International Journal of Biometeorology

, Volume 50, Issue 6, pp 335–341

The effect of birthplace on heat tolerance and mortality in Milan, Italy, 1980–1989


    • Dipartimento di BiologiaUniversità di Pisa e IFC-CNR-Pisa
  • Vito M. R. Muggeo
    • Dipartimento di Statistica e Matematica, “S. Vianelli”Università di Palermo
  • Rosanna Cusimano
    • Dipartimento di Prevenzione, Servizio di Sanità PubblicaEpidemiologia e Medicina Preventiva
Original Article

DOI: 10.1007/s00484-006-0035-x

Cite this article as:
Vigotti, M.A., Muggeo, V.M.R. & Cusimano, R. Int J Biometeorol (2006) 50: 335. doi:10.1007/s00484-006-0035-x


The temperature–mortality relationship follows a well-known J-V shaped pattern with mortality excesses recorded at cold and hot temperatures, and minimum at some optimal value, referred as Minimum Mortality Temperature (MMT). As the MMT, which is used to measure the population heat-tolerance, is higher for people living in warmer places, it has been argued that populations will adapt to temperature changes. We tested this notion by taking advantage of a huge migratory flow that occurred in Italy during the 1950s, when a large number of unemployed people moved from the southern to the industrializing north-western regions. We have analyzed mortality–temperature relationships in Milan residents, split by groups identified by area of birth. In order to obtain estimates of the temperature-related risks, log-linear models have been used to fit daily death count data as a function of different explanatory variables. Results suggest that mortality risks differ by birthplace, regardless of the place of residence, namely heat tolerance in adult life could be modulated by outdoor temperature experienced early in life. This indicates that no complete adaptation might occur with rising external environmental temperatures.


BirthplaceClimate change effectsHeat toleranceMigrantsMortality


The relationship between temperature and mortality has a well-known ‘J- or V-shape’ with mortality excesses at extreme cold and hot conditions. Between these extremes, mortality reaches its minimum at an optimal temperature value sometimes referred as ‘Minimum Mortality Temperature’ (MMT), which is usually taken as a rough measure of heat-tolerance of people living in that area (Kunst et al. 1993; The Eurowinter Group 1997; Beniston 2002; Keatinge and Donaldson 2000a,b; Curriero et al. 2002). The value of the MMT seems to be substantially higher in warmer regions (Curriero et al. 2002; Saez et al. 1995; Braga et al. 2001), while the effects of colder days seem to be stronger in warmer areas and effects of hotter days greater in colder areas. This supports the suggestion that humans adapt themselves to climatic conditions of the area where they live. When external temperatures change, either because of new local weather conditions or because people have moved to climatically different areas, the questions are whether and how the human body responds to the new conditions, by some physiological variations or by some adaptation strategy (Collins 1999; Leblanc 1988).

A possible tool to answer this question is to investigate the temperature effects in migrant populations which are exposed to environmental conditions different from those of their native areas. In Italy, during the 1950s and 1960s of the past century, more than three million Italian citizens moved around the country, mainly from southern regions towards the developed and industrial north-western regions, in particular Piedmont and Lombardy. This huge migration flow was a quite homogenous group of young adult males looking for jobs; in fact, 64% of these Italian migrants, in 1956–57, were aged 15–50 years, with a peak at 25–30 years (Golini 1974). Even by the time of the 1981 census, 17,6% of residents in the district of Milan, the main town of Lombardy, had been born in the South (National Institute of Statistics, Census 1981).

Birthplace has turned out to be an important factor in standard mortality analyses: in Italy, the first mortality atlases showed lear geographical trends in mortality decreasing with latitude (Facchini et al. 1985; Cislaghi et al. 1989), and mortality analyses for natives (born and died in the same area) compared to the migrated group (born in the south and died in the north) showed that, among residents in north-western areas, mortality was lower for those born in the south (Vigotti et al. 1988; Vigotti 1996). Furthermore, these mortality differences were found to be stronger among males of low socio-economic status and to weaken with time from migration (Costa et al. 1990). Substantially, considering the first generation of these migrants, the differences observed among people living in the same areas seems to mirror the differences existing between north and south.

This paper investigates short-term effects of temperature on mortality among two groups of residents in Milan split by birthplace; those born in north Italy and those migrated from south Italy. Moreover, for useful comparisons, temperature effects on residents in Palermo, the main town of Sicily, were also analyzed. The main focus will be on the mortality–temperature relationship and the MMT location. We are interested in evaluating whether early acclimatization can regulate, to some extent, heat tolerance; in this respect, the birthplace is considered as an indicator of the place where the early life was spent.

Materials and methods

The daily series analyzed in this paper have already been used in previous studies on short-term effects of air pollution on health: the 1980–1989 Milan dataset was analyzed in the European project APHEA-1 (Katsouyanni et al. 1995), and the Palermo dataset, including the period August 1996–December 1999, was analyzed in the MISA-1 study, the Italian meta-analysis (Biggeri et al. 2001), for the period 1997–1999.

Mortality and meteorological data

Anonymous data from death certificates came from Municipalities registries: only deaths for all natural causes (International Classification of Diseases, codes 1–799) and for both sexes were considered. The birthplace information is the one recorded on the death certificates, while no information on the year of immigration was available.

Temperature and relative humidity hourly data in Milan came from the meteorological station, which is centrally located, while in Palermo they came from the airport. Daily mean values were computed if 75% of hours was present and missing values were imputed as described in the previous cited projects.

Population subgroups

In 1980–1989, about 1,4 million people were living in Milan city (National Institute of Statistics, Census 1981). Mortality data of residents were analyzed according to two birthplaces: those ‘born in Lombardy’, the region around Milan (61.1% of all deaths) located north of 45°N of latitude, and those 'born in Sicily', the most southern region in Italy (3.6% of all deaths), located at a latitude south of 38,5°N. To obtain more accurate results, we also considered two wider groups of residents: (1) the group of people born in any northern region, including Lombardy, north of 44.5°N and located in the closed basin of the lowland of the Po river below the Alps Mountains and over the Apennine Mountains, where climate is typically continental (73.9% of all deaths); and (2) the group of those born in any southern region, including Sicily, south of 41.5° N, with a typical Mediterranean climate (15.6% of all deaths). Table 1 displays some descriptive statistics for the selected population groups, including those resident in Palermo.
Table 1

Summary of distribution of daily death, mean temperature and relative humidity in Milan, 1980–1989, and Palermo, 1997–1999


Daily values










Milan 1980–1989

  All residents








 Residents by birthplace

  Born in Lombardy








  Born in Sicily








  Born in the north








  Born in the south








Palermo, 1997–1999

  All residents









Milan 1980–1989








Palermo 1997–1999









Milan 1980–1989








Palermo 1997–1999








Milan resident deaths are divided by the birthplace

Meteorological differences

As reported in Table 1, the distributions of daily temperatures, measured in Lombardy and in Sicily, showed different values at low temperatures (the minimum was −6°C in Lombardy and ±2°C in Sicily), but not at higher ones (the maximum was 31.3°C in Lombardy and 32.3 °C in Sicily). However, the percentiles distributions are quite different, and the standard deviation was higher in Milan.

As data refer to different periods, a further comparison was done among temperature values published in the Official Meteorological Statistics (National Institute of Statistics, 1980–1989): details are omitted, but data showed similar patterns, providing evidence that the two periods are thus sufficiently comparable.

Statistical methods

Several methods have been used to estimate the temperature effects, while accounting for the V-like curve, including quadratic terms or two straight lines connected at the MMT taken to be known (Kunst et al. 1993; Saez et al. 1995; Diaz et al. 2002). According to the latter approach, left and right slopes represent respectively cold and hot risk for 1°C increase. In this paper, we also use such a piecewise linear approximation, but allowing the threshold value to be estimated in order to account for uncertainty of the threshold estimate (Muggeo 2004).

A segmented relationship between the response and a continuous variable X is generally modelled by including in the predictor the non-linear function
$$\beta X + \beta _{1} {\left( {X - \psi _{1} } \right)}_{ + } $$
being \({\left( {X - \psi _{1} } \right)}_{ + } = {\left( {X - \psi _{1} } \right)}\) if X1 and zero otherwise. According to the above terms, β is the left-slope and β1 is the difference-in-slope parameter when X exceeds its threshold value ψ1, the so-called breakpoint or MMT. Further changes in slope, as in ‘Lombardy natives’ later discussed, may be accounted by additional ‘(X−ψ)+- like’ terms with own breakpoint and difference-in-slope parameters.
Thus, accounting for confounding variables x, the Poisson log-linear model, for the generic observed series Y and for two breakpoints in the temperature relationship, is:
$$ {\text{Log E}}{\left[ {\text{Y}} \right]} = \delta ^{\prime } x + \beta Temp + \beta _{1} {\left( {Temp - \psi _{1} } \right)}_{ + } + \beta _{2} {\left( {Temp - \psi _{2} } \right)}_{ + } $$
where Temp is the temperature and δ includes parameters of confounding variables x such as day, year, months, days of week, holiday, influenza epidemics, relative humidity (unlagged, via regression splines with 4 df) and air pollution (unlagged particulate matter). This parameterization allows the estimatimation of the relative risks (RR) corresponding to different temperature intervals; hence for Temp1 RR it is exp{β}, for ψ1<Temp2 RR is exp{β+β1} and finally for Temp2 RR equals exp(β+β12). To estimate this model, we used a recent method which can overcome difficulties in breakpoint estimation (Muggeo 2003). This method yields point estimates and full covariance matrix for all parameters of the model and needs starting values just for the breakpoints, which are easily obtained by the smoothed scatter-plots. The temperature variable evaluated in the model, Temp, has been considered as the mean lag 0–1, i.e. mean of current and previous day, because it was the measurement associated with the best log-likelihood among other lags (lag 0, lag 1–2,..). Finally, possible over-dispersion was accounted for by the Pearson statistic and residual autocorrelation was checked by the partial autocorrelation function.

Seasonality modelling

Seasonality modelling has always been a crucial step in epidemiological time-series-based studies due to the strong seasonal pattern in mortality data. The modern approach includes parametric or nonparametric smoothers (Dominici et al. 2002), such as smoothingsplines where a difficult task concerns the selection of the amount of smoothing . Although several criteria have been discussed, including the Akaike’s or Bayesian Information Criteria or cross-validation, a unique decision has not been reached. Due to strong seasonality featuring the death counts and temperature series, the mortality–temperature relationship itself is strongly influenced by the amount of smoothing of the seasonality smoother. Different amounts of smoothing can lead to over- or under-estimates of the temperature-induced risks, yielding ‘unexpected’ bias in the estimates. ‘Unexpected’ because there is no guarantee over which is the direction of such bias.

Hence, to account for seasonality, the categorical variables ‘month’ and ‘year’ along with the quantitative ‘day’ have been used. Thus, while unexpected bias in the cold and hot estimated risks is reduced, possible over- or under-estimates leave unchanged the comparisons of major interest to us.


The smoothed estimates of mortality–temperature curves from fitted Poisson models are displayed in Fig. 1 for different groups of residents in Milan and for all residents in Palermo. A J-like relationship appears in each selected sub-groups, but the shapes differ. Not surprisingly, the curve of those born in Lombardy is very similar to the one for all residents and, as temperature increases, it exhibits two breakpoints with a very steep right slope. On the other hand, the curve of immigrants from Sicily shows just one breakpoint and it rises less steeply at hotter temperatures. The curves for the wider groups, i.e. people born in all northern regions and of immigrants from all northern regions, showed identical pattern but narrower confidence limits due to larger series (plots not shown).
Fig. 1

Smoothed mortality–temperature relationship for different groups: a all Milan residents; b Milan residents born in Lombardy ; c Milan resident born in Sicily; d all Palermo residents . Log (fitted), reported on Y axis, represent the log of expected values of mortality due to temperature effect

In Fig. 1d, the mortality–temperature curve for Palermo residents is displayed; just like Sicilian migrants in Milan, the curve exhibits only one breakpoint and a mild heat slope.

To obtain temperature-related risks, a piecewise parameterization was used, as discussed above; based on plots of the smoothed curves, one and two breakpoints were estimated for immigrants and natives, respectively, while a one-breakpoint relationship was fitted for the all residents in Palermo.

Table 2 reports the results from the fitted models: the breakpoints are expressed in degrees Celsius (°C) and temperature effects are provided as percent variation in the relative risk (RR) of dying for an increase of ±1°C in daily mean temperature above the estimated threshold.
Table 2

Estimates (95%CI) of temperature breakpoints ψ and % changes in relative risks of dying for a 1°C increase above the estimated thresholds in daily temperature, among residents in Palermo and in Milan by birthplace


Palermo 1996–1999


Milan residents by birth-place 1980–1989

All residents

Sicily region

Lombardy region

Southern regions

Northern regions

Breakpoints (°C)


23.2 (21.2, 25.1)

23.6 (20.8, 26.4)

18.9 (14.9, 22.9)

22.2 (20.4, 23.9)

18.8 (14.9, 22.9)


25.9(25.1, 26.7)

25.9 (25.1, 26.7)

RR (% change)


6.25 (3.80, 8.76)

6.56 (−1.25, 14.9)

0.61 (−0.51, 1.75)

4.45 (1.94, 7.02)

0.61 (−0.51, 1.75)


10.7 (6.37, 15.1)

10.6 (6.37, 15.1)

*The former breakpoint is the Minimum Mortality Temperature

All those born in Sicily or in any southern regions share similar MMT values, ranging from 22.2°C to 23.6°C, and also the estimated percentage RRs are quite similar, regardless of their residence; in particular, the percent RR increase is substantially the same between residents in Palermo (6.25%) and Sicilian migrants residing in Milan (6.56%). Unfortunately, for the last group, the small number of events leads to a rather uncertain estimate of the heat-RR including the zero.

For all residents born in Lombardy or in any northern region, the results show a reduced heat tolerance highlighted by a lower MMT value (estimated at around 19°C) and a considerable RR (10.6%) when temperature exceeds the latter threshold at 26°C.

Both MMT and heat induced RR might be considered to be different between migrants and natives; in fact, pair-wise comparisons reveal that one point estimate is not enclosed within the confidence interval of the corresponding estimate in the other group. O’Neill et al. (2003) have used such a guideline to compare estimates of cold- and heat-related risks between pairs of two groups. Actually, such differences may be found in all but one case; in fact, this is not true in comparing heat RR in Lombardy’s natives versus Sicily’s migrants, since neither of the estimates is outside the other’s confidence interval. However, as discussed above, the confidence limits estimated for the heat RR in Sicily’s migrants are quite wide due to the group’s small size.

Besides heat effects, another noteworthy finding concerns the effect of influenza epidemics. Among Milan residents born in the north, the percent increase in the risk of dying for influenza epidemics is 7.1% (95% CI: 4.9, 9.3), while among those born in the south percentage variations are wider: 11.6% (95% CI: 2.8, 21.1) and 10.5% (95%CI: 6.3, 14.9) among immigrants from Sicily and from the south, respectively. The estimate for residents in Palermo is 10.8% (95%CI: 4.3, 17.8).

Sensitivity analysis

A concern of this study may be the number of breakpoints in the native groups as suggested by plots in Fig. 1. To assess possible effects of the amount of smoothing, we used more degrees of freedom to smooth mortality–temperature relationships in the migrant groups: the relevant fitted curves appeared wiggly on the left side with no evidence for an additional breakpoint. Second, for residents born in Lombardy, we also fitted a model with only one breakpoint: the estimated MMT was found at 25.6°C (95% CI: 25.1, 26.1) and the percentage change in the heat risk was 11.7% (95% CI: 8.9, 14.6) for 1°C rise in daily temperature. However, using a simple Poisson model, the AIC selected a model with two breakpoints (AIC=21,394.3) rather than just one (AIC=21,404.8), suggesting that the former model should be preferred.

We also studied the ‘emptying effect’ of August due to summer holidays, and results excluding August did not differ: on the other end, extreme temperatures were recorded mainly in July.


In this study, the effects of daily temperature on natural mortality have been analyzed among residents in Milan, who were born (and likely have grown up) in Lombardy or in Sicily, two areas of Italy geographically and climatically different, and among residents in Palermo, mostly born in Sicily. Important climate differences exist in Italy between southern and northern areas, especially considering areas in the lowland of the Po river, on the northern side of the Apennine Mountains not moderated by the sea influence. Daily temperature in Milan shows a wider standard deviation than in Palermo, and this difference is larger in the warmer season (standard deviations equal to 4.59°C and 2.98°C in Milan and Palermo, respectively) than in the cooler one (standard deviations equal to 4.48°C and 3.89°C). Thus, due to somewhat notable climatic inequalities between north and south, it is reasonable to assess possible differences in response to the same exposure temperature among people with a presumed different acclimatization. The effects of temperature, expressed by the heat-related risk and the MMT, turn out to depend on the birthplace: people born in the same southern area, either migrated to the north or still resident in Palermo, share similar effects and exhibit a better tolerance to heat with respect to people born in the northern area. The MMT is higher, and the heat relative risk is lower for people born in south. Results from both groups of migrants (born in Sicily and born in the south) appear substantially similar, with wider confidence intervals for the estimates of the smaller group. This concordance in results gives more strength to the suggested hypothesis: early acclimatization seems to be an important factor in fixing the heat tolerance, and therefore people born in warmer areas bear the heat better than people born in colder areas.

Reasons of such differences are not clear, however. We do not know whether variation in thermal tolerance corresponds to some physiological variation during adult life or to some strategy of adaptation of human behavior (Collins 1999; Leblanc 1988). As the ability to adapt to environmental stresses declines with age, it is plausible to suppose that the human thermoregulatory system reaches, in the first part of life, some equilibrium with the external environment, where the individual may be expected to spend his or her life. On this hypothesis, individuals who have migrated as young adults to climatically different countries should maintain the same thermal tolerance developed in youth at their birthplace, and would suffer from different climatic conditions afterwards; this hypothesis may explain these results quite parsimoniously. Yet another explanation might be found in the variability of summer temperature which is higher in Milan than Palermo. Even if variance of summertime temperatures has been positively associated with higher mortality excesses (Braga et al. 2002; Kalkstein 1993), this motivation is not able per se to explain differences between residents exposed to the same temperatures.

Further explanations could be looked for in a different structure of the selected groups: migrants could be a biased group because of their low socio-economic levels or because of a “healthy migrant effect”. We also found that migrants from Sicily living in Milan are younger than natives probably because some retired migrants returned to Sicily; in fact, 71.9% of all deaths are over 65 years of age compared to 75.2% for those born in Lombardy. Nevertheless, even if these factors could account for the differences with natives in Milan, they do not explain the similarities with residents in Palermo, Sicily. Instead, some behavioral or cultural attitude associated with temperature, e.g. diet, may instead play a role in these similarities.

Finally, we think that the use of air conditioning may play a role, among Milan natives, in determining the peculiar mortality–temperature relationship with two breakpoints and justify the lower RR increase between them.

We also found that the effect of influenza epidemics on mortality seems to be similar between southern migrants living in Milan and residents in Palermo. We do not have an explanation for it, although it strengthens the similarities by birthplace. Recently, Jongbloet (2003) has suggested that a geographical gradient by latitude is likely in many constitutional diseases, where geographical location may play a role in disease causation together with lifestyles and generic behaviors.

This study does not address an important issue useful to go deeply into the role of the birthplace: the duration of or age at immigration. We are aware of such limits, but unfortunately such information was not available in our records.

A minor lack of the present study concerns the methodology: the lag structure used to model the temperature effects does not allow us to debate the cold effects or to investigate delayed effects and possible mortality displacements, i.e. harvesting; distributed lag model could be employed instead. However, our results remain valid in order to understand and quantify the short term effects of heat.

To our knowledge, the present study is the first where the birthplace is investigated as a possible factor in determining the temperature–mortality relationship regardless of the environmental exposure. The migrants constitute a useful population group where the hypothesis of early acclimatization (as opposed to ongoing adaptation) and possible consequences on the adult life may be studied. Although our results seem to support the hypothesis in favour of early acclimatization, clearly more research is needed to investigate further the role of birthplace on heat thermal tolerance; such a topic could be examined in similar ecological studies on routinely collected data, involving larger cities with non-negligible proportions of immigrants.

Copyright information

© ISB 2006