Home
The University of Adelaide
| The University of Adelaide |
Home |
Departments |
Search
|
 |
||
 |
 |
  |
 |
||||||||||||||||||||||||||||||||
|
|
Modelling Cooling Tower Risk for Legionnaires' Disease using Bayesian Networks and Geographic Information SystemsWilmot, P 1 ., Pradhan, M. 2 , Bentham, R. 3 , Hakendorf, R. 4 1 GISCA Adelaide 2 Health Informatics, University of Adelaide . 3 Department of Environmental Health, Flinders University 4 Epidemiology Unit, Flinders Medical Centre In Australia Legionnaires' Disease affects 250 persons annually with the source of infection often never found. Associations between Legionnaires' Disease and cooling towers have been well documented and vary geographically. Exposure of Legionella pneumophila is thought to originate from an amplified source and released in the form of an aerosol into the air where susceptible persons can inhale this bacteria. The project establishes a Bayesian Belief Network (BBN) to model the uncertainty of aerosols released from cooling towers and Geographic Information System (GIS) to create a wind dispersion model and identify potential cooling towers as the source of infection. The result was an application that identified cooling towers that are potentially the source of infection given the location of the individual. A cooling tower register was established over a five year period and past cases were incorporated into a desktop GIS. Identifying these potential sources may reduce future numbers of sporadic cases each year. It also allows environmental health officers to locate, identify, inspect and sample cooling towers in a manner faster than current practices. The project used census information to determine population densities surrounding cooling towers and identified cooling towers that have a proportionally higher risk of infecting the population based on the demographics of persons surrounding the towers. The paper has demonstrated the use of GIS and BBN in environmental epidemiology and the power of spatial information in the area of health. IntroductionThe recent outbreak of Legionnaires' disease at the Melbourne Aquarium has prompted widespread interest in the growth of Legionella in cooling towers. Cooling towers are heat rejecting devices which are usually associated with air conditioners and generally found on top of buildings (Figure 1). Cooling towers recycle and cool water through an evaporative process. The aquatic environment is ideal for Legionella bacteria to grow. If a person inhales the bacteria deep into the lungs then the chance of a bacterial infection increases. The person may contract what is commonly called Legionnaires' Disease. The source of infection is often never found. Published reports over the last 24 years since Legionnaires' disease was first recognised have detailed a number of risk factors. There have been very few published reports in which Legionella culture results from cooling towers have been equated to a quantifiable risk of infection[ 1, 2] . Of these reports, none has been based upon multiple culture results taken from single systems over prolonged periods. As such there is a lack of a scientific support for Legionella culture as a risk management tool. To improve the detection and prediction of Legionalla outbreaks we developed a decision support system that integrates a Bayesian Belief network (BBN) and a Geographical Information System (GIS). This system considers information about the Legionalla bacteria, and the factors that promote its growth in cooling towers.
Figure 1. Cross section of a cooling towers circulation system
Source : The National Occupational Health and Safety Commission booklet titled "Legionnaires Disease and Related Illnesses", 1989, p9. BackgroundPublished reports over the last 24 years since Legionnaires' disease was first recognised have detailed a number of risk factors. There have been very few published reports in which Legionella culture results from cooling towers have been equated to a quantifiable risk of infection[ 1, 2] . Of these reports, none has been based upon multiple culture results taken from single systems over prolonged periods. As such there is a lack of a scientific support for Legionella culture as a risk management tool. Cooling tower maintenance and biocide dosage have been shown to be major contributors to risk management and review papers have stated that with proper management Legionella can be readily controlled [3] . Correctly fitted and designed drift eliminators are important in minimising the release of bacteria in aerosol from the systems. Regular cleaning regimes and inspection and review of the cooling towers and water treatment equipment are also important aspects of a control strategy [ 4,5] . Operation of cooling towers, and cooling water temperature have been shown to positively correlate with concentrations of Legionella in cooling towers. Intermittent operation has been proposed as a common factor in many cooling tower associated outbreaks [6]. The size and surface area to volume ratio of the cooling water systems are also significantly associated with risk of outbreaks. It is notable that outbreaks of Legionnaires' disease have not been associated with large cooling water systems [6] . Prevailing weather conditions such as cloud cover, ultra violet light intensity, air temperature and relative humidity have been shown to influence the survival and dispersion of Legionella e in aerosol. In many cases wind direction has been a major factor determining whether significant exposure to aerosol could occur [ 7,8] . Climatic conditions determine how far the viable bacteria may be transmitted from the cooling tower. Air temperature and humidity may also influence operation and cooling water temperature [6] . The above criteria have all been the subject of review articles [ 3,9] . These factors are interrelated and combine to create a risk of multiplication and dissemination of the bacterium. These factors can be identified as risks without the need for Legionella culture. A further confounding factor in predicting risk has been the identification of susceptible populations. Living in proximity to a cooling tower has been demonstrated to increase risk of sporadic cases of Legionnaires' disease[ 10] . However, the demographic identification of susceptible populations and their proximity to cooling towers has not been addressed outside the context of nosocomial infection. Model constructionA literature review was conducted into the known factors regarding the growth and spread of Legionella in cooling towers. The creation of the Belief Network Model was undertaken using the GeNIe software. The belief network consisted of nodes representing uncertain events or random variables and arcs which connect these nodes or dependencies. The arcs represented directions of probabilistic dependency between the parent node and that of the child. Figure 2 illustrates nodes as circles or ovals and the arcs that connect these nodes together. The structure of the model is graphical and qualitative, used to show the interactions between variables. The model was based on factors affecting Legionella growth in cooling towers and cooling tower risk supported in the literature. The qualitative model was developed and later verified by Bentham, R.H.(pers.comm.)
Figure 2. Final Belief Network Model
The GeNIE software allowed the user to enter conditional probability statements with values that reflected beliefs. Each node had a minimum of two states, state0 and state1. The state for the humidity node, for example was either high humidity or low humidity. Other nodes were similarly represented as the true or false argument ie biocides were used or not used. Given prior information (knowledge) about cooling towers a model of the relative aerosol risk from a cooling tower could be developed. After the Belief model had been constructed and the conditional probabilities entered, the model could be tested. Given the worst case scenario, the evidence for the following nodes were set.
The Belief Network also allowed the user to develop guidelines for reducing the potential aerosol risk of a cooling tower, for example, if prior knowledge about the cooling tower existed, such as:
The probability of a case given the cooling tower was based on three conditions.
Dispersion modelWe constructed a binormal plume dispersion model to update the probability of a cooling tower infection given a case of Legionella. If a case falls within the dispersion model then there is a chance that the Legionella bacteria could have travelled from that cooling tower. Figure 3 indicates a case within the plume dispersion model from a cooling tower.
Figure 3. A case within the dispersion model
If a case falls outside the extent of the dispersion model, the chance that it originated from that cooling tower remains highly unlikely and is disregarded as a possible source of infection. Figure 4 indicates the case outside the wind dispersion model.
Figure 4. A case outside the dispersion model
The probability of cooling tower risk P(CT) was calculated using the Bayesian Belief Network, and the probability of a case given cooling tower P(Case | CT) calculated using the formula derived in Equation 3-7. The P(Case) as previously mentioned is a constant over this area. Using Bayes' Rule, we could multiply the P(Case | CT) with P(CT) to satisfy the requirements of Bayes' Rule and determine the P( Cooling Tower | Case). The resulting probabilities were calculated for each cooling tower for each case that fell within the dispersion model. The probability of a cooling tower as the source of infection given a case can be calculated and viewed within the GIS system. This result could be updated for any additional cases added. Population risk and inspection policyA cooling tower only represents a risk if there is a population surrounding it. The degree of risk depends on the amount of aerosols being released from the cooling tower and how susceptible the population is that surrounds the cooling tower. A cooling tower with extreme Legionella levels with no population surrounding the cooling tower does not possess a risk of infection to the population. However, in a possible worst case scenario, if there exists a high density of persons surrounding the cooling tower given the same tower aerosol risk, then there is a potentially larger risk of infection to the population. Persons most at risk of infection, based on past outbreaks and literature find men 2.5 times more likely to contract the disease than females [11] . Persons with depressed immune systems, are smokers, and greater than 60 years of age are at a higher risk of infection. For the purposes of this study, only the numbers of men and women by age in a Collection District were used, based upon the 1996 Australian Census. Furthermore, given the population surrounding the cooling tower an Infected Risk Index (IRI) could be established based on the number of males and females over 60 years of age. If additional data became available such as the locations of immunosuppressant persons or locations of hospitals etc then this population at risk could be updated. The Population at Risk or Infected Risk Index was based on the population surrounding the cooling tower and the cooling tower risk P(CT | Case).The IRI calculated every person within a 1.5 km radius of a cooling tower was infected by Legionella and died, resulting in the total loss of the population (Figure 5). The result ranked individual cooling towers based on the surrounding population and the risk of that cooling tower.
Figure 5. Population selected based on 1.5 km radius around a cooling tower
The resulting output from the system is a recommended ranked list for inspection of cooling towers that will minimise harm to the community (Figure 6).
Figure 6. An Infective Risk Index (IRI) based on population surrounding the cooling towers.
DiscussionWe have described the integration of a statistical model with a GIS, along with demographic data regarding the proximity of susceptible populations. The GIS used data from a register of cooling towers located in the Adelaide Metropolitan Area, South Australia . These data combined with wind directions and plume dispersions are used to predict cooling towers with the greatest likelihood of association with a case of Legionnaires' disease. Inputting patient movement data regarding location into the model enables ranking of the most likely cooling tower associated with a given case. In an outbreak situation the system identifies the most likely cooling tower sources of infection after each case is entered. Using the register of cooling towers, the contact details for the owner / operator and exact location of the system can be accessed immediately. This provides a desktop investigation system capable of locating most likely sources of outbreaks within minutes of accessing patient data. We plan to validate this model using historical data, but we could also evaluate the efficacy of the system using a prospective design if funding were available. The model could also provide a comprehensive surveillance system for cooling tower maintenance. It could be quite readily adapted to produce a web-based surveillance system Australia-wide, with access for local councils and state health departments. Routine maintenance procedures by owner/operators and water treatment personnel could be electronically reported to the system and risk assessments for each system could be made. The system provides a risk assessment based upon scientifically established risk factors. A cooling tower in a given location can be assessed for risk associated with Legionnaires' disease without using culture for the organism. More pertinently, the system can predict quantifiable risk on a daily basis according to climatic conditions and cooling tower specifications; it will modify risk according to system maintenance and operation characteristics. Unreliability in Legionella culture is well established, and the low reliability in assessing risk from these results, has probably been the reasoning behind not instituting compulsory testing. Unknowns regarding virulence within species and strains, and host immune status further complicate culture result interpretations. We suggest that, though culture may be valid as a support to a maintenance strategy, a system based on established risk factors is more functional as a surveillance, risk assessment and management tool. The economic advantages of an interactive computer based GIS surveillance system over routine sampling protocols and conventional outbreak investigations would be substantial. References
|
|
|||||||||||||||||||||||||||||
