates by department and between those reported by the MSPP and the U.S. Centers for Disease Control and Prevention (15) (Spearman correlation coefficient for ordering, 0.97 [

*P*< 0.001]; Spearman correlation coefficient for case dates, 0.92 [

*P*< 0.001]) (Figure 1). Cumulative hospitalized case counts projected by the model agreed closely with those reported by the MSPP (Appendix Figure 3).

### Estimated Time to Peak and Duration of Epidemic Under Initial Conditions

**Supplement**).

### Optimal Intervention Strategies

*top*). Nonetheless, the benefit associated with vaccination was far greater than that of allocation of clean water (Figure 3,

*bottom*). We estimated that 1.7 to 2.0 times as many people would need to be given access to clean water to achieve an equal reduction in cholera cases that can be achieved through optimal allocation of vaccine (Appendix Table 3). The combination of clean water and vaccination was projected to have a superadditive effect (projected reduction in cases resulting from both interventions was greater than the sum of the effects of individual interventions) (Appendix Figure 4).

### Changing Dynamics of Infection Over Time

_{0}(2.90), with a decrease in R

_{0}by an average of 1.8% per day (Figure 4). By contrasting this epidemic with simulations using an R

_{0}of 2.78 (as in the base case) or 2.90, we estimate that disease-control interventions have probably prevented thousands of cases of cholera (as of February 2011).

## Discussion

*V. cholerae*strain responsible for the current epidemic was presumably introduced by a person or persons visiting Haiti from South Asia (38). Regardless of the route of introduction, the limited availability of clean drinking water and sewage treatment in Haiti have permitted the rapid emergence of a cholera epidemic that is now active in all regions of the country. As of February 2011, more than 4000 cholera-related deaths have been recorded (2).

*initial*epidemic dynamics in the absence of an intervention; the animated representation of the course of regional epidemics (

**Supplement**) ignores control measures, which are likely to reduce the severity of regional epidemics (and may already be doing so, thanks to the hard work and dedication of government health workers and local and international relief workers). Our subsequent refitted model suggests that these efforts have resulted in a substantial decrease in the transmissibility of cholera in Haiti.

## References

**International Monetary Fund and International Development Association.**Haiti: Enhanced Initiative for Heavily Indebted Poor Countries—Completion Point Document. IMF Country Report No. 09/288. Washington, DC: International Monetary Fund; 2009. Accessed atwww.imf.org/external/pubs/ft/scr/2009/cr09288.pdfon 17 December 2010.**Documentation of the Haitian cholera epidemic**, 2010–11. Ministère de la Santé Publique et de la Population, Republique d'Haiti. Accessed athttp://mspp.gouv.ht/site/index.php?option=com_content&view=article&id=57&itemid=1on 10 January 2011.- Centers for Disease Control and Prevention (CDC)Update: cholera outbreak—Haiti, 2010.MMWR Morb Mortal Wkly Rep20105914739
- Centers for Disease Control and Prevention (CDC)Cholera outbreak—Haiti, October 2010.MMWR Morb Mortal Wkly Rep2010591411
**Sontag D.**In Haiti, capital braces for a cholera outbreak. New York Times. 24 October 2010:A1. Accessed atwww.nytimes.com/2010/10/25/world/americas/25haiti.htmlon 6 January 2011.**Knox R.**Doctors urge cholera vaccine for Haiti, neighbors. NPR.org. 10 December 2010. Accessed atwww.npr.org/2010/12/10/131950133/doctors-urge-cholera-vaccine-for-haiti-neighborson 15 December 2010.**ProMED-mail subject PRO/EDR>Cholera—Haiti (28)**: update & Dominican Rep. Archive number 20101214.4441. Published 14 December 2010. Accessed atwww.promedmail.orgon 5 January 2011.- ,MishraS,FismanDN.BoilyMCThe ABC of terms used in mathematical models of infectious diseases.J Epidemiol Community Health2011658794
- .FismanDPandemic Influenza Outbreak Research Modelling Team (Pan-InfORM)Modelling an influenza pandemic: a guide for the perplexed.CMAJ20091811713
- ,AndersonRM.MayRMInfectious Diseases of Humans: Dynamics and Control.New YorkOxford Univ Pr1991
- ,ViboudC,BjørnstadON,SmithDL,SimonsenL,MillerMA.GrenfellBTSynchrony, waves, and spatial hierarchies in the spread of influenza.Science200631244751
- ,CrépeyP.BarthélemyMDetecting robust patterns in the spread of epidemics: a case study of influenza in the United States and France.Am J Epidemiol2007166124451
- Centers for Disease Control and Prevention (CDC)Update: outbreak of cholera—Haiti, 2010.MMWR Morb Mortal Wkly Rep201059158690
- ,BrownsteinJS,FreifeldCC,ReisBY.MandlKDSurveillance Sans Frontières: Internet-based emerging infectious disease intelligence and the HealthMap project.PLoS Med20085151
- ,KeelingMJ.RohaniPModeling Infectious Diseases in Humans and Animals.Princeton, NJPrinceton Univ Pr200923390
- ,TienJH.EarnDJMultiple transmission pathways and disease dynamics in a waterborne pathogen model.Bull Math Biol201072150633
- ,TuiteAR,GreerAL,WhelanM,WinterAL,LeeB. et al.YanPEstimated epidemiologic parameters and morbidity associated with pandemic H1N1 influenza.CMAJ20101821316
- ,PascualM,KoelleK.DobsonAPHyperinfectivity in cholera: a new mechanism for an old epidemiological model? [Letter].PLoS Med20063280
- ,TienJH,PoinarHN,FismanDN.EarnDJHerald waves of cholera in nineteenth century London.J R Soc Interface2010
- ,Miller NeilanRL,SchaeferE,GaffH,FisterKR.LenhartSModeling optimal intervention strategies for cholera.Bull Math Biol201072200418
- .TauxeRCholera.Evans AS, Brachman PSBacterial Infections of Humans: Epidemiology and Control. 3rd ed.New YorkKluwer Academic/Plenum Publishers199822342
- .GuthmannJPEpidemic cholera in Latin America: spread and routes of transmission.J Trop Med Hyg19959841927
- ,McIntyreRC,TiraT,FloodT.BlakePAModes of transmission of cholera in a newly infected population on an atoll: implications for control measures.Lancet197913114
- ,RabbaniGH.GreenoughWB3rdFood as a vehicle of transmission of cholera.J Diarrhoeal Dis Res19991719
- ,AlbertMJ,NeiraM.MotarjemiYThe role of food in the epidemiology of cholera.World Health Stat Q1997501118
- ,HoodMA,NessGE.RodrickGEIsolation of
*Vibrio cholerae*serotype O1 from the eastern oyster,*Crassostrea virginica*.Appl Environ Microbiol19814155960 **Institute Haitien de Statistique et d'Informatique.**Population totale, population de 18 ans et plus, menages et densites estimes en 2009. Port-au-Prince, Haiti: Institute Haitien de Statistique et d'Informatique; 2009. Accessed atwww.ihsi.ht/pdf/projection/poptotal&menagdens_estim2009.pdfon 11 January 2011.- ,GravesPM,DeeksJJ,DemicheliV.JeffersonTVaccines for preventing cholera: killed whole cell or other subunit vaccines (injected).Cochrane Database Syst Rev2010CD000974
- .Canadian Immunization Guide.OttawaPublic Health Agency of Canada200615865
- ,JohannessonM,PliskinJS.WeinsteinMCA note on QALYs, time tradeoff, and discounting.Med Decis Making19941418893
- ,GriffithDC,Kelly-HopeLA.MillerMAReview of reported cholera outbreaks worldwide, 1995-2005.Am J Trop Med Hyg2006759737
- ,ChinCS,SorensonJ,HarrisJB,RobinsWP,CharlesRC. et al.Jean-CharlesRRThe origin of the Haitian cholera outbreak strain.N Engl J Med20113643342
- ,BhartiN,XiaY,BjornstadON.GrenfellBTMeasles on the edge: coastal heterogeneities and infection dynamics.PLoS One200831941
- ,EggoRM,CauchemezS.FergusonNMSpatial dynamics of the 1918 influenza pandemic in England, Wales and the United States.J R Soc Interface2011823343
- ,IversLC,FarmerP,AlmazorCP.LéandreFFive complementary interventions to slow cholera: Haiti.Lancet2010376204851

### Appendix

#### Model Details

*i*th patch (with

*i*from 1 to 10):

*s*,

_{i}*x*, and

_{i}*r*are the proportions of the population in the susceptible, infectious, and recovered (immune) or removed (through cholera-specific mortality) states, respectively, and

_{i}*w*is the concentration of

_{i}*V. cholerae*in local waters, rescaled by using the parameter [xgr ] to reflect the burden of infection in the population. Rescaling is described in detail by Tien and Earn (18). The parameter γ is the inverse of the mean duration of infectiousness (that is, the recovery rate in persons who survive and time to death in those who do not), and µ is the birth rate and noncholera mortality rate, which can effectively be ignored on the short time frame associated with this epidemic.

*i*th patch is:

*and β*

_{Xi}*represent the transmission rate from infectious persons and water in the*

_{wi}*i*th patch to susceptible persons in the

*i*th patch, and [thgr ]

*represents the influence of infection prevalence in the*

_{ij}*j*th patch on incidence in the

*i*th patch, according to the relation:

*p*and

_{i}*p*are population sizes,

_{j}*d*is the distance between 2 patches, and

*n*is a power that determines the strength of the dependence of transmission rate on distance.

#### Model Parameterization

_{0}for cholera can be expressed in terms of infectiousness of water sources and infectiousness of person-to-person contact, and it is inversely related to the rate of recovery from disease and the all-cause mortality rate in the population (18), such that for a single-patch model:

*and µ are small relative to β*

_{I}*and γ:*

_{w}#### Optimization of Control Strategies

##### Vaccination

*v*is the probability of vaccination within a department.

_{i}##### Provision of Clean Water

*c*is the probability of provision of clean water within a department. As with vaccination, there was no preferential provision of clean water to susceptible persons.

_{i}
## This feature is available only to Registered Users

Subscribe/Learn More## 0 Comments