The use of discriminant analysis in the detection of geographic types of asthma

The use of discriminant analysis in the detection of geographic types of asthma

THE USE OF DISCRIMINANT DETECTION OF GEOGRAPHIC ANALYSIS IN THE TYPES OF ASTHMA G. G. GILES Department of Geography. Umversny of Tasmania Abstra...

809KB Sizes 17 Downloads 5 Views

THE USE OF DISCRIMINANT DETECTION OF GEOGRAPHIC

ANALYSIS IN THE TYPES OF ASTHMA

G. G. GILES Department

of Geography.

Umversny

of Tasmania

Abstract-The abtltty of resptratory symptom htstorres to predict asthmattcs’ restdenttal environment IS exammed by usmg data from the 1961 birth cohort of Tasmania Multtple dtscrtmmant analyses demonstrate that asthmattcs hving m four differing physical environments wtthm the state can be successfully discrtmtnated using these vartables. Cartographtc analyses reinforce theories of geographic environmental mteractton by illustratmg the statistrcally significant surpluses and deficits of each geographic type. The spattal concentration of each geographic type matches tts source area and exhibits a distance decay effect An exammatlon of physical and climatic data for each of the four areas indicates the Importance of temperature. altitude and pollution in the pathogenesis of childhood asthma morbidity.

The measurement of disease-environment Interaction is an issue of longstandmg concern both to medical geographersand to epidemiologists. These relationships have been most easily demonstrated in the patterns of common infections [l]. Variations of these epidemics over space. through time and within populations have obvrous envtronmental determmants. The ecologtcal limiting factors of either the pathogen or its vector(s), the establishment of cultural practices hazardous to health and the susceptibility/resistance of individuals are all examples of environmental parameters that influence the appearance of the disease [Z]. Chronic drseases are not so simple; a multifactorial aetiology is commonly implicated m the preclmical phase. during the chnical manifestatrons of an ldentlfiable disease state and m the subsequent natural history and degree of morbidity experienced. Many factors of aettological importance are, of course, envnonmental. Physical, psycho-social and brological elements of the environment all play varying roles singly or in combination at different stages m the course of the illness. There can be, for example, 20 year lags between the Initial envtronmental insult and the appearance of clinical lesions diagnosed as cancer. Many naturally occurring geogens and also mtroduced anthropogemc substances can have a cumulative effect. Some are harmless unless a synergistic substance IS present. Others have additive effects and yet others give multiplicative results. Furthermore, such environmental agents continue to interact with an established chronic complaint by the provtslon of stressful or restorative stimuli that can do much either to exacerbate or to ameliorate the symptoms and ultimately the progress or remission of the disease. Because of aetrologtcal complexity and the infinite permutations of environmental interactions that can take place within the comparatively long pre-climcal phase of many chronic conditions, the ensuing pattern of disease in any populatton IS likely to be extremely variable. Heart disease. cancer. bronchrtls. asthma. arthritis and hypertension are all labels for syndromes that are each described by a unique symptom constellation but at the mdivtdual case level are remarkably variable m the degree of each sya atom experienced. Case variability affects diagnostic definition and this

adds more “noise” to epidemiological patterns. Some cases are symptomattcally so mild that the sufferers never present themselves for diagnosis in the first place-the onset of incapacity being so slow that it is accepted as normal. The data available for most epidemiological studres of chronic disease reflect all of these effects and other sources of error that can seriously misrepresent the real. community prevalence. Errors m data structure, collection and processing are notoriously difficult to control. Given the limitations of the best data that can be obtained, the complex web of inter-related effects stretching across space and time may appear to be intractable. Fortunately, this is not the case. Although the variation between cases may be high, by using careful controls. groups of individuals can be found who share basic environmental, and humanecological parameters to the extent that their withingroup variation is extremely low. If environment is taken as the sum total of all effects acting upon an individual since conception, individuals sharing similar environments should demonstrate similar disease patterns. Indeed, because of the stratification of society by age, sex, occupation, socio-economic status, residence and other variables, tt is reasonable to assume that persons of one closely defined group will differ in disease characteristics from other differently defined groups. This rationale underlies the use of cluster analysis techniques by epidemiologists. By defining tight, minimum variance, clusters one is able to classify any morbid population into sub-groups possessing common aetiologies or levels of morbidity. Contrasting clusters can then be compared to elucidate causal or irritant factors in the environment. The opposite of this approach is to select environments that differ with respect to the variables suspected to be of interest and then compare the difference in natural history experienced by individuals residing m those environments. This is a medical-geographtc perspective. The geegrapher assumes that dtscrete regions will enjoy a high degree of homogeneity in elements of the physical environment and hence in patterns of environment-sensitive diseases. In comparing dissimilar environments he hopes to identify the important

G. G. GlLES

126

factors that influence the type and extent of complaint suffered. Marked spatial variation has been demonstrated in the mcldence of several diseases [3]. Although many of these spatial patterns remam poorly understood, an examination of areas containing statistically significant concentrations and deficits of cases has occasionally led to the discovery of causal/irritant factors in the environment [4]. Chronic diseases. because of their long time span. offer more opportunity for environmental interplay. In a low-migration, stable population the period of residence m one location is likely to be long. Such a situation is conducive to the development of “geographic types” of chronic disease history. If individuals suffering from the same chronic complaint living in the same area experience closely similar environmental histories they are likely to be distinguishable from individuals living in other areas possessing different environmeutal hlstories. When within a dlseased population, geographic types can be defined to the extent that given a history of symptoms one can predict the area of residence of any individual with a greater probability than that arrived at purely by chance, then environmental influence will not only have been shown to exist but also will have been shown to be measurable.

AIMS AND

METHODS

This research was designed to detect geographical differentiation in the natural history of asthma, a common chronic disorder. Asthma is used here as a collective term for the entire spectrum of wheezy breathing. This syndrome was chosen for study for three reasons. First, unusually high-quality data were available upon the prevalence of asthma in a birth cohort of Tasmania, an island state of Australia. Second, its community prevalence in the study area was quite high; about 169/, of the population of 7 year olds living in the state. And third, the asthma syndrome was well documented in the literature for its environmental, especially meteorotropic, relationships

CSI. It was

hypothesised that sufferers from an environmentally sensitive condition like asthma who spent their maximum period of residence in dissimilar geographic areas would demonstrate differences in morbidity. Furthermore, a comparison of area1 differences m morbidity would help identify the recognizable envlronmental elements. The research therefore centred around three objectives: (a) the selection of discrete, homogeneous geographic areas (b) the comparison of group characteristics of asthmatic individuals in those areas, and (c) the identification of physical environmental variables that could be implicated in asthma pathogenesis. Multiple discrimmant analysis [6] was chosen to identify geographic types within the wheezy breathing/asthma syndrome. This technique is designed’ to separate statistically two or more groups using a set of variables upon which the groups should theoretically differ.

au

Fig.

. ONE

PRIMARY

SCHOOI

1 The dlstrlbutlon

of groups and prImarc schools

their

constituent

1. Northwest, 2. Burme. 3. Devonport, 4, N.W. Inland. 5. Tamar Estuary, 6. Launceston. 7 Northeast. 8. East Coast. 9 Interior. 10. Hobart. II. Huon. 12. West Coast.

DATA

The data used in this study were collected by the Asthma Foundation of Tasmania in 1968 [7]. In that year every child in the 1961 birth cohort was examined for evidence, past and present. of chest illness. Of over 8000 individuals, about 16” “. had had a history of wheezing. It was this population of wheezers that was chosen for analysis. Not only was age controlled for. but all the information was collected by the same team of thoroughly briefed medical officers. thereby minimising inter-observer errors. Each of the 1387 cases was described’ by over 40 variables. These included historlcal details, clinical signs and anthropometric and spirometric measurements. When binary variables were removed, 27 interval-level variables remained, which were standardised prior to analysis (Table 1). The first 13 were responses to a health history questionnaire. The remainder were collected in the course of a clinical examination or were combined scores derived from both historical and clinical evidence. The geographic location of each case was described by the primary school of daily attendance. The sites of these schools are indicated on Fig. 1. GROUP

DEFINITION

Before the dlscriminant analyses could be executed the groups to be discriminated had to be defined. It was decided to start with a large number of small, relatively homogeneous physical environments and in the course of successive analyses to discard any areas with unacceptable proportions of misclassifications among their residents. In this way only a few strong types were expected to remain at the conclusion. Twelve areas were selected for the initial analysis (Fig. 1). This exercise drew a great deal upon detailed local knowledge from many sources. Places were included on the basis of popular reputation, general or asthamatics’ anecdotes. practitioners’ opinions

The use of discriminant Table

1. Vartables

analysis

available

227

as dtscrimmators

Mean

Standard deviation

1.36 1.31 3.50 5.72 2.08 2.58 2.71 1.81 5.01 2.21 2.32 2.69 118.50 1.73 1.95 1.27 1151.21 1449.04 1621.58 1917.19 0.47

0.60 0.6 1 1.76 1.95 0.90 1.66 1.33 1.81 2.35 1.13 1.74 1.55 13.79 0.44 0.22 0.67 247.91 336.49 343.06 537.61 0.63

1.02 1.65

0.19 0.54

1.03

0.89

0.02 0.48

0.18 0.88

Descriptton No of episodes of pneumoma No. of episodes of haves p.a. Ttme elapsed smce last eptsode of wheezing Frequency of occurrence of wheezmg Average length of episode of wheezmg Age of onset of wheezmg Total number of eptsodes of wheezing Time elapsed smce last eptsode of cough Frequency of occurrence of cough Average length of episode of cough Age of onset of cough Total number of episodes of cough Hetght in cm Degree of nasal obstructton Degree of post-nasal dtscharge Presence of rhonct and riles Forced exptratory volume f set Forced expnatory volume 1 set Vttal capacity Maxtmum exptratory flow rate Composite score htstory and presence of allergtes Composne score htstory and presence of wheezmg Compostte score htstory and presence of cough Compostte score history and presence of upper respiratory mfecttons Compostte score htstory and presence of deformities Compostte score htstory and presence of eczema

and urban structure, it was suspected that m some of the areas the within-group variance in environmental elements would be greater than the between-group variance and that some groups, therefore. would not serve as strong geographic types. Some areas, for example, will attract more in-migration than others and their population morbidity averages will be based on variable environmental histories and hence regress toward the state norm. Alternatively, many locations will have grossly similar physical environments and will again tend toward a normal.

were chosen for their unique climatic and physiographic characteristics and some for their urban-industrial pollution. Much of the information available was not supported by hard data. The best data obtainable were physicalclimatic averages. Some of these are given in Table 2 to illustrate the climatic variation withm the state. Each area is described by meteorological data collected from a representative station within its boundaries. Although each of the areas shown on the map possesses a unique mrx of climate, topography. industry Others

Table

2. Physical

Northwest Burme Devonport Inland N W Tamar Launceston N. East E Coast lntertor Hobart Huon W Coast * 86 percenttle

characteristics

Altttude (metres)

Ramfall (cm)

4.6 7.3 12.2 269 7 152 106 7 199.9 30.5 349.9 55.2 39.9 17’2

92.9 100.5 92.4 1209 86.5 71.4 98.4 68.4 55.1 62.2 87.8 250.7

of January

of a representative

Ramdays

maximum-14

site m each area

Max. temp. (‘CJ

Mm. Temp. (‘CJ

Range* f-c)

16.6 16.8 16.7 15.6 17.0 17.4 16.5 17.1 16.5 16.8 16.9 15.7

9.2 8.3 8.0 57 8.2 7.2 7.1 9.4 3.7 8.3 55 6.4

21.0 224 23.3 24.6 24.6 27.2 26.1 19.6 32.5 24 6 27.2 25.6

201 162 138 143 148 131 143 99 140 162 149 248 percentile

of July muumum

_“X

G.

Table

3. Group

means

G.

GILES

for each variable Northwest

Pneumonia Hives Age of onset of wheezing Duration of cough Height Forced expiratory volume ) set Vital capacity Maximum expiratory flow rate Index of wheeziness Index of productive cough Index of upper respiratory infection Index of deformities

1.37 1.47 2.50 1.55 120.62 1131.76 1611.26 1996.66 1.00 1.45 0.55 0.00

Table 4. Standardised

discriminant Function

Forced expiratory volume $ set Maximum expiratory flow rate Vital capacity Average duration of cough Height Upper respiratory infections

lnterlor

1.43 1.29 2.41 2.55 120.39 1058.40 1520.23 1732.95 0.99 1.76 1.27 0.00

I .26 I .19 2.40 2.49 122.73 1241.00 1728.71 2162.77 1.00 1.89

1.21 1.24 2.95 2.09 118.41 I180.07 1621.57 2082.48 0.98 171 1.17 0.0’

1.06 0.00

Spinal-thoracic deformity Index of wheezing Incidence of pneumonia

function

1

0.90 -0.90 -0.72 0.52 0.47 0.26

coefficients*

Function

2

Function

-0:2

0.59 +

-

+ 0.53 -0.58 -0.47

o.lo 0.27 -

-0.26 + +

-0.41 0.29 0.22

+ + -

+

0.34 -

0.22 0.50

0.32 46.5 0.49 0.54 109.791 39 < 0.001

0.23 33.3 0.43 0.72 60.104 24 < 0.001

0.14 20.2 0.35 0.88 23.23 1 11 0.016

Age of onset of wheezing

* Coefficients

Hobart

for height and spirometry and Hobart possesses the lowest. The Northwest has the lowest value for productive cough and for its duration. Huon is characterised by low values for pneumonia. hives and wheezing and the highest values for infantile feeding difficulties, age of onset of productive cough. and deformities of the spine and thorax. From these variables, discriminant functions were computed. The first three functions gave significant increases in discriminatory power and were retained for classificatory purposes: the remaining, nonsignificant functions were discarded. The structure of the functions is illustrated in Table 4. Function 1. which explains over 46”” of the between-group discriminatory variance, is positively related to a history of pneumonia. the average duration of productive cough and the forced expiratory volume in half a ,second but is negatively related to vital capacity and maximum expiratory flow rate. This function is obviously describing a parameter of pulmonary health linked to pathological changes caused by pneumonia and extended bouts of bronchial infection.

The 12 groups entered in the initial analysis were a first approximation. The intention was to conduct more than one analysis and to discard any area that was unsatisfactorily discriminated. After 4 separate analyses, at the end of each of which the 2 worstclassified groups were deleted, the final analysis was conducted upon 4 groups. The groups entered in this analysis were: the Northwest, the Interior, the Huon and Hobart. These four residential environments contained the extremes of the ranges of altitude, latitude, urbanisation and continentality within the state. The discriminant analysis chosen was a step-wise model. That is, the best discriminating variable was chosen first, and then successive choices of variables were made that increased the discriminating power of the model. This process continued until the gain was no longer statistically significant. In all, 13 variables were selected in this stepwise fashion. These are listed in Table 3 which also gives the group means for each variable. The means are interesting in their own right: for example, the Interior possesses the highest mean

Eigenvalue Relative percentage Canonical correlation Wilks’ lambda Chl-square Degrees of freedom Significance

m the analysis

values

ANALYSIS

Incidence of hives Index of productive

entered

cough

less than 0.2 are indicated

by + or -.

3

729

The use of discrrminant analysrs Table 5. Predrction results

Actual group

No. of cases

Northwest

45

Hobart

91

, Interror

39

Huon

47

Ungrouped cases

1165

TOTAL _

1387

Northwest 21 46.7% 10 11.0% 4 10.3% 6 12.8% 255 21.9% 296 21.3%

Hobart

Interior

9 20.07” 45 49.50/, 6 15.47; 9 19.1% 321 27.6:; 390 28.1%

8 17.8:; 19 20.9:/, 20 51.3% 9 19.1% 281 24.6:; 343 24.7%

Huon 7 15.6”,a 17 18.77; 9 23.17; 23 48.9”/:, 302 25.97; 358 25.87;

Percent of “Grouped” cases correctly classified: 49.1%.

Function 2, explaining a further 3376 of the variance, is characterised by positive relationships with hives and wheezing and negative relationships with age of onset, upper respiratory infections and thoracic

deformities. This parameter is related to early-onset, allergic, wheezing of a mild nature. Function 3 explains the remaining 20% of the variance and is related to the history and presence of productive cough and the forced expiratory volume in half a second. This parameter is more difficult to describe as it loads on fewer variables but is probably related to bronchial asthma of a more severe, incapacitating nature than that described by function 2. CLASSIFICATION The next step was the production of a set of classification coefficients that would transform the raw vari-

ables for each case into probabilities of membership for each group. Each case is thus allocated to the group for which it has the highest probability. For the originally “grouped” cases, the proportion of correct allocations is calculated to give an estimate of the predictive quality of the model. The ungrouped cases’ allocations are then examined to see how accurately a case of unknown group affiliation can be located with respect to its closest group. The results are given in Table 5. Just under 50% of the grouped cases were correctly assigned. This proportion is about twice the proportion expected purely by chance (25%). Although the three discriminant functions were found to predict group membership correctly for 50% of the cases with known affiliation, the initiallyungrouped cases had also been classified and their residence was known. It remained to be discovered whether these extra cases would enhance the prediction results and also whether the grouped and ungrouped cases, taken together, could demonstrate geographic patterns of statistical significance. SPATIAL ANALYSIS

These questions were answered by combining the grouped and ungrouped cases and mapping them with respect to their primary school locations. High and low prevalence schools for each type were

selected by eye and combined with contiguous neighbouring schools of similar or more extreme prevalence until the proportion of that type began to drop or increase sharply. At this point the cluster of schools was delineated with a line spaced at mid distance between the peripheral schools belonging to the cluster and schools external to it. The proportion of cases of each type within the cluster’s population was then tested for significant deviation from the state’s norm. The norms were taken from Table 5 (e.g. the norm for the Northwest type was taken as 21.3%). The norms of all 4 types were each over 20%. To test for the significance of each cluster’s deviation from the norm, use was made of the binomial distribution of proportions. This exercise was facilitated by the use of tables [8]. N is the total number of asthmatics in the cluster; ni is the number of asthmatics of type i in the cluster and pi is the percentage of type i in the total population @i is the norm for type i). There is a separate table for each N. For each ni in a given table for N, tolerance limits are given for the acceptable range of proportions for nJN at both the P < 0.05 and P < 0.01 levels. If Pi is contained within the tolerance range for any N and ni, then ni does not deviate significantly from pi. If pi falls below the tolerance range, hi is significantly greater than the norm at that significance level. If pi is above the tolerance level, t+ en ni is significantly less than the norm at that level of significance. The process was executed iteratively for each type until all high and low areas were examined for significant deviations.

RESULTS AND DISCUSSION

Figure 2 presents the results of the spatial clustering. The probability maps obtained are similar to those of the Choynowski/McGlashan [9] method but are based upon the binomial rather than the Poisson distribution. All of the shaded areas shown are significant at P < 0.01 unless otherwise stated. In addition, an attempt has been made to indicate “high spots” within the areas of significance. With large numbers it is often possible to obtain a significant deviation with a remarkably small increase in proportion. Each area type has an intense local effect surrounded by an area

G.

@

NORTHWEST

G.

GILES

(21.3%)

@

HOBART

:

>state norm

._

Excluded
Significant

0

INTERIOR

@.&;:, . ‘1_ ;; ’

I. +cs _.-

norm

-----Source

‘-T

..

(28.1%)

__*

b” \

, . ,,

\ .. 3X

‘!i

14’.

area

or.20

(24.7%)

@

HUON

(258%)

do

N = Total “i=

number

Number p 60.01

of

of osthmatlcs type

unless

i

osthmotxs

l (p < 0.05)

Fig. 2. Probability maps of the discrimmant groups.

of overall increased prevalence that is statistically significant. The blank areas on the map are excluded areas that are close to the norm in prevalence. When Figs 1 and 2 are compared it 1s possible to assess the spatial persistence of the types. The constituent members of the type groups were from quite small geographic areas; however, all types demonstrate in the maps much larger areas of above average prevalence. The source area for each type usually has the highest concentration of that type. This core locality is surrounded by an area of lowered prevalence that remains significantly higher than the norm. A distance decay gradient is observed about these core areas but is disturbed by the population’s affiliation to other types. The Northwest type has a state prevalence of about 219; (Fig. 2a). In its source area its prevalence IS twice the norm at 431,;. When the peripheral areas are added the prevalence drops to 36”~; which is still stgnificantly greater than the norm at P < 0.01. Northwest type asthma is absent from the central area of the state P < 0.01 as thus area is dominated by Interior and Huon types. Hobart type asthma (Fig. 2b) was represented by the school populations in the core of the industrial manufacturing areas of the city. With a state prevalence of ZS’:, the only hrgh areas are the western suburbs of Hobart 44’1; P s 0.01 and an area around Georgetown at the mouth of the Tamar estuary

P < 0.05. Other urban and suburban areas were close to the state norm. Similarly to Northwest type, Hobart asthma was absent from the central area and most of the rural districts P < 0.01. The far Northwest area was also low P < 0.05. Interior type (Fig. 2c) accounted for nearly 25”; of the cases. Its high prevalence area closely parallels Hobart type’s low prevalence area. The source area contains 619; of its own type: its larger region 37”” P < 0.01. The only significantly low area discovered was the industrial area of Hobart 169, P < 0.05. Huon type (Fig. 2d) had a prevalence of 51”, m the Huon area, twice the norm. Its wider domain mcluded the Interior source area and gained a prevalence of 389,. The suburbs of Hobart, possessing only close to average proportions, formed an enclave within this area. The north east sector of the state, an area of high Interior type prevalence, was significantly low for Huon type P < 0.01. By examining Table 2 and the additional detailed informatton upon climatic averages m Table 6 one can begin to Identify the links between climatic envrronments and the drscriminant groups. When extremes in the spatial patterns in Fig. 2 correspond to extremes in the chmatrc data evidence for environmental interaction is strongly supported. Actually. the maps suggest two pairs of contrasting dimensrons. Hobart contrasts with Interror and Northwest wrth Huon

The use of discrimmant

analysis

Table 6. Chmattc averages for representative

Awwal

stations

Northwest

Huon

Hobart

Interior

13.0 15.4 7 1

11.6 15.3 n.a n.a.

12.1 15.3 13

10.3 14.9 138 87

20.9 12.5 8.4

21.3 8.8 12.5

21.8 11.9 9.9

22.8 7.6 15.2

12.5 6.1 64

11.5 0.0 11.5

11.6 4.5 7.1

10.8 -0.3 11.2

averuges

9 a.m. temperature 3 p.m. temperature Frosts. light t < 2 C) Frosts, heavy (i O-C) Monthly

231

averages

January Maximum temperature Mimmum temperature Temperature range July Maximum temperature Mmimum temperature Temperature range

The first pair, Hobart and Interior, has obvious differences m environment; they measure the opposite ends of the rural-urban continuum. The Hobart group comes from the polluted. industrial parts of the city; the Interior group from the clean, open bushlands, 300m higher and 100 km inland. When the group means are compared, Hobart’s spirometry values are the lowest and Interior’s are the highest. The other notable differences in symptoms are the increased weighting of upper respiratory infection and the increased incidence of pneumonia in the Hobart type. The extremes in spirometry may well be related to a beneficial effect of altitude in the Interior and a harmful effect of low level pollution in Hobart. Sulphur dioxide, in particular, has a marked relationship to increased airways resistance. Interior residents are subjected also to a greater range in diurnal and seasonal temperatures. Although only 50 km from the nearest coast, topography and altitude produce a degree of continentality in the area. This can be seen from the number of frost days, the January average temperatures, the annual range and the low prectpitation. The differentiation of these two areas is probably related to the first discriminant function which loaded highly on the spirometry variables. The Northwest and Huon types are both from rurally-based samples; one in the far north of the state. the other in the far south. In comparison to the Northwest. the Huon type loads highly on age of onset. the average length of attack of productive coughmg. the history, of upper respiratory infection and thoracic deformities. It loads negatively on hives and height The Northwest has slightly lower sptrometry values. a low value for upper respiratory mfections and the lowest value for average length of duration for producttve cough. Age of onset is about 6 months earlier. on average. m the Northwest. The separation of these two areas is probably related to the second and thud discrimtnant functions. Function 2 reflects the intermittent. short duration. early onset. allergic wheezmg of the Northwest whtle Function 3 reflects the extended bouts of cripplmg bronchial asthma of later childhood This would appear to be in\ olved wtth differences rn temperature. Although the Huon area IS by no means continental. it has the most southerly latitude of the groups. Its climatic

variables from Table 6 are closest to those of the Interior. The average July minimum of 0°C is 5°C lower than the Northwest and emphasises the increased probability of days of frost. The Northwest, although not the warmest of the areas, does experience the smallest range of temperature and the highest July minimum. The moderating effect of the ocean is important in this location. In the Huon area, latitude and cold air drainage in winter counteract the maritime influence. More evidence is obtained from the Huon map when it is compared to that of the Interior. One notices that Huon type’s larger area of significantly high prevalence embraces the central portion of the state-the Interior source area. If the suggestion of low temperature being culpable for Huon type asthma is correct, it would be expected to show this overlap with the cooler Interior environment. Further, the Northwest type “warm environment asthma” is conspicuous by its absence from the Interior source area. CONCLUSION

The discriminant and cartographic analyses have thus highlighted three environmental and three asthmatic health dimensions that appear to interact with each other to produce the broad patterns of asthma morbidity in Tasmania. The major climatic parameter would be an aspect of temperature, probably the prolonged cold periods experienced in particular environments acting in conjunction with the strong fluctuation in temperature common to continental areas. High altitude may also be indirectly involved in asthma pathogenesis m that reduced pressure. reduced partial pressure of oxygen and reduced pollution may have beneficial pulmonary effects The third envtronmental element is the price paid for hvmg in urban environments: the reduced pulmonary efficiency caused by domestic. vehicular and mdustrial pollution. The inferred relationships between the dtscrimmant functions and the environment dimensions are summartsed in Table 7. To a large extent this analysts has been exploratory The relattonships outlined above are fairly conjectural. Enough evidence has been produced. however.

G G

232 Table

7. Inferred relationshlps between discrlmmant tlons and envxonmental variables Function 1 Pulmonary function

Temperature Altitude Pollution

+ -

Function Mild

2

func-

Function 3 BronchIal

wheezmg

mfectlon

+ -

+ +

GILES

extremes of one variable at a time could also be helpful m clarifymg exact relationships. Continumg studies involve the use of multiple regression techniques and time series spectral analyses of longltudlnal morbidity records from the differmg geographic areas. .4cknowledgements_The author wishes to thank the Asthma Foundation of Tasmama for then financlal support and their permIssIon Birth Cohort Survey.

to structure hypotheses that can be tested in future work. Whatever further research determines to be the exact relationships between all of the factors involved, it has been demonstrated that with a chronic complaint, asthma, geographic types can be successfully distinguished from one another. In this four group example the highest level of predictive accuracy was 617, (in the source area of the Interior type of asthma). Imperfect predictive ability is, in part, due to the heterogeneity of the syndrome and, in part, due to migration noise that was not controlled for in the data. With more stringent con-

trols and increased selectivity in group membership prediction can be made far more successful. For example, all of the individuals located by primary school were assumed to have spent their maximum period of residence in that location. This assumption is unrealistic for the Hobart area but is quite reasonable in isolated, rural localities where out-migration dominates. Another method of increasing the resolution of the analysis would be to take different sub-populations of wheezers separately. In the present analysis the entire population was included. Individuals ranged from those who had wheezed once to those with continuous symptoms. It can be hypothesised that any environmental effects would have had differential impacts on these and other sub-types. Two-group discriminant analyses looking at the

to use the data

from

the

1961

REFERENCES A. T A. Atlases in medical geography 1. Learmonth 195&1970: A review In Medical Geography. Technrques and Field Studies (Edited by McGlashan N. D.) kethuen, London, 1972. of vectored &eases. In 2. Knight C. G. The ecoeraDhv The- Geography of- H&lb -und Disease (Edited by Hunter J. M.). Umverslty of North Carolina at Chapel Hill. Department of Geography, 1974. 3. Howe G. M. National Atlas of Disease Mortaliry rn rhe United Kingdom. Nelson, London, 1970. N. D. Food contammants and oeso4. McGlashan phageal cancer. In McGlashan N. D (Ed.), op. clr.. 1972. S. W Medical Biorneteorology. pp. 4645. Tromp 475. Elsevier. Amsterdam. 1963. W. R. Discriminant analysis. In Stat~std 6. Klecka Package for the Social Sciences, 2nd edn (Edited by Nle N. H. et al.) pp. 434-467. McGraw Hill, New York. 1975. 1. Gibson H. B.. Silverstone H.. Gandevia B. and Hall G. J. L. Respiratory disorders in seven-year-old ctuldren in Tasmania: Alms. methods and administratlon of the survey. Med. J. Aust. 2, 201, 1969. 8. Diem K. and Lentner C. (Eds) Documenta Geigy. Screnrijic Tables (7th edn), pp. 85-103. Geigy. Basle. 1970. N. D. Uses of the Poisson probability 9 McGlashan model with human populations. Pacific Viewpoint 17. 167, 1976.