Throughout history, technological progress has transformed population health, but the distributional effects of these gains are unclear. New substitutes for older, more expensive health technologies can produce convergence in population health outcomes but may also be prone to elite capture and thus divergence. We study the case of penicillin using detailed historical mortality statistics and exploiting its abruptly timed introduction in Italy after WWII. We find that penicillin reduced both the mean and standard deviation of infectious disease mortality, leading to substantial convergence across disparate regions of Italy. Our results do not appear to be driven by competing risks or confounded by mortality patterns associated with WWII.

Mortality convergence, Mortality decline, Historical demography, Penicillin, Infectious disease mortality

Introduction

Technological progress in medicine has been described as a leading explanation for the increase in life expectancy in the post–World War II (WWII) era (Acemoglu and Johnson 2007; Davis 1956; Easterlin 1999; Mokyr 2002; Preston 1975) and is often proposed as a solution to health challenges in lower-income countries. Technological progress, however, can also have unintended consequences for the distribution of disease. If solely elites can afford a health technology or a technology is a private good that might substitute for the provision of public goods, such innovation could exacerbate health disparities (Ashraf et al. 2016; Kremer and Willis 2016; Mosca 1939; Olson 1965).^¹ Alternatively, breakthrough health technologies that are less expensive may have more widespread use than older, more expensive ones, resulting in population health convergence.

In this study, we analyze the effects of the introduction of penicillin, arguably one of the most important medical technologies of the twentieth century (Tomes 1990, 1998). Discovered to kill staphylococcus bacteria by Alexander Fleming in 1928—and successfully isolated and produced by Howard Florey, Ernst Chain, and Norman Heatley in 1939—this new “miracle drug” quickly became the first-line treatment for pneumonia, diphtheria, syphilis, gonorrhea, scarlet fever, and other infectious diseases (Dowling 1977; Levy 1992).^² Penicillin's population health benefits were preceded by those of sulfa agents, the first chemotherapy developed to fight infection.^³Jayachandran et al. (2010) found that the introduction of sulfa drugs led to substantial reductions in maternal mortality and pneumonia deaths in the historical United States.^⁴

We build on the Jayachandran et al. (2010) study in two ways: (1) by exploring the effect of penicillin on average mortality in Italy following its introduction by the United Nations in the aftermath of WWII; and (2) more generally, by documenting the effects of the technology on the distribution of mortality across regions of Italy. At the end of WWII, economic elites were disproportionately concentrated in the northern regions of Italy (and this was reinforced by the disproportionate allocation of funds to northern regions under the Marshall Plan, for example; Graziani 1985). Some scholars have suggested that the distribution of new technologies, even inexpensive ones, often benefit the elite first (Brenzel and Claquin 1994). Others have suggested that more inexpensive and portable technologies that do not require large-scale infrastructure investments are less prone to elite capture and have considerable potential to improve population health and promote health equity (Acemoglu and Robinson 2008; Mosca 1939; Olson 1965). For penicillin, the former possibility could be particularly true given that political decision-making often targeted expensive water and sanitation infrastructure—which can greatly reduce the spread of infectious diseases—toward more elite neighborhoods and communities (Bigatti 2014; Massarutto 2011; Picci 2002; Troesken 2004).^⁵

To study the distributional consequences of the introduction of penicillin, we assemble a data set of Italian vital records spanning much of the twentieth century (Atella, Francisci et al. 2017). Focusing on the period 1924–1955, we first establish penicillin's average effect on mortality by interacting the exact timing of penicillin's introduction with causes of death sensitive to penicillin, controlling for time- and region-level effects as well as regional linear time trends. We then estimate penicillin's effect on the distribution of mortality in four ways: (1) analyzing changes in the age distribution of deaths over time, using Kolmogorov-Smirnov (K-S) tests to detect significant differences in these distributions year by year; (2) using an econometric approach akin to our framework for estimating mean reductions in mortality rates but testing for differential declines by initial levels of infectious disease death rates; (3) testing for β convergence, following the literature on macroeconomic growth and economic convergence (Barro and Sala-i-Martin 1992); and (4) testing for σ convergence by estimating the relationship between penicillin's introduction and changes in the standard deviation of regional mortality rates (Janssen et al. 2016).

On average, we find timely reductions in penicillin-sensitive disease mortality rates across Italy that closely coincide with penicillin's 1947 introduction. We find no trend differences between penicillin-sensitive and penicillin-insensitive diseases before 1947, and the subsequent decline of about 0.3 deaths per 1,000 annually represents a 58% reduction relative to the average rate of penicillin-sensitive deaths in earlier years. Then, using all four approaches, we find clear evidence of mortality rate convergence across regions of Italy.

We also consider three important potential concerns with our analyses. The first is competing risks. Because our estimation framework relies on comparisons between penicillin-sensitive and penicillin-insensitive diseases, reductions in penicillin-sensitive deaths could mechanically increase noncommunicable disease mortality as those benefiting from penicillin live long enough to suffer and die from noninfectious causes. We show that competing risks do not play a major role in practice because a comparable decline in penicillin-sensitive death rates is present when we examine a simple change over time (i.e., not comparing with mortality rates from noncommunicable diseases). Moreover, even when the sample time frame is limited to a very short window following the introduction of penicillin (assuming that the competing risks phenomenon takes time to emerge), we still find statistically significant and quantitatively meaningful effects. The second potential concern is that because penicillin was introduced shortly after the end of WWII and infectious disease mortality rates commonly surge during wartime, the decline of these rates relative to noninfectious mortality rates could partly reflect the end of conflict (and mean regression), as Erdem et al. (2011) and Zapor and Moran (2005) found among military personnel. Unfortunately, we do not have data on military personnel and how many returned to particular areas of Italy. However, our results are similar across areas with varying degrees of war-related destruction and are robust to excluding the years of most intense conflict in Italy (1943–1945) from our estimation. Third, we consider the potential role of migration; although we can only imperfectly explore its role, we do not believe that it is primarily responsible for our results.

Background

Early Efforts to Combat Infectious Disease Mortality in Italy

At the time of unification in 1861, life expectancy at birth in Italy was approximately 29 years, and the crude death rate was about 35 per 1,000 people (Atella, Francisci et al. 2017). In 1887, Francesco Crispi's government introduced the country's first sanitary reforms under the Crispi-Pagliani law.^⁶ However, infectious disease deaths in Italy did not begin to decline until the early twentieth century, which historians have linked to improvements in municipal hygiene under legal provisions governingwater quality and sanitation in urban areas (Giovannini 1996; Giuntini 1999; Pogliano 1984). During fascist rule (1920–1943), public health initiatives targeted the triple endemic diseases of malaria, syphilis, and tuberculosis.^⁷

Italy's nascent healthcare system and reforms often failed to reach its poorest members (Giovannini 1996; Giuntini 1999; Pogliano 1984).^⁸ Municipal authorities were responsible for providing healthcare to the indigent. In practice, however, private charitable organizations provided these services (Opere Pie, or Pious Organizations”).^⁹ An 1884 royal decree established that private firms could invest in the water sector (Ermano and Massarutto 2012). Massarutto (2011) argued that this decree and the ensuing privatization directed water supply toward the wealthy in urban areas. As late as the 1950s, only 52% of private houses in Italy received potable water (Doria 2010), and only 7% had all three utilities: potable water, adequate sanitation, and electricity (Barra Bagnasco 1996).^¹⁰

Average mortality rates in Italy declined during the early- to mid-1900s (see Figure 1), although wide regional health disparities in infectious disease mortality were prevalent at the beginning of this period (see Figure 2, panel a). Nationally, life expectancy at birth in Italy rose to 40–45 years but varied by as much as 12 years across regions.^¹¹ These regional disparities peaked in 1925–1930, remained high until WWII, and then declined sharply at the end of the war. Understanding whether the introduction of penicillin contributed to the compression of regional mortality rates is a central focus of this paper.

The Advent of Penicillin in Italy

The supply of penicillin in Italy was widely available beginning in 1947 (Luzzi 2004). In its early stage of distribution (1945 and 1946), penicillin was available in Italy through international aid only: it was imported via the United Nations Relief and Rehabilitation Administration and was limited in quantity. Given the flourishing black market at that time, those who had enough money and connections had access to the drug. (The initial distribution of penicillin in Italy by the U.S. Army started with the main intent of using it to treat the spreading of venereal diseases.) ENDIMEA (Ente Nazionale Distribuzione Medicinali agli Alleati) was established to resolve this problem, beginning its work on October 1, 1944. On January 1, 1945, each provincial health office began to communicate to the General Directorate of Public Health (DGSP), within the Ministry of the Interior, estimates of the demand for medicines for the quarter. Based on the requests and the availability of drugs, the DGSP granted the drugs to the applicants, instructing ENDIMEA to ship the product to the local private wholesalers. Once the wholesalers collected the drugs, they were responsible for the distribution of medicines to the provincial health offices. The latter was also obliged to follow the drug from the wholesaler warehouse to the pharmacies or hospitals in order to prevent theft and illegal sales.^¹²

Italian patients could receive the new drug at no cost if requested by their physicians. Provincial health offices were held accountable for any medicine shortages, and pharmacists received a high margin for the sale of antibiotics, which discouraged the emergence of an underground market favoring the wealthy (Battini 1946; Luzzi 2004).

Data

Vital Statistics Data

The Italian National Statistical Office (ISTAT) provides national vital statistics in annual Health Statistics Yearbooks (Annuario di Statistiche Sanitarie) starting from 1887 (see section A of the online appendix for a complete description of the vital statistics). Vital statistics for 1924–1955 were digitized at the region, year, and cause-of-death level.^¹³

Harmonizing Regions

ISTAT data provide death counts for Italy's regions over time. Because of changes in administrative regional borders over our period of investigation, we had to aggregate some regions. From 1924 until the end of WWII, Italy was organized into 18 administrative regions, with only minor changes across region borders. The only exception was the establishment of Valle d'Aosta in September 1945. For this reason, to obtain a harmonized regional data set from 1924 to 1954, we treat Piedmont and Valle d'Aosta as a single region. Thus, our sample includes the following 18 regions: Piedmont and Valle d'Aosta, Lombardy, Trentino Alto Adige, Veneto, Friuli Venezia Giulia, Liguria, Emilia Romagna, Tuscany, Umbria, Marche, Lazio, Abruzzi and Molise, Campania, Apulia, Basilicata, Calabria, Sicily, and Sardinia.^¹⁴

Harmonizing Cause of Death

Officially, ISTAT assumed the responsibility of collecting individual-level mortality data only in 1930, following Law No. 2238 issued in December 1929. Previously, the statistical office was part of the Ministry of Agriculture, Industry and Trade, and collected statistics related to causes of death (“Indagine sulle cause di morte”). ISTAT collects mortality data using an official reporting format (Scheda di morte) following international standards recommended by the World Health Organization (WHO).^¹⁵ These individual death certificates consist of two parts. A general practitioner or coroner certifies the first part and ascribes a cause. The recorded initial cause—disease or trauma—may have led to additional complications but initiated a causal chain leading to death. Diagnostic codes are then assigned to each death using WHO International Classification of Diseases criteria. A municipal civil registrar completes the second part of the death certificate, which includes information about the demographic and social characteristics of the deceased.

Because causes of death reported in Italy's vital records have changed over time, we use categories that can be consistently identified and tracked across all study years, effectively adopting the classification used in 1956–1957 for all years of our analysis (1924–1955).^¹⁶ We then classify each of these causes that can be consistently tracked over time according to whether it can be treated with penicillin. Ultimately, this process yields 79 penicillin-insensitive, 15 penicillin-sensitive, and 15 unclassified diseases.^¹⁷ In our analysis of cause-specific mortality, we exclude unclassified deaths.

Constructing a War-Related Destruction Indicator

Because no official statistics contain information on destruction (which would be difficult to define and compute in any case), we measure WWII-related destruction using a proxy based on the region-level number of military and civilian deaths directly related to war between 1940 and 1945 (ISTAT 1957).^¹⁸ We obtain these statistics by collecting the number missing and dead due to the aerial and terrestrial bombing on the Italian territory. Clearly, destruction and the number missing and dead are directly correlated. In fact, as Atella, Di Porto, and Kopinska (2017) showed, where the war front lasted longer (i.e., Gustav line and Winter line), the local population witnessed a more intense military presence as well as more civilian deaths, often as a direct consequence of the destruction caused by the aerial and terrestrial bombing.

Constructing Rates

Finally, using death counts by region, year, and cause, we construct mortality rates using data on regional populations in Italy over time. Specifically, to create population denominators, we use regional population counts from Italy's decennial population censuses provided by ISTAT. For intercensal years, population estimates accounting for births and deaths provide estimates of this population construction process using several sources of population data (Vecchi 2017). (See section A of the online appendix for more details on population estimates.)

Our final sample contains region, year, cause-of-death observations for 1924–1959. Table A3 in the online appendix shows descriptive statistics for all-cause mortality from penicillin-sensitive and penicillin-insensitive diseases before and after penicillin introduction.

Human Mortality Database

We use national-level data on age-specific deaths from the Human Mortality Database (HMD) in two ways (HMD n.d.).^¹⁹ First, because the regional vital statistics do not contain information about deaths by age, precluding age adjustment, we control for age-specific population counts.^²⁰ Second, we directly examine changes in the age distribution of deaths over time as an alternative strategy for studying mortality convergence.

Graphical Analysis

Mortality Decline by Cause

As Figure 1 shows, Italy's total mortality rate fell substantially between 1924 and 1955. However, this decline varied dramatically by cause and, in particular, by penicillin sensitivity. Panel a of Figure 2 shows that with the official introduction of penicillin in 1947, the decline in regional mortality rates for penicillin-sensitive causes accelerated sharply. Moreover, the variance in regional death rates sensitive to penicillin became markedly more compressed beginning in 1947, with substantial convergence occurring by the mid-1950s. Panel b of Figure 2 demonstrates that regional mortality rates for noninfectious causes (or penicillin-insensitive diseases) did not decline over the entire period, nor did their variance.^²¹

Mortality Decline by Age

Figure 3 depicts the age distribution of period life table deaths in Italy for selected years between 1924 and 1955. The figure reveals substantial reductions in infant and child mortality over time (which was generally due to infectious causes, most of which are penicillin-sensitive; Vercelli et al. 2014). As a result, deaths appear to have become more concentrated at older ages. This pattern is consistent with convergence, although the timing of this compression is less readily evident (discussed in more detail in the Results).

Average Life Expectancy and Standard Deviation of Age at Death

Figure 4 shows both life expectancy at birth and the standard deviation of age at death across regions—a common indicator used in the literature to measure mortality convergence and lifespan inequality (Edwards and Tuljapurkar 2005; Gillespie et al. 2014)—by year. Life expectancy at birth increased from 51.5 in 1924 to 57.6 in 1939, declined abruptly during WWII until the 1943 Italian armistice, recovered quickly to its prewar level, and continued to increase in the postwar years. The standard deviation of age at death declined little during the prewar period (as life expectancy at birth was rising), remained relatively constant during WWII, and then declined precipitously beginning in the late 1940s (around the time that penicillin was introduced) and throughout the postwar period. This rapid decline beginning in the latter 1940s is suggestive of mortality convergence.

Standard Deviation of Regional Death Rates by Cause

Figure 5 shows the standard deviation of regional death rates separately for all causes, penicillin-sensitive diseases, and penicillin-insensitive diseases over time. Before WWII, the standard deviation trajectories of penicillin-sensitive and penicillin-insensitive disease mortality rates were similar. Both experienced disruptions during the war. After 1947, however, the standard deviation of penicillin-sensitive disease mortality rates across regions converged, whereas the standard deviation of penicillin-insensitive mortality rate slightly diverged.

Estimation

Mortality Decline

Building on graphical evidence shown in Figures 2 and 4, our empirical strategy tests for sharply timed differential trend breaks in penicillin-sensitive disease mortality rates (relative to penicillin-insensitive disease mortality rates) coincident with the introduction of penicillin in 1947. Specifically, we estimate

m_{i c t} = α + γ_{i} + δ_{t} + θ_{c} + β (S e n s i t i v e_{c} \cdot I_{t}^{p o s t}) + X_{i t} \cdot ψ + γ_{i} \cdot t + ε_{i c t},

(1)

for regions i, causes of death c, and years t. m_ict is a region-, cause-, and year-specific death rate (specified both in level and natural log form). $I_{t}^{p o s t}$ is a dummy variable equal to 1 for observations in 1947 or later, Sensitive is a dummy variable for whether cause c is sensitive to penicillin, and X is a vector of covariates including interpolated population estimates in discrete age categories. We include year, region, and disease fixed effects; our preferred specification also includes linear time trends interacted with region dummy variables. Standard errors are clustered by region, and we report p values corresponding to both wild and block bootstrapped standard errors.^²²

We also assess the temporal dynamics of the introduction of penicillin by substituting a vector of year dummy variables for

I_{t}^{p o s t}

in :

m_{i c t} = α + γ_{i} + δ_{t} + θ_{c} + \sum_{j} ρ_{j} (S e n s i t i v e_{c} \cdot I_{t}^{j}) + X_{i t} \cdot ψ + γ_{i} \cdot t + ε_{i c t},

(2)

where $I_{t}^{j}$ is a vector of j year dummy variables, and all other variables are defined as before.

Mortality Convergence

To study national-level mortality convergence, we first analyze changes in the distribution of age at death (Fries 1980; Kannisto 2000; Wilmoth and Horiuchi 1999). Specifically, we use Kolmogorov-Smirnov (K-S) tests to formally assess whether the timing of statistically significant changes in the distribution of age at death coincides with the introduction of penicillin in 1947, comparing the distribution in each year between 1924 and 1954 with the last year in our sample, 1955.^²³

Second, we test for mortality convergence associated with the introduction of penicillin using an econometric framework similar to the one we use for estimating mean reductions in penicillin-sensitive disease mortality rates. Specifically, we reestimate Eq. (1), stratifying by the pre-1947 level of both penicillin-sensitive disease mortality rates and overall regional mortality rates.

Third, we test for β convergence, or convergence in levels of mortality rates, following the approach of Barro and Sala-i-Martin (1992) to study convergence in gross domestic product (GDP) per capita across countries. This approach assesses whether regions with higher pre-1947 penicillin-sensitive and penicillin-insensitive disease mortality rates converged toward regions with lower mortality rates by subperiods (before and after the introduction of penicillin in 1929–1946 and 1947–1955, respectively). Specifically, we estimate

\frac{(m_{i c t} - m_{i c t_{0}})}{(t - t_{0})} = α + γ_{i} + β m_{i c t_{0}} + ε_{i t},

(3)

where m_ict is the mortality rate in region i due to cause c in initial year t₀ (t₀ = 1924 for the first subperiod and 1947 for the second subperiod), and β is the parameter of interest. Finally, we test for σ convergence, or convergence in the standard deviation of mortality rates, across regions of Italy by reestimating Eq. (1), using the standard deviation of cause-specific mortality rates by region as the dependent variable (Janssen et al. 2016; Young et al. 2008).

Results

Mortality Decline

Table 1 reports results obtained by estimating Eq. (1). Conditioning on year, region, and disease fixed effects as well as the regional age distribution, mortality rates from penicillin-sensitive diseases relative to penicillin-insensitive diseases fell by approximately 0.3 per 1,000 after the introduction of penicillin—a difference that is statistically distinguishable from 0 ( p < .001). This figure represents a 58% reduction relative to the mean mortality rate among deaths due to these causes before 1947 (0.469; see Table A3, online appendix).^²⁴ Column 2 adds region-specific linear time trends. In both columns, the estimate of β is robust, with the mortality rate decline associated with the introduction of penicillin remaining at about 0.3 per 1,000. Columns 3 and 4 repeat this estimation for log mortality, showing robust reductions in penicillin-sensitive disease mortality rates.

Figure 6 examines the dynamic pattern of mortality decline associated with the introduction of penicillin in Italy, showing estimates and 95% confidence intervals for each year-specific ρ_j in Eq. (2). The figure shows little evidence of preexisting declines in penicillin-sensitive disease mortality rates (relative to insensitive ones) before the introduction of penicillin. Year-specific estimates show a decline with the introduction of penicillin in 1947, becoming negative and statistically different from 0 by 1948 (and in all subsequent years) with the diffusion of penicillin. These year-specific estimates approach a decline of 2.8 per 1,000 deaths by 1955, the end of our study period.

Mortality Convergence

Figure 7 first shows year-by-year K-S p values for tests of differences between each year's distribution of age at death against the distribution in 1955 (the last year in our sample). These p values remain constant at nearly 0 for all years before the introduction of antibiotics, indicating strongly significant differences from 1955. These p values rapidly rise, becoming nonsignificant right after 1947 ( p > .05), thus indicating no difference from 1955. This timely compression in Italy's age distribution of deaths is highly consistent with the notion that the introduction of antibiotics led to mortality rate convergence.

Table 2 reports estimates obtained from Eq. (1), stratified by level of penicillin-sensitive disease mortality rates before 1947 (using quartiles of pre-1947 penicillin-sensitive disease mortality). The first and fourth columns show that after 1947, penicillin-sensitive disease mortality rates declined more in regions with higher (quartile 4) initial rates (by 0.315 per 1,000) than in those with lower (quartile 1) initial rates (0.24 per 1,000)—and significantly so. Compared with the mean of the pre-1947 penicillin-sensitive cause-specific mortality, these changes represent reductions of 67% and 51%, respectively.

Table 3 reports β convergence estimates from Eq. (3). Penicillin-sensitive mortality rates were almost four times greater after the introduction of penicillin than before (with β estimates of –0.021 vs. −0.075, respectively). By contrast, for penicillin-insensitive mortality rates, the estimate is not statistically significant for the pre-1947 period and is positive and statistically different from 0 thereafter, suggesting divergence in penicillin-insensitive disease mortality rates.

Finally, Table 4 (columns 1 and 2) presents estimates of σ convergence, with the standard deviation of deaths across regions and years for each of the two disease categories as the dependent variable. We find a decline in the standard deviation of penicillin-sensitive disease mortality after the introduction of penicillin (−0.113, p < .001) but not a decline in the standard deviation of penicillin-insensitive disease mortality (−0.004). The results remain consistent after truncating the sample in 1950 (row 2): the standard deviation of the mortality rate declines significantly for penicillin-sensitive diseases (−0.090, p < .001) but not for penicillin-insensitive diseases (−0.009).^²⁵

Identification Issues and Threats to Internal Validity

In this section, we present and discuss several identification issues and potential treats that could affect our interpretation of the effects of the introduction of antibiotics on infectious disease mortality. First, we consider the problem of competing risks—specifically, the survival benefit of antibiotics increased the exposure of the population to death from noninfectious causes. Second, we discuss how the end to World War II may have affected infectious disease mortality in Italy during our sample period. Third, we address the problem of potential selective migration.

Competing Risks

In our context, the problem of competing risks concerns a possible increase in noncommunicable disease mortality resulting from fewer people dying from infectious disease mortality due to the introduction of penicillin. This phenomenon can influence our estimates because we compare death rates from penicillin-sensitive and penicillin-insensitive diseases over time (Honoré and Lleras-Muney 2006; Peterson 1976; Tsiatis 1975). In this section, we study the influence of competing risks on our estimation of the contribution of penicillin to the mortality decline in Italy.^²⁶ We present a simple framework for defining the particular competing-risk problem inherent in our main estimating equation and describe an alternative approach to bound the estimates.

Period 0. In any period, deaths can be divided into mortality from two etiologies, E = {I, NC}, where I indicates infectious diseases and NC reflects noncommunicable diseases:

D = I + N C .

(4)

Deaths (D) from etiology E can be expressed as the mortality rate from E multiplied by the number of susceptible from that etiology at time t: E = µ^e,t ⋅ S^E^,t. We assume that the mortality rate from noncommunicable diseases is exogenous to penicillin and that N⁰ persons are surviving after Period 0 in this closed population without births or migration.

Period 1. Penicillin ( p) is introduced at the beginning of Period 1. This introduction immediately treats and cures those afflicted with infectious diseases, leading to a direct decline in mortality associated with diseases of infectious etiology:

D^{1} = S^{I}^{, 1} µ^{I}^{, 1} (p) + S^{N C}^{, 1} µ^{N C} .

(5)

Period 2. By Period 2, penicillin shows indirect effects: weaker individuals who otherwise would have died from infectious disease are now alive (e.g., the harvesting effect of infections has been reduced). These spared lives increase the size of the population susceptible to noncommunicable diseases:

S^{N C}^{, 2} = N^{0} - S^{I}^{, 1} µ^{I}^{, 1} (p) - S^{N C}^{, 1} µ^{N C} .

(6)

Recall that ∂µ^I,i / ∂p < 0 by the pharmacology of penicillin and ∂S^NC^,2 / ∂p > 0 by Eq. (6). Hence, penicillin's total effect over time reflects its direct effect on infectious deaths and its indirect effect on the population susceptible to noncommunicable diseases.

Our difference-in-differences estimate compares the changes in deaths related to infectious and noncommunicable diseases before and after the introduction of penicillin:^²⁷

(S^{I, 2} µ^{I, 2} (p) - S^{I, 0} µ^{I, 0}) - (S^{N C, 2} (p) µ^{N C} - S^{N C, 0} µ^{N C}) .

(7)

Given that the first term is negative (by the direct effects of penicillin) and the second, in the absence of an overall secular decline in mortality, is positive (by the indirect effects of penicillin), our estimates are biased away from the null.^²⁸ In our setting, though, mortality may exhibit a prominent secular decline for all causes, making difference-in-differences the preferred framework. Thus, estimates from both the simple and double differences are biased—the former by a secular decline and misattribution of penicillin to secular patterns and the latter by competing risks.

To gauge the relative importance of these two sources of bias, we zoom in on the period very soon after penicillin was introduced (i.e., Period 1) so that our difference-in-differences estimator is not influenced by competing risks (i.e., there is no indirect effect on noncommunicable diseases): $(S^{I, 1} µ^{I, 1} (p) - S^{I, 0} µ^{I, 0}) - (S^{N C, 1} µ^{N C} - S^{N C, 0} µ^{N C})$ ⁠. We are more likely to isolate the direct effect of penicillin on infectious disease mortality for the period shortly after penicillin's introduction because the harvesting and death for more chronic, noncommunicable entities would not have had time to occur. Specifically, we estimate the effect of penicillin on penicillin-sensitive (i.e., infectious) and penicillin-insensitive (i.e., noncommunicable) causes pre-1947 versus post-1947 using the full period ending in 1955 versus a truncated period ending in 1950, three years after penicillin's introduction in Italy.

These results are shown in Table 4. For the truncated period, the results demonstrate that both penicillin-sensitive and penicillin-insensitive disease mortality declined (row 2, columns 3 and 4), although the effect is much larger for the former. The estimates point to an important secular decline in mortality, independent of penicillin, hence the negative post coefficient in column 4. The difference-in-differences estimate ([–0.486] – [–0.363]) is approximately −0.12, which is smaller than the main estimates shown in Table 1 and a lower bound on penicillin's effects.

For the full period (ending in 1955), the effect of penicillin's introduction on insensitive disease mortality is more positive, reflecting the competing-risk issue. The magnitude of this change, however, is slight and statistically indistinguishable from the estimates using the 1950 cutoff. Moreover, the increasing difference-in-differences point estimates over time are attributable to the first term in Eq. (2): the direct effect of penicillin on infectious disease mortality (see row 3 in Table 4). As discussed earlier, this likely reflects the technology's diffusion.

Turning to sigma convergence, we see no major concern for competing risks in any specification. Convergence is limited to infectious causes.

In summary, the framework and results presented in this subsection show that the competing-risks problem in our context is small, given that a longer time window does not greatly bias the second term in Eq. (7) in a positive direction. On the other hand, controlling for a secular decline in mortality—which is what differencing out of noncommunicable causes affords—is important for obtaining less biased estimates.

The End of WWII

Another potential concern with our results is the influence of the end of WWII on infectious disease rates. Because infectious disease mortality rates commonly surge during wartime (Erdem et al. 2011; Zapor and Moran 2005), their decline relative to noninfectious mortality rates could partly reflect the end of conflict. Providing at least some prima facie evidence that this is not an important concern in our case – Figure 2 (panel a) shows that penicillin-sensitive disease mortality rates declined until 1930, generally remained stable through 1945 (rather than rising), and then resumed their decline. However, we probe this issue further in two ways.

First, we test for differential penicillin effects in areas with varying degrees of war-related destruction. Specifically, we reestimate Eq. (1) separately for regions with above- and below-median degrees of war intensity, measured using conflict-related mortality rates between 1940 and 1945.^²⁹ Columns 5 to 8 of Table 2 report estimates separately for regions with varying intensity of exposure to WWII (defined by quartiles of war-related destruction). In general, the penicillin effect is larger in areas with lower (quartile 1) war intensity (0.297 per 1,000) relative to those with higher (quartile 4) war intensity (0.252 per 1,000), and this difference is statistically significant.^³⁰

Second, we exclude the years of greatest conflict in Italy (1943–1945) from our sample and reestimate Eq. (1). Table C1 in the online appendix shows these results by quartiles of pre-1947 penicillin-sensitive disease mortality. The estimates are statistically equivalent to those that include 1943–1945.

The Role of Potential Selective Migration

A third concern related to the end of WWII is that patterns of differential migration could explain our main results. For this concern to be valid, selective migration (internal or international) would have to be plausibly related to the relative risk of penicillin-sensitive (vs. penicillin-insensitive) diseases, given a likely correlation between the relative risk of penicillin-sensitive diseases and socioeconomic status. However, three historical facts suggest that this concern may not be important in practice. First, evidence suggests that postwar migration did not change the composition of Italians along the dimensions of gender or household structure (specifically, individuals living alone or in single-parent families with a woman as the head of the household) (Gomellini et al. 2017). Second, some suggestive empirical evidence suggests that internal postwar migration in Italy was unrelated to health status (Atella et al. 2019). Third, following WWII, it took time for Italy to open its borders and organize migration procedures; between 1941 and 1950, migration rates in Italy were the lowest ever recorded, with the exception of the preceding decade under the country's fascist regime (Caselli and Capocaccia 1989; Del Boca and Venturini, 2005).

Policy Implications

Our findings focus on a single technology and a single context. However, this specific technology, penicillin, has been highly impactful in addressing a range of prevalent and fatal infectious diseases around the world. Further, the context of Italy's epidemiological transition during the twentieth century reflects a population health transformation that many countries worldwide have experienced.

Nonetheless, considerable scope for further population health gains remains to be realized through low marginal cost innovations in countries not yet undergoing, or completing, their epidemiological transitions. This is true not only for average rates of mortality (due to infectious diseases or other causes amenable to basic, good-quality primary care, for example) but also, importantly, for their distribution within populations (i.e., for health equity). In recent years, the three leading causes of death among infants and young children (as a share of all deaths in those respective age groups) have been acute respiratory infections, diarrheal diseases, and malaria (WHO 2018). In principle, such diseases can be addressed or managed with long-standing medical interventions or even adequate water and sanitation infrastructure (Alsan and Goldin 2019; Bhalotra et al. 2018; Cutler and Miller 2005), but they also are readily amenable to substitute technologies with low marginal costs. Globally, considerable population health convergence has already been achieved with the latter (Jamison et al. 2013); a notable example is the well-known “child survival revolution” launched by UNICEF during the 1980s, which helped to boost population coverage rates of Expanded Program on Immunization vaccinations in lower-income countries and made important contributions to global health equity (Jolly 2001). Nonetheless, considerable scope remains for further convergence in population through existing technologies with low marginal costs and the promise of new ones.

Conclusion

Although technological progress in health has produced dramatic gains in life expectancy over the past century around the globe, it can also have unintended consequences for the distribution of disease. On one hand, the accessibility and adoption of such technologies may be greater among elites, thereby exacerbating health disparities. Alternatively, health technologies that are inexpensive and substitute for older, more expensive ones may disproportionately benefit those who are poorer and less healthy, leading to population health convergence. Notably, little empirical evidence has shed light on the consequences of major new health technologies for the distribution of health in populations.

Studying the seminal case of penicillin and focusing on Italy using newly digitized vital statistics covering several decades throughout the twentieth century, we find that the introduction of penicillin significantly reduced average infectious disease mortality rates but also led to a substantial reduction in the variance of mortality rates across Italian regions. Specifically, the standard deviation in age at death (S₀) fell by nearly eight years—a decline of about 30%.

These results do not appear to be driven by competing risks or by population processes related to the end of WWII. More generally, our findings are consistent with previous research investigating cross-country population health convergence after WWII (Acemoglu and Johnson 2007; Deaton 2006). Overall, our results reinforce the idea that some forms of technological progress in health can be a powerful force in reducing health inequalities.

Acknowledgments

We would like to thank Giovanni Vecchi for useful discussion and comments during the early stage of this research project. Marcella Alsan gratefully acknowledges support from the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD), Grant 5K01-HD084709. Lena Schoemaker, Afia Khan, Anlu Xing, and Morgan Foy provided excellent research assistance.

Notes

Following seminal work on elites and elite capture in the political economy literature (e.g., Acemoglu and Robinson 2008; Bardhan and Mookherjee 2000), we consider elites to be those with disproportionate political power.

In a high-profile article, Florey et al. (1943) described the effectiveness of penicillin in treating wounded soldiers in North Africa. Some scholars have estimated that penicillin saved at least 300,000 lives during WWII alone (Dowling 1977; Levy 1992; Ratcliff 1945). It was only thanks to Margaret Hutchinson Rousseau, an American chemical engineer who designed the first commercial penicillin production plant, that penicillin was massively produced and made available worldwide.

Sulfa drugs are distinct from beta-lactam antibiotics, of which penicillin was the first developed (Mandell et al. 2010).

⁴

Conybeare (1948), Loudon (1988), and Mackenbach and Looman (1998) also found suggestive evidence on the importance of antibiotics. By contrast, some studies found that penicillin did not have the dramatic effect commonly attributed to it (e.g., on Finland and Sweden, see Hemminki and Paakkulainen 1976).

⁵

Technological progress may be necessary but not sufficient for health convergence: institutions may be critical to ensure that benefits are distributed to those most in need.

⁶

Historians interpret the establishment of a health system as one of the most important achievements of Italian “political and moral life” at the beginning of the new Kingdom (Croce [1928], cited in Cosmacini [2005:345]).

⁷

The most important provisions focused on malaria were issued between 1923 and 1934 through a series of laws designed to arrive at a thorough land reclamation operation (bonifica integrale). The Consolidation Act on the reclamation of marshlands was approved with Royal Decree No. 3256 on December 30, 1923. Law No. 3134 on December 24, 1928, called the Mussolini Law, granted financial resources to land reclamation and provided for integration of the drinking water supply and the construction of rural buildings, hamlets, and roads.

⁸

The law’s principles of universalism became effective only in 1978, when the Italian National Health Service System was established.

⁹

The nonpoor generally used services provided by individual private-practice doctors (medici libero-esercenti) who made home visits.

¹⁰

Until the end of WWII, national public spending for infrastructure was concentrated in the North. Regarding public investments in water, an 1884 royal decree created a dualism in the management of water services: water was public, but private companies could invest in the sector after having been awarded government licenses (Ermano 2012). As a result, water was largely provided to wealthy families in urban areas (Massarutto 2011). Despite many subsequent laws governing water management issued between 1861 and 1950, universalism in the water sector was not achieved until the late 1950s (Mantelli and Temporelli 2007). Modern water infrastructure in poorer southern regions was completed only in the late 1980s (Mantelli and Temporelli 2007).

¹¹

During the 1920s, the regional variation in life expectancy at birth was almost as large as that estimated for Indian states between 2011 and 2016, ranging from 61.5 years in Madhya Pradesh to 77 years in Kerala (Ponnapalli et al. 2013).

¹²

In the spring of 1945, the drug became available via physician prescription but only in the city public health offices (Ufficio d’Igiene). In 1947, the pharmaceutical company SPA Milan became the first private Italian company to market antibiotics, selling Supercillin (in vials) and Prontocillin (tablets). A year later, the Anglo-American monopoly over penicillin was broken by Domenico Marotta, then-director of the Istituto Superiore di Sanit, who called Ernst Chain to lead the International Centre for Microbiological Chemistry in Rome.

¹³

To the best of our knowledge, formal demographic analyses of the quality and completeness of Italy’s historical mortality statistics have not been published. However, Italian vital statistics back to the nineteenth century are included in the Human Mortality Database (HMD), and data quality is a central criterion for HMD inclusion. The HMD is available at http://www.mortality.org/. Data documentation for the Italy HMD is available at http://www.mortality.org/hmd/ITA/InputDB/ITAcom.pdf.

¹⁴

In December 1963, the region of Molise was established by splitting the Abruzzi and Molise into two distinct regions, giving Italy its current administrative structure based on 20 regions.

¹⁵

Before the post-WWII establishment of the United Nations, the Health Organisation of the League of Nations, which was established in 1919, provided the guidelines.

¹⁶

The 17 causes of death that we use are infective and parasitic diseases; tumors; allergic and endocrine gland diseases; blood and hematopoietic diseases; psychic and personality disorders; nervous system diseases; circulatory system diseases; respiratory system diseases; digestive system diseases; genitourinary system diseases; complications of pregnancy; skin and tissue diseases; bones and locomotive organs diseases; congenital malformations; early childhood particular diseases; senility and pathologic states; and accidents, traumatisms, and poisonings. The coding system changed from the International Analytical Classification during 1924–1955 to the ISTAT Intermediate Classification in subsequent years. Moreover, from 1958 onward, the inclusion of new death-related causes led to a higher disaggregation of diseases than in 1956–1957.

¹⁷

See section B of the online appendix for the list of penicillin-sensitive and penicillin-insensitive diseases.

¹⁸

We classify the war victims by the region of actual death, delivering a space-time quantification of war conflicts. The WWII severity indicator in Italy comes from an official ISTAT (1957) publication. The intensity of war destruction is quantified relative to the median. Regions with above-median conflict deaths are Piedmont-Valle d’Aosta, Veneto, Friuli Venezia Giulia, Emilia Romagna, Toscana, Umbria, Marche, Lazio, Lombardy, Trentino Alto Adige, Abruzzi and Molise, Campania, Puglia, Basilicata, Campania, Sicilia, and Sardegna. See Atella, Di Porto, and Kopinska (2017) for a detailed description of the data used. We then compute WWII-related death rates by dividing death counts by the average regional population in 1940–1945.

¹⁹

National-level HMD estimates are constructed by HMD investigators using vital statistics, population censuses, and population estimates directly from ISTAT and from other researchers working on behalf of ISTAT. Death counts in the Italian vital statistics are based on the de facto population (popolazione presente) until 1980 and on the de jure population (popolazione residente) thereafter. Therefore, mortality rates before 1981 in the HMD are based on population estimates of the de facto population, calculated from census counts to consider this change in the coverage of vital statistics (Glei 2015). We also adjust death counts from the vital statistics to include missing military deaths during World Wars I and II and to redistribute deaths spatially by age and calendar year (Jdanov et al. 2008). For 1937–1951, we use intercensal survival methods to derive population estimates using pre- and postwar census counts (Jdanov et al. 2008).

²⁰

These data are also available online at http://www.mortality.org/.

²¹

More work is required to address concerns about the confounding role of WWII given that infectious disease deaths often exceeded casualties directly due to conflict (Erdem et al. 2011; Zapor and Moran 2005). We address this issue later in the Mortality Convergence section. Further, one might also argue that WWII promoted internal migration flows from poorer to richer regions, thus contributing to higher differences among regional mortality rates. Atella et al. (2019) showed that migration within Italian regions began in earnest in 1951.

²²

To assess the robustness of our results, we also estimate variants of Eq. (1), using both levels and log specifications and including region-specific linear time trends.

²³

We use a two-sample K-S statistic to test whether two empirical one-dimensional distribution functions (F_m(x), G_n(x)) differ from each other. The K-S statistic is defined by the formula D_mn = sup_x |F_m(x) −G_n(x)|. The null hypothesis that the two samples come from the same distribution is rejected at the α level of significance when $D_{m n} > c (α) \sqrt{m + n / m n}$ ⁠, where $c (α) = \sqrt{- 1 / 2 ln(α / 2)}$ (Stephens 1974).

²⁴

To estimate this effect we divide the post-1947 mean average reduction in cause-specific mortality (the regression coefficient in the first column of Table 1, −0.272) by the mean mortality rate for penicillin-sensitive disease causes in the baseline period, between 1924 and 1946.

²⁵

We use a truncated sample to test the potential threat of competing risks in the interpretation of the effects of the introduction of penicillin on infectious disease mortality, including our interpretation of σ convergence (see the Competing Risks section).

²⁶

Technically, any problem of competing risks would amount to a violation of the stable unit treatment value assumption (SUTVA), which is necessary for unbiased estimates in a difference-in-difference framework such as ours.

²⁷

In our empirical framework, Eq. (7) is normalized by the total population. The population susceptible to infectious diseases at the beginning of Period 2 also experiences a positive effect, but we ignore that detail because it is dwarfed by the first-order effect of penicillin in reducing the mortality associated with many communicable causes.

²⁸

In the presence of a secular decline, the second term will be less negative than in the counterfactual of no indirect effect of penicillin.

²⁹

These data can be found in ISTAT (1957) and are available online at https://ebiblio.istat.it/digibib/Cause%20di%20morte/IST3413mortiedispersipercausebellicheanni1940_45+OCRottimizz.pdf.

³⁰

We test for the significance of the difference between the coefficients obtained in the separate models by pooling the samples used in both models, and specifying a triple interaction that includes an indicator for high and low destruction. The coefficient for this interactions term is statistically significant (p values < .001).

References

Acemoglu, D., & Johnson, S. (

2007

Disease and development: The effect of life expectancy on growth

Journal of Political Economy

115

925

–

985

	Cause-Specific Mortality per 1,000		Log of Cause-Specific Mortality per 1,000
	(1)	(2)	(3)	(4)
Sensitive × Post-1947	−0.272^***	−0.272^***	−0.611^***	−0.612^***
	(0.008)	(0.008)	(0.023)	(0.023)
Block Bootstrap SE	0.008	0.008	0.022	0.023
Block Bootstrap p Value	.002	.002	.002	.002
Wild Bootstrap SE	0.008	0.008	0.023	0.023
Wild Bootstrap p Value	.002	.002	.002	.002
Number of Observations	20,898	20,898	20,892	20,892
R²	.867	.868	.887	.887
Year Fixed Effects	Yes	Yes	Yes	Yes
Region Fixed Effects	Yes	Yes	Yes	Yes
Disease Fixed Effects	Yes	Yes	Yes	Yes
Region Trends	No	Yes	No	Yes
Number of Clusters	18	18	18	18

	Quartiles (Q) of Pre-1947 Penicillin-Sensitive Disease Mortality				Sensitivity to WWII Mortality Quartiles of War-Related Destruction
	Q1 (lowest)	Q2	Q3	Q4 (highest)	Q1 (lowest)	Q2	Q3	Q4 (highest)
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Sensitive × Post-1947	−0.240^***	−0.263^***	−0.277^***	−0.315^***	−0.297^***	−0.274^***	−0.261^***	−0.252^***
	(0.012)	(0.014)	(0.008)	(0.013)	(0.015)	(0.019)	(0.019)	(0.006)
Block Bootstrap SE	0.012	0.013	0.008	0.012	0.014	0.019	0.019	0.005
Block Bootstrap p Value	.002	.002	.002	.002	.002	.002	.002	.002
Wild Bootstrap SE	0.011	0.012	0.007	0.011	0.013	0.017	0.017	0.005
Wild Bootstrap p Value	.002	.002	.002	.002	.002	.002	.002	.002
Number of Observations	5,750	4,664	5,827	4,657	5,825	4,659	5,746	4,668
R²	.911	.883	.884	.889	.875	.881	.888	.925
Year Fixed Effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Region Fixed Effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Disease Fixed Effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Region Trends	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Number of Clusters	5	4	5	4	5	4	5	4

	Pre-Penicillin Period (1924–1946)		Post-Penicillin Period (1947–1955)
	High Penicillin-Sensitive Disease Mortality	Low Penicillin-Sensitive Disease Mortality	High Penicillin-Sensitive Disease Mortality	Low Penicillin-Sensitive Disease Mortality
	(1)	(2)	(3)	(4)
Baseline Mortality (1924 or 1947)	−0.021^***	−0.002	−0.075^***	0.009^***
	(0.001)	(0.002)	(0.002)	(0.002)
Number of Observations	162	424	161	508
R²	.913	.059	.936	.149
Region Fixed Effects	Yes	Yes	Yes	Yes

	Standard Deviation of the Mortality Rate (σ Convergence)		Mortality Rate
	Penicillin-Sensitive	Penicillin-Insensitive	Penicillin-Sensitive	Penicillin-Insensitive
	(1)	(2)	(3)	(4)
Post-1947: Full Sample (period ending 1955)	−0.113^***	−0.004	−0.600^***	−0.340^***
	(0.014)	(0.009)	(0.075)	(0.095)
Post-1947: Truncated Sample (period ending 1950)	−0.090^***	−0.009	−0.486^***	−0.363^***
	(0.014)	(0.009)	(0.075)	(0.097)
Difference in Post-1947 Coefficients (1955 – 1950)	−0.023	0.005	−0.114	0.023

Technological Progress and Health Convergence: The Case of Penicillin in Postwar Italy

Abstract

Introduction

Background

Early Efforts to Combat Infectious Disease Mortality in Italy

The Advent of Penicillin in Italy

Data

Vital Statistics Data

Harmonizing Regions

Harmonizing Cause of Death

Constructing a War-Related Destruction Indicator

Constructing Rates

Human Mortality Database

Graphical Analysis

Mortality Decline by Cause

Mortality Decline by Age

Average Life Expectancy and Standard Deviation of Age at Death

Standard Deviation of Regional Death Rates by Cause

Estimation

Mortality Decline

Mortality Convergence

Results

Mortality Decline

Mortality Convergence

Identification Issues and Threats to Internal Validity

Competing Risks

The End of WWII

The Role of Potential Selective Migration

Policy Implications

Conclusion

Acknowledgments

Notes

References

Supplementary data

Data & Figures

Contents

Supplements

Supplementary data

References

Related

Citing articles via

Email alerts

Related Articles

Related Topics

This Feature Is Available To Subscribers Only