551_556_Kennedy_SPPE_919 08.07.2005 09:47 Uhr Seite 551
Soc Psychiatry Psychiatr Epidemiol (2005) 40 : 551–556
DOI 10.1007/s00127-005-0919-9
ORIGINAL PAPER
Conor O’Neill · Alan Kelly · Hamish Sinclair · Harry Kennedy
Deprivation: Different implications for forensic psychiatric need in urban and rural areas
Accepted: 9 February 2005
■ Abstract Background Ecological relationships between deprivation and forensic psychiatric admission rates may differ in urban and rural areas. Aims The aim of the study was to compare the relationship between material deprivation and forensic admission rates in rural and urban areas for a whole-national service in Ireland over a 3-year period. Method All Irish forensic admissions from 1997 to 1999 were allocated to the appropriate small area. Material deprivation scores were calculated from census data. Mean annual admission rates and Bayesian standardised forensic admission ratios for small areas were aggregated by material deprivation score and population density. Results At small area level, there were significant non-linear increases in forensic admissions with increasing deprivation. The increases in urban areas (population density > 10/ hectare) were absent in less densely populated areas. Conclusions Deprivation alone may not be the key factor in predicting forensic service utilisation. Factors associated with specifically urban deprived areas may be of greater relevance in planning services. ■ Key words forensic – psychiatry – admission rates – deprivation – population density
and with use of secure psychiatric beds (McCrone et al. 2001) where local service characteristics and population density may also be relevant. The use of forensic secure bed admission rates is said to be as good or better than general adult psychiatric admission rates as an index of population need because of the independent ‘gatekeeper’ role of the courts. The need for secure psychiatric beds is of particular importance when planning tertiary services (Coid et al. 2001a) and when assessing the effectiveness of existing services (Coid 2001b). Such studies do not always allow generalisation across countries. Composite indices of deprivation rely on census items such as ethnicity or unemployment which have variable definitions. Differences between rural and urban populations may be underestimated by composite socio-economic deprivation scores due to assumptions regarding construct validity. We therefore used population density which allows robust regional and national comparisons. We used an index of deprivation composed of material measures only. We hypothesised no association between material deprivation, population density and forensic admission rates at district or small area level in Ireland. We also hypothesised that for equivalent levels of material deprivation, forensic admission rates are the same in urban and rural locations.
Introduction Subjects and methods Indices of socio-economic deprivation correlate with psychiatric hospital admission rates (Jarman et al. 1992)
C. O’Neill, MRCPsych · Dr. H. Kennedy, FRCPsych () Central Mental Hospital Dundrum, Dublin 14, Ireland Tel.: +353-1/2989266 Fax: +353-1/2963411 E-Mail:
[email protected]
A computerised case register recorded demographic and diagnostic characteristics of all admissions via the criminal justice system to the Irish forensic psychiatric service during the three calendar years 1997–1999. Because we were interested in service need, all new admissions and re-admissions were counted as discrete episodes unless the interval between discharge and readmission was less than one week. An incidence rate over a 3-year period, counting each individual only once, would give a misleading impression of service utilisation. During this period, all admissions via the courts and prisons were to the Central Mental Hospital, Dundrum. Because this service is centralised nationally, it is possible to minimise variations in local practice due to assessment standards, admission criteria and resource
SPPE 919
A. Kelly, PhD · H. Sinclair, PhD Dept. of Community Health and General Practice Trinity College Dublin, Ireland
■ Case definition
551_556_Kennedy_SPPE_919 08.07.2005 09:47 Uhr Seite 552
552 availability, factors which might influence similar studies of general adult psychiatric services across the ten Health Boards and various private sector providers servicing the same population. ■ Sociogeographic measures All admissions were assigned to the appropriate Health Board and District Electoral Division based on address or place of arrest. There are ten Health Boards with approximately equal populations (Table 1). A District Electoral Division (DED) is the smallest census enumeration unit. There are 3,421 DEDs in Ireland, with an average population of 1,060 people (SD 1606.3). The median population density for DEDs is 0.23 people per hectare (range 0.01–155.94 per hectare). The population density for small areas in Ireland has a nonlinear distribution. A division above and below ten persons per hectare (85th centile in Ireland, population 1,046,072 people aged 15–64) approximately corresponds to the median population density in other jurisdictions, e. g. England and Wales combined. This urbanrural split divides the admissions into 381 urban and 95 rural. Material deprivation was measured using a validated five-point scale (SAHRU 1996). This utilises DED census data for unemployment, household overcrowding, car ownership, low social class and living in rented accommodation. This is similar to the material deprivation scales developed by Townsend et al. (1988) and Carstairs (1995). Population density was obtained from the 1996 Irish Census (C. S. O. 1997). ■ Statistical analysis All correlations were performed using Spearman’s rank correlation to avoid assumptions regarding distribution. Population density was not significantly correlated with deprivation at Health Board level (Spearman r = –0.127, n = 10, p = 0.728).At small area (DED) level, the correlation between deprivation and population density is even smaller (Spearman r = 0.09, n = 3421, p < 0.001). Forensic admission rates were calculated from census data. Rates are given per 100,000 aged 15–64. Because forensic admissions are relatively rare events, not all small areas were represented during the study period. A Bayesian spatial modelling technique was employed to smooth the small area estimates (Mollie 1999). An age-standardised forensic admission ratio (similar to a standardised mortality ratio) was calculated for all small areas (DEDs), using an algorithm which starts with median admission ratios for represented DEDs in an adjacency matrix. For rare events such as forensic admissions at DED level, this allows estimates [Bayesian Standardised Forensic Ad-
mission Ratios (B-SFAR)] which can be used for service planning (Kelly 1999). The B-SFAR is expressed as the mean B-SFAR of constituent DEDs for Health Boards and other population aggregates. Because this study is primarily concerned with service utilisation, admissions were counted rather than individuals, since counting individuals would underestimate service use. At Health Board level, the admission rate correlated strongly with the B-SFAR (Spearman r = 0.976, df = 9, p < 0.001) (see Table 1). At small area (DED) level, where admissions are rare, the correlation is less strong (Spearman r = 0.433, n = 3421, p < 0.001). This suggests that the B-SFAR does not distort the raw data. ■ Sample There were 476 admissions of 348 individuals via the criminal justice system between 1997 and 1999. In all, 418 (88 %) admissions were of males. Mean age was 30.9 years (SD 10.5). A total of 140 (29 %) cases had a primary ICD-10 diagnosis of schizophrenia, 111 (23 %) personality disorder and 89 (19 %) affective disorder. Only 58 (12 %) were married, and only 64 (13.4 %) were in full-time employment. Mean length of stay during the study period was 77.8 days (SD 129.9). Only 176 (37 %) had been detained for violent offences. ■ Case mix: urban and rural comparisons A simple division into those detained because of violent or non-violent offences reveals no significant difference between urban and rural populations (χ2 = 2.36, df = 1, p = 0.125). Using the scale devised by Coid and Kahtan (2000) modified for the Irish jurisdiction, where 1 represents the lowest need for security and 4 represents the highest, there were no cases rated to level 1 (hospital transfers for selfharm only); 274 at level 2 (property or minor violent offences); 121 at level 3 (violence against the person or sexual offences which were not life-threatening); and 81 at level 4 (life-threatening violence or rape; a full set of definitions as modified for the Irish jurisdiction is available from the authors). The differences between rural and urban areas do not reach significance when those rated 2 (property or minor violent offences) are compared with 3 and 4 combined (more serious violent and sexual offences against the person) (χ2 = 3.18, df = 1, p = 0.075). There were 198 admissions (41.3 % of total) with a diagnosis on discharge of schizophrenia or other psychosis (ICD-10 codes 0.00–9.00 organic psychosis, 20.00–31.2, 31.9–32.0 schizophrenia spectrum disorders, delusional disorder, mania and psychotic depression). There was an apparent excess of patients from urban DEDs
Table 1 Mean annual forensic psychiatric admission rates via the criminal justice system per 100,000 population aged 15–64 years and Bayesian-predicted age-standardised forensic admission ratios, material deprivation scores and population densities by Health Board in Ireland; 1997–1999 Health Board
Population aged 15–64
Material Deprivation Score
Population density (Persons per hectare)
Number of admissions 1997–1999
Forensic admission rate/100.000 aged 15–64
95% CI limits
Mean B-SFAR value
95% CI limits
ERHA-North ERHA-SW ERHA-East Midland North-Eastern Mid-Western Western Southern South-Eastern North-Western Total
309,045 352,135 215,437 128,625 193,667 203,709 218,799 352,507 249,334 129,793 2,352,781
2.7 2.6 1.9 2.2 2.2 2.1 2.3 2.2 2.3 3.0 2.3
8.59 2.00 2.11 0.31 0.48 0.38 0.25 0.44 0.41 0.25 0.33
135 123 44 25 34 27 22 34 22 10 476
14.56 11.64 6.81 6.48 5.85 4.42 3.35 3.22 2.94 2.57 6.74
(12.21–17.23) (9.68–13.89) (4.95–9.13) (4.19–9.56) (4.05–8.18) (2.91–6.43) (2.10–5.07) (2.23–4.49) (1.84–4.45) (1.23–4.72) (6.15–7.38)
182.9 119.9 47.4 30.8 26.5 29.8 19.8 16.7 19.1 16.5 35.7
(127.3–238.5) (90.6–149.2) (36.1–58.8) (24.0–37.6) (23.2–29.8) (23.4–36.3) (17.6–22.0) (13.0–20.4) (17.6–20.7) (12.8–20.2) (32.3–39.2)
551_556_Kennedy_SPPE_919 08.07.2005 09:47 Uhr Seite 553
553 schizophrenia or other psychosis (χ2 = 6.37, df = 1,
with a diagnosis of p = 0.012). This suggests that if there is any variation in the threshold for admission, it operates so that those from urban areas with less severe mental disorders are less likely to be admitted.
Results ■ Hypothesis 1: Association between deprivation, population density and forensic admission rates at district (Health Board) or small area level There were highly significant differences in mean annual age-standardised forensic admission rates between Irish Health Boards (Table 1). When admission rates for DEDs are averaged across Health Boards, analysis of variance is highly significant (ANOVA F = 15.4, df = 9, p < 0.001). However, this difference is exaggerated by the large number of rural DEDs with no admissions. The Bayesian-adjusted standardised forensic admission ratio for DEDs (B-SFAR) overcomes this problem, and confirms highly significant differences between Health Boards (ANOVA F = 63.1, df = 9, p < 0.001). There was no correlation between forensic admission rates and deprivation at health board level (Spearman r = –0.117, p = 0.748). There was a significant association between forensic admission rates and population density at Health Board level (Spearman r = 0.729, p = 0.017). Using B-SFAR as the dependent variable, there was also no correlation with deprivation at Health Board level (Spearman r = –0.111, n = 10, p = 0.761), and population density was correlated with mean B-SFAR for the Health Board (Spearman r = 0.681, n = 10, p = 0.03). At small area level, significant non-linear increases were found with increasing material deprivation in both mean age-standardised forensic admission rates and Bayesian-Standardised Forensic Admission Ratios (Table 2, Fig. 1). Similarly, when DEDs were divided into deciles according to population density (Table 3, Fig. 2), there was a significant increase in both forensic admission rates (ANOVA F = 18.15, df = 9, p < 0.001) and BSFAR (ANOVA F = 54.49, df = 9, p < 0.001).When divided into quintiles according to population density, this also held true for admission rates (ANOVA F = 32.17, df = 4, p < 0.001) and B-SFAR (ANOVA F = 95.45, df = 4, p < 0.001).
Table 2 Mean annual forensic psychiatric admission rates* via the criminal justice system per 100,000 population aged 15–64 years and Bayesian-predicted age-standardised forensic admission ratios (B-SFAR) for Irish small areas aggregated according to material deprivation score 1997–1999
Fig. 1 Mean Bayesian-predicted forensic admission ratios (B-SFAR) for Irish small area aggregates at five levels of material deprivation (Error bars indicate 95% confidence intervals)
■ Hypothesis 2: For equivalent levels of material deprivation, forensic admission rates are the same in urban and rural locations Urban small areas (i. e. with population density > 10 per hectare) demonstrated an exponential increase in agestandardised admission rates and B-SFAR values with increasing deprivation (Table 4, Fig. 3), while no such effect existed in rural small areas. Regardless of the population density cut-off taken to define the rural-urban divide, the same pattern of significant difference was always found. Rural, low-population density small area aggregates did not have significantly higher forensic admission rates or B-SFARs when more deprived, while ‘urban’ high population density areas had significant non-linear increases in forensic admission rates with increasing deprivation
Discussion These findings suggest that the relationship between utilisation of forensic psychiatric services and deprivation is more complex than a simple linear correlation.
Deprivation score
Number of DEDs
Cases
Population aged 15–64
Mean annual admissions per 100.000 aged 15–64
95% CI
Mean B-SFAR value
95% CI (lower)
95 % CI (upper)
5 4 3 2 1
135 429 606 1,460 791
125 175 55 80 41
140,034 434,400 329,071 790,015 659,261
29.75 13.43 5.57 3.38 2.07
24.77–35.45 11.51–15.57 4.20–7.25 2.68–4.20 1.49–2.81
207.9 98.9 56.9 48.0 50.9
147.9 81.1 50.9 46.1 47.1
267.9 116.7 62.9 49.9 54.7
551_556_Kennedy_SPPE_919 08.07.2005 09:47 Uhr Seite 554
554 Table 3 Age-standardised forensic admission rates and Bayesian-derived standardised forensic admission ratios for Irish small areas aggregated into deciles of population density
Decile of density
Upper limit of density range
Population aged 15–64
Cases
Forensic admission rate
95% CI lower
95% CI upper
B-SFAR (based on median)
95% CI lower
95% CI upper
10 9 8 7 6 5 4 3 2 1 Total
155.94 26.32 0.87 0.39 0.28 0.23 0.19 0.16 0.13 0.10 155.94
637,152 803,145 261,374 151,487 124,171 99,374 86,102 72,370 71,806 45,800 2,352,781
268 154 19 11 4 4 3 6 5 2 476
14.02 6.39 2.42 2.42 1.07 1.34 1.16 2.76 2.32 1.46 6.74
12.39 5.42 1.46 1.21 0.29 0.37 0.24 1.01 0.75 0.18 6.15
15.80 7.48 3.78 4.33 2.75 3.44 3.39 6.02 5.42 5.26 7.38
136.9 64.3 24.6 21.8 18.0 18.7 18.1 20.0 18.7 16.3 35.7
109.9 48.4 18.8 18.8 16.7 17.0 16.8 17.7 17.4 15.2 32.3
164.0 80.2 30.3 24.7 19.3 20.4 19.4 22.2 20.0 17.5 39.2
Fig. 2 Mean Bayesian-predicted forensic admission ratios (B-SFAR) for Irish small area aggregates at ten deciles of population density (Error bars indicate 95% confidence intervals)
Ecological studies do not demonstrate causal links between environmental variables and epidemiological rates. However, there are plausible theoretical explanations for the associations found here. Taking the epidemiological triad of host, environment and agent, ecological studies tell us little about susceptibility at the host level. A morbid idea or practice can behave like a noxious agent, spreading in predictable epidemic and endemic ways relevant to this study, e. g. heroin abuse (De Alarcon 1969; further discussed in Shepherd 1983). These ‘epidemics’ follow mathematical patterns of spread and cessation (Daley and Gani 1999). Over-crowding may in itself be a stressor, giving rise to mental illnesses, but disorders spread more quickly and maintain a higher endemic prevalence where there are large numbers of susceptible individuals in close proximity. In relation to this study,‘ecological’ susceptibility factors may include unemployment, ready availability of street drugs and poor urban planning (Dalton 1992). Just as disorders spread more quickly in areas of high population density, the endemic level may also de-
Table 4 Mean annual forensic psychiatric admission rates and Bayesian-adjusted mean age-standardised forensic admission rates (B-SFAR) at different levels of deprivation in small area aggregates with population densities above and below ten persons per hectare in Ireland: 1997–1999 Population density
Deprivation score
Above ten per hectare
5 4 3 2 1
Below ten per hectare
5 4 3 2 1
Number of DEDs
Cases
Population aged 15–64
Mean annual admissions per 100,000 aged 15–64
95% CI limits
Mean B-SFAR value
95% CI limits
68 133 56 98 149
122 151 33 42 28
113,018 283,340 100,737 217,037 331,940
35.98 17.76 10.92 6.45 2.81
29.88–42.96 15.04–20.83 7.52–15.34 4.65–8.71 1.87–4.06
307.53 163.02 88.99 63.851 51.47
204.11–410.9498 117.84–208.20 43.88–134.10 47.84–79.86 36.03–66.90
67 296 550 1,362 642
3 24 22 38 13
27,016 151,060 228,334 572,978 327,321
3.70 5.30 3.21 2.21 1.32
0.76–10.82 3.39–7.88 2.01–4.86 1.56–3.03 0.70–2.26
23.19 25.64 21.4963 19.649 20.23
10.59–35.79 16.854–34.42 19.23–23.76 18.76–20.54 18.80–21.65
551_556_Kennedy_SPPE_919 08.07.2005 09:47 Uhr Seite 555
555
Also, however important they may be, service factors may tend only to obscure the underlying pathogenic factors. We have not included local service factors in this analysis because ecological studies are in general better at generating hypotheses and establishing the existence of an association rather than estimating its strength (Richardson 1992). Variations and trends in population, deprivation and service factors, over time and across places may also influence the data. For practical purposes, what is needed is a sample with adequate numbers for statistical power, collected over sufficiently short periods to be relevant to a single service model. This is the principal reason for the use of Bayesian-adjusted admission ratios in this study (Clayton and Bernadinelli 1992).
■ Comparisons Fig. 3 Mean Bayesian-predicted forensic admission ratios (B-SFAR) for Irish small area aggregates at five levels of material deprivation in areas with population densities above and below ten persons per hectare (Error bars indicate 95% confidence intervals)
pend on the size of the population at risk and increases are higher where the prevailing rate is high, a critical mass effect (Lester 1988). Urban planning, population density and design are said to be important determinants of urban crime rates (Newman 1972). However, city environments can undergo surprisingly rapid changes over time (Jacobs 1962; Porter 1994; Prunty 1998), hence the advantage of shortening the sample time required by using Bayesian methods. This study shows what appears to be a qualitative difference between rural and urban small areas in keeping with these classic epidemiological models. Increasing material deprivation is associated with increased forensic psychiatric admission rates, but only in urban, densely populated areas.
■ Methodological limitations International comparisons and benchmarks using measures of deprivation (e. g. unemployment) are difficult since census and Government statistics are often defined differently in each jurisdiction. Population density has the advantage of being clear and universal. This paper does not take account of variations in service provision, e. g. at Health Board level. We have described these variations in the population studied (O’Neill et al. 2002). McCrone et al. (2001) suggest that provision of locked (low-secure) wards at the sub-regional level is an important determinant of regional forensic admission rates, but there is an important distinction between acute and long-term locked low-secure beds. Faced with such distinctions, a true measure of service provision is difficult to standardise or measure, at least in ways that enable international comparisons.
An exponential relationship exists between forensic admission rates and deprivation, though only in urban areas. This exponential relationship in urban areas may reflect a critical mass effect of adjacent densely populated communities, vulnerable due to material deprivation. A concrete explanation for this urban effect may lie in the exponential relationship between deprivation or population density, and rates of suicide and violent crime including homicide, described in London (Kennedy et al. 1999). An exponential relationship between indicted crime and material deprivation has also been described in the population studied here (Bacik et al. 2000). Since suicide and homicide rates are not ‘capped’ by resource availability, their exponential relationship with population density may be a more accurate reflection of the underlying population process than the linear increases in forensic admission rates observed in England and Wales (Coid et al. 2001), where service factors appeared to account for a large part of the variance (McCrone et al. 2001). Social cohesion may have some protective effect in deprived rural areas. Social cohesion is, however, difficult to define and to measure. Before accepting that ‘cohesion’ is the protective factor, it would first be necessary to discount the epidemiologically protective effect of isolation. Illegal intoxicants are less readily available, opportunities for acquisitive crime and provocations for violent crime are all less probable in areas with low population density.
■ Implications for future research These findings are ecological associations and do not imply causation. These results do permit international benchmarking against population density. Relationships between hospital admission rates and population deprivation scores have been incorporated into formulae to predict admission rates and, thus, allocate Health Board funding. It has been pointed out that when ap-
551_556_Kennedy_SPPE_919 08.07.2005 09:47 Uhr Seite 556
556
plied to general psychiatric settings, such formulae have had limited predictive utility and may have led to diversion of funds away from deprived, densely populated inner-city areas (Glover et al. 1999; MILMIS Group 1995). However theoretically attractive it might be to continue to refine definitions and measures of material deprivation and social cohesion, a simple measure such as population density seems to be as powerful. We suggest that a valuable future seam of research can be mined by examining the relative importance of population density and critical mass of adjacent densely populated areas. ■ Acknowledgement This study was funded by a grant from the East Coast Area Health Board.
References 1. Bacik I, Kelly A, O’Connell M, Sinclair H (2000) Crime and Poverty in Dublin: an analysis of the association between community deprivation, District Court appearance and sentence severity. Dublin: Round Hall Press 2. Carstairs V (1995) Deprivation indices: their interpretation and use in relation to health. J Epidemiol Community Health 49 (Suppl. 2):83–88 3. Clayton D, Bernadinelli L (1992) Bayesian Methods for Mapping Disease Risk. In: Elliott P, Cuzick J, English D, Stern R (eds) Geographical and Environmental Epidemiology: Methods for SmallArea Studies. World Health Organisation and Oxford University Press: Oxford, pp 205–220 4. Coid J, Kahtan N, Cook A, Gault S, Jarman B (2001a) Predicting admission rates to secure forensic psychiatric services. Psychol Med 31:531–539 5. Coid J, Kahtan N, Gault S, Cook A, Jarman B (2001b) Medium secure forensic psychiatry services: Comparison of seven English Health regions. Br J Psychiatry 178:55–61 6. Coid J (1998) Socio-economic deprivation and admission rates to secure forensic psychiatric services. Psychiatr Bull 22:294–297 7. Coid J, Kahtan N (2000) An instrument to measure security needs of patients in medium security. J Forensic Psychiatry 11(1):119–134 8. CSO (1997) Census 96. Stationery Office: Dublin 9. Daley DJ, Kendall DG (1964) Epidemics and rumours. Nature 204:1118 10. Daley DJ, Gani J (1999) Epidemic Modelling, Cambridge University Press: Cambridge, pp 133–153 11. Dalton T (1992) Urban Crime and Disorder: Report of the Interdepartmental Group. The Stationery Office: Dublin 12. De Alaracon R (1969) The Spread of Heroin Abuse in a Community. WHO Bulletin on Narcotics 29:17 13. Faris REL, Dunham HW (1960) Mental Disorders in Urban Areas: An Ecological Study of Schizophrenia and Other Psychoses. (2nd Edition) Hafner: New York 14. Glover GR, Robin E, Emami J, Arabscheibani GR (1998) A needs index for mental health care. Soc Psychiatry Psychiatr Epidemiol 33:89–96
15. Jacobs J (1962) The Death and Life of Great American Cities. Jonathan Cape (London) 16. Jarman B, Hirsch S, White P, Driscoll R (1992) Predicting Psychiatric Admission Rates. BMJ 304:1146–1151 17. Jonas K (1992) Modelling and suicide: a test of the Werther effect. Br J Soc Psychol 31(Pt 4):295–306 18. Kelly A (1999) Case studies in Bayesian disease mapping for health and health service research in Ireland. In: Lee A (ed) Disease Mapping and Risk Assessment for Public Health. John Wiley and Sons: London, pp 349–363 19. Kennedy HG, Iveson RCY, Hill O (1999) Violence, homicide and suicide: Strong correlation and wide variation across districts. Br J Psychiatry 175:462–466 20. Lester D (1988) A critical-mass theory of national suicide rates. Suicide and Life-Threatening Behaviour 18(3):279–284 21. Lewis G, David A, Andreasson S, Allebeck P (1992) Schizophrenia and city life. Lancet 340:137–140 22. McCrone L, Leese M, Glover G, Thornicroft G (2001) Social deprivation and the use of secure psychiatric beds. J Forensic Psychiatry 12:434–445 23. Mollie A (1999) Bayesian and empirical Bayes approaches to disease mapping. In: Lawson et al. (eds) Disease Mapping and Risk Assessment for Public Health. John Wiley and Sons: London, pp 15–29 24. Newman O (1972) Defensible Space: People and Design in the Violent City. Macmillan: London 25. O’Neill C, Kelly A, Sinclair H, Kennedy HG (2002) Interactions of general and forensic psychiatric services. Ir J Psychol Med (in press) 26. Porter R (1994) London: A Social History. Hamish Hamilton: London 27. Prunty J (1998) Dublin Slums 1800–1925: A study in Urban Geography. Irish Academic Press: Dublin 28. Richardson S (1992) Statistical Methods for Geographical Correlation Studies. In: Elliott P, Cuzick J, English D, Stern R (eds) Geographical and Environmental Epidemiology: Methods for Small-Area Studies.World Health Organisation and Oxford University Press, pp 181–204 29. Shepherd M (1983) Epidemiology and clinical psychiatry. In: The Psychosocial Matrix of Psychiatry. Collected Papers, Tavistock: London, pp 31–49 30. Statham DJ, Heath AC, Madden PAF, Bucholz KK, Bierut L, Dinwiddie SH, et al. (1998) Suicidal behaviour, an epidemiological and genetic study. Psychol Med 28:839–855 31. Stack S (1996) The effect of the media on suicide: evidence from Japan, 1955–1985. Suicide and Life-Threatening behaviour. 26(2):132–142 32. Schmidtke A, Hafner H (1988) The Werther effect after television films: new evidence for an old hypothesis. Psychol Med 18(3): 665–676 33. Torrey EF (1995) Jails and prisons – America’s new mental hospitals. Am J Public Health 85(12):1611–1613 34. Torrey EF (1997) Out of the Shadows: Confronting America’s Mental illness Crisis. John Wiley and Sons: New York 35. Townsend P, Phillimore P, Beattie A (1988) Health and Deprivation: inequality and the north. Croom Helm: New York 36. Vilani S (2001) Impact of media on children and adolescents: a 10-year review of the research. J Am Acad Child Adolesc Psychiatry 40(4):392–401