Contagious Transmission in a Swedish National Sample of Alcohol Use Disorders as a Function of Geographical Proximity Among Siblings and Propinquity-of-Rearing Defined Acquaintances

Abstract

Objective:

The purpose of this study was to determine whether alcohol use disorder (AUD) can be transmitted contagiously in siblings and likely acquaintances growing up close to one another (Propinquity-of-Rearing Defined Acquaintances [PRDAs]).

Method:

PRDAs were pairs of same-age subjects growing up within 1 km of each other and sharing the same school class, where one of whom (PRDA1) was first registered for AUD at age 15 or older. Using adult residential location, we predicted proximity-dependent risk for an AUD first registration in a second PRDA within 3 years of PRDA1's registration.

Results:

In 150,195 informative siblings, cohabitation status (hazard ratio [HR] = 1.22, 95% confidence interval [CI] [1.08, 1.37]), but not proximity, predicted risk for AUD onset. In 114,375 informative PRDA pairs, a log model fit best, predicting lower risk the greater the distance (HR = 0.88, 95% CI [0.84, 0.92]) with risks for AUD at 10, 50, and 100 km from affected PRDA1 cases equaling, respectively: 0.73 [0.66, 0.82], 0.60 [0.51, 0.72], and 0.55 [0.45, 0.68]. Within PRDA acquaintanceships, results resembled those found among PRDA pairs. The proximity-dependent contagious risk for AUD among PRDA pairs was attenuated by increasing age, lower genetic risk, and higher educational attainment.

Conclusions:

Cohabitation but not distance predicted transmission of AUD between siblings. However, contagious transmission of AUD among acquaintances growing up and attending school together was present and attenuated by increasing distance in adulthood. The impact of adult proximity on transmission was moderated by age, educational attainment, and genetic risk for AUD. Our results provide support for the validity of contagion models for AUD.

The notion of contagion [of alcoholism] qualifies here … [as] the impact of the other's sickness on oneself, by physical and social proximity to the drinker, insofar as the conditions for contagion to be possible include not only sharing the same physical (domestic) space, but also the existence of a social bond. (Fainzang, 1996, p. 473)

Alcohol Use Disorder (AUD) runs strongly in families (Cotton, 1979). Twin and adoption studies suggest that this transmission results largely from genetic factors, but familial–environmental factors are also important (Verhulst et al., 2015). Major efforts are also underway to identify the molecular genetic variants that have an impact on risk for AUD (Adkins et al., 2017; Lai et al., 2019; Mallard et al., 2022; Walters et al., 2018). Considerably less attention has been paid to clarifying how AUD is environmentally transmitted within families and groups of acquaintances or friends.

We here explore one such potentially important mechanism: contagion. As articulated in the above quote from Fainzang, this theory postulated that one individual, can, through a process of modeling and social learning (Bandura, 1986), increase the risk of AUD developing in a relative or friend who is in close touch with them. Such transmission within social networks, sometimes called “social contagion” (Christakis & Fowler, 2013), has been established for the use of common psychoactive substances including alcohol (Rosenquist et al., 2010; Slomkowski et al., 2009) and tobacco (Christakis & Fowler, 2008) as well as other complex biobehavioral syndromes such as obesity (Chris-takis & Fowler, 2007). Consistent with this hypothesis, we have previously shown in a Swedish national sample that, within marital pairs, an onset of AUD in one marital partner is associated with a large and rapid rise in risk of AUD registration in their spouse (Kendler et al., 2018). Furthermore, we found evidence for contagious transmission for AUD in parent–offspring and sibling pairs who reside in the same household, neighborhood, or municipality (Kendler et al., 2020b).

In this report, we expand our studies of the contagious transmission of AUD in two ways. First, we move beyond our prior simple classification of physical proximity (cohabitation vs. residence in small communities vs. in large metropolitan areas) and examine physical distances between the places of residence of sibling pairs. Second, we expand beyond the consideration of spouses or biological relatives to consider AUD transmission among friends/acquaintances, using the admittedly awkward (but accurate) term, Propinquity-of-Rearing Defined Acquaintances (PRDA). Traditional studies of social networks have relied on self-report data (where individuals nominate their friends) available only in special and inevitably small samples. In a previous article focused on drug use disorder (DUD), we proposed an alternative to such self-report data using the detailed geographical data available in the Swedish registries (Kendler et al., 2019a). We there demonstrated evidence for contagious transmission of DUD across pairs of PRDA.

As detailed below, we here improve on the measures previously used to define PRDAs. We now require them to not only live in the same small community within 1 km of each other for at least 10 of the first 15 years but also to (a) be in the same school at age 16, and (b) be broadly similar to one another in their academic performance.

Method

We collected information on individuals from Swedish population-based registers with national coverage linking each person's unique personal identification number, replaced with a pseudonymized serial number by Statistics Sweden to preserve confidentiality. Ethical approval was granted by the regional ethical review board in Lund (No. 2008/409 and later amendments). Participant consent was not required because the study was based on secondary, pseudonymized data. This study was conducted at Lund University and Virginia Commonwealth University.

We created two data sets. In the first, we included all full sibling pairs in which at least one in the pair, Sibling 1 (S1), was registered for AUD between ages 15 and 40 (for our definition of AUD, see Appendix Table 1). (A supplemental appendix appears as an online-only addendum to this article on the journal's website.) Furthermore, we required that both siblings were born in 1950 or later, had a maximum age difference of 10 years, that Sibling 2 (S2) was registered in Sweden and at least 18 years old at the time of registration for AUD in S1, and that the registration in S1 occurred in 1990 or later. For all pairs, we included information about place of residence at time of AUD registration in S1. The location of each individual has exact coordinates, which, for confidentiality reasons, were delivered to us truncated into 250 m × 250 m squares from which we calculated direct distances between places of residence. We used Cox proportional hazards model in which the main predictor was distance in kilometers between siblings at S1's registration. Follow-up began for S2 at S1's registration until registration for AUD—our main outcome variable—as well as death, emigration, and end of follow-up (3 years after S1's registration) but at the latest December 31, 2018. We tested several models allowing the relationship between distance and the outcome variable to differ and if it should include an extra effect if S1 and S2 cohabitated. The best-fitting model was chosen by Akaike's Information Criterion (AIC; Akaike, 1987). In the models, we controlled for age in S2 at S1's registration, S1 − S2 age differences, sex composition of the pair, and Small Areas for Market Statistics (SAMS) density, which reflected, within the SAMS where S2 resided at S1's registration, the proportion of AUD registrations in individuals with ±5 years age difference to S2 within a 3-year interval around that registration. For the analysis of sibships (three or more siblings), we included, in the Cox proportional hazards model, a separate stratum, and thereby a separate baseline hazard function, for each sibship.

For the second data set, we selected all individuals born in Sweden from 1975 to 1990. Based on information from Statistics Sweden, we created all possible pairs based on the following seven criteria: (a) same year of birth, (b) same SAMS during at least 10 of their first 15 years, (c) maximum 1 km distance during ages 0–15 (mean across 15 years), (d) same SAMS at age 15, (e) same school at age 16, (f) same year of graduation, and (g) maximum 1 SD difference in academic achievement at age 16 (for definition, see Appendix Table 1). Furthermore, among the potential acquaintance pairs, all biological relatives up to first cousins and stepsiblings were excluded.

Our database also included dates of first registration of AUD. We then selected all pairs in which at least one member (whom we termed PRDA1) was registered for AUD between ages 15 and 40 and measured the distance between them at the time of this AUD registration. This measure we call “distance in adulthood.” The follow-up time for PRDA2 began the day of AUD registration for PRDA1 and continued until registration for AUD (again, our main outcome variable), death, emigration, December 31, 2018, or end of follow-up (3 years after PRDA1's registration). To control for contextual effects of PRDA2's area of residence, we also included SAMS density of AUD. For each individual, we calculated a familial genetic risk score (FGRS) for AUD based on the morbid risk in their first- to fifth-degree relatives. For details, see Appendix Table 2. We also examined individual academic achievement obtained from the National School Registry as a grade point average at the end of grade nine (age 16).

We then performed Cox proportional hazards models for the risk of AUD in PRDA2 in the 3 years after PRDA1's first AUD registration as a function of the distance between PRDA1 and PRDA2 at the time of PRDA1's registration. In our Cox analyses, we examined various functions to best capture the relationship between adult distance and PRDA2 risk (e.g., linear, quadratic, logarithmic, splines) and chose the best fit model by AIC. In additional models, we first controlled for sex in PRDA2, distance in childhood, and SAMS density and then for indices of susceptibility, age at registration of AUD, FGRSAUD, and academic achievement in PRDA2 as well. In the models, we controlled for the fact that one PRDA1 may have multiple PRDA2s with a robust sandwich estimator. In an additional acquaintanceship model, we included a separate stratum for each unique PRDA1, allowing each PRDA1 and his/her PRDA to have a separate baseline hazard function. In the next analyses, we included interaction terms between distance in adulthood and the three indices of susceptibility. For the three interaction models, we used linear probability models so that the interactions were measured on the additive scale. All statistical analyses were performed in SAS 9.4 (SAS Institute Inc., Cary, NC). In accord with common practice in reporting descriptive data from epidemiological cohorts, we focus on the confidence intervals (CI) in our estimates rather than p values.

Results

Descriptive findings

As depicted in Table 1, we identified, for our analyses, 150,195 informative sibling and 114,375 informative PRDA pairs. Within 3 years of the onset of AUD in S1, 1.7% of the S2 subjects also had a first AUD registration. The parallel figure for the PRDA2 individuals was lower, equaling 0.61%. The average S1−S2 distance was only slightly shorter when S2 developed an AUD onset (27.3 km) than when they did not (29.8 km), but a considerably higher proportion of the affected siblings were cohabiting with the affected S1 (17.9%) than for unaffected siblings (11.6%). By contrast, the mean PRDA1 − PRDA2 distance was considerably shorter when PRDA2 was affected (76.7 km) versus not affected (98.7 km). Of note, PRDA2s who developed AUD were younger, had a higher genetic risk, had lower academic achievement, and lived in areas with a higher AUD density than PRDA2s who did not develop AUD.

Table 1.

Descriptive statistics on full sibling and Propinquity-of-Rearing Defined Acquaintances (PRDA) pairs from the Swedish population where at least one in the pair is registered for alcohol use disorder

Relationship	Alcohol use disorder
	All	Male to male	Female to female	Male to female	Female to male
Full siblings
n pairs	150,195	51,078	23,836	52,079	23,202
% in S2 (within 3 years)	1.7%	2.4%	1.5%	0.9%	2.3%
Sibling 2 affected, M or %
Distance in km	27.3	26.1	24.3	30.8	28.6
Distance in min.	25.8	24.6	23.5	28.6	27.2
Same household	17.9%	21.6%	11.1%	14.0%	17.8%
Age at registration S1	34.4	33.7	36.4	35.3	33.4
Age difference	3.9	3.9	3.8	3.8	3.9
Year of birth S1	1973	1973	1972	1972	1974
Year of birth S2	1973	1973	1972	1972	1974
SAMS density S1 (±3)	2.4%	2.4%	2.5%	2.4%	2.4%
SAMS density S2 (±3)	2.2%	2.1%	2.4%	2.2%	2.2%
Sibling 2 not affected, M or %
Distance in km	29.8	27.1	31.5	30.5	32.1
Distance in min.	27.8	25.5	29.4	28.5	29.8
Same household	11.6%	13.6%	10.1%	10.3%	12.1%
Age at registration S1	36.7	36.5	37.6	36.6	36.6
Age difference	4.1	4.1	4.0	4.1	4.0
Year of birth S1	1971	1971	1971	1970	1972
Year of birth S2	1970	1970	1970	1970	1971
SAMS density S1 (±3)	2.4%	2.3%	2.4%	2.4%	2.4%
SAMS density S2 (±3)	1.6%	1.5%	1.6%	1.5%	1.6%
PRDA
n pairs	114,375	40,626	17,456	27,149	29,144
% in PRDA2 (within 3 years)	0.61%	0.85%	0.47%	0.48%	0.48%
PRDA 2 affected, M (SD)
Distance in adulthood in km	76.7 (173)	56.4 (143)	88.9 (150)	99.0 (230)	99.1 (190)
Distance in childhood in km	0.5 (0.3)	0.4 (0.3)	0.4 (0.3)	0.5 (0.2)	0.5 (0.2)
Age at registration in PRDA1	26.3 (5.7)	26.1 (5.4)	27.5 (6.4)	25.6 (5.4)	26.9 (6.1)
SAMS density PRDA2	1.4 (1.0)	1.4 (1.1)	1.4 (0.9)	1.2 (0.9)	1.5 (1.1)
Genetic risk PRDA2	0.4 (1.2)	0.4 (1.2)	0.4 (1.4)	0.3 (1.0)	0.2 (1.2)
Academic achievement PRDA2	-0.5 (0.8)	-0.6 (0.8)	-0.6 (0.9)	-0.5 (0.9)	-0.4 (0.8)
PRDA 2 not affected, M (SD)
Distance in adulthood in km	98.7 (174)	88.0 (164)	113.2 (150)	102.1 (177)	101.8 (178)
Distance in childhood in km	0.5 (0.3)	0.5 (0.3)	0.5 (0.3)	0.5 (0.3)	0.5 (0.3)
Age at registration in PRDA1	28.1 (6.0)	28.1 (5.8)	28.1 (6.2)	28.2 (6.0)	28.1 (6.0)
SAMS density PRDA2	1.1 (0.9)	1.1 (1.0)	1.1 (1.0)	1.1 (0.9)	1.1 (0.9)
Genetic risk PRDA2	0.0 (0.9)	0.0 (0.9)	0.0 (0.9)	0.0 (0.9)	0.0 (0.9)
Academic achievement PRDA2	-0.1 (0.8)	-0.1 (0.8)	-0.0 (0.8)	-0.1 (0.8)	-0.1 (0.8

Notes: S1 = sibling 1; S2 = sibling 2; min. = minutes; SAMS = Small Areas for Market Statistics.

Sibling pair analyses

We first tested, as a dichotomous positive predictor variable in a Cox regression, cohabitation status. This was significantly associated with S2's risk for developing AUD: hazard ratio (HR) [95% CI] = 1.22 [1.08, 1.37]. In sibships with one S1 and multiple S2s, we saw a similar relationship: HR = 1.29 [1.02, 1.63]. This latter result is important because it controls for most potentially confounding background factors in comparing the risk for AUD onset in siblings of the same individual who were versus were not cohabiting with S1 at the time of his/her AUD onset.

The left column of Table 2 depicts the model fit for the effects of S1 − S2 proximity on risk for AUD in S2. The best fit was seen for a model incorporating linear and quadratic effects. However, neither the linear effect (1.001 [0.999, 1.004], p = .24) nor the quadratic effect (1.000 [1.000, 1.000], p = .07) was statistically significant.

Table 2.

Akaike Information Criterion values for models of the impact of distance on risk for AUD in the siblings not cohabiting or Propinquityof-Rearing Defined Acquaintances (PRDA) within 3 years of AUD onset in proband

Variable	Siblings	PRDA
No distance effect	48,858.3	16,159.814
Linear effect only	48,858.5	16,151.250
Linear + Quadratic effect	48,856.7	16,150.597
Log	48,860.0	16,118.408

Notes: Best fit model in bold. AUD = alcohol use disorder.

Propinquity-of-Rearing Defined Acquaintances Analyses— Main effects

As seen in the right column of Table 2, the log function was the best fit model to describe the relationship between AUD transmission and geographical proximity of our PRDA1 and PRDA2 pairs. This effect was significant (HR = 0.87 [0.83, 0.91]) (Model A Table 3) with a HR less than one meaning that risk decreased with increasing distance between the two acquaintances. We then added three covariates for PRDA2 to the prediction in Model B, Table 3: sex, distance between PRDA1 and PRDA2 in childhood, and SAMS density ofAUD. The latter variable is important because it controls for potential selection effects whereby PRDA2 resides in geographical areas with low or high AUD risk. Of note, including these variables has only a minimal effect on the prediction of AUD risk in PRDA2: 0.88 [0.84, 0.92].

Table 3.

Results from best-fit model for the impact of distance on risk for AUD in PRDA

Variable	Model A PRDA pair HR [95% CI]	Model B PRDA pair HR [95% CI]	Model C PRDA pair HR [95% CI]	Model D Acquaintanceship HR [95% CI]
Log of distance	0.87 [0.83, 0.91]	0.88 [0.84, 0.92]	0.94 [0.90, 0.98]	0.93 [0.88, 0.99]
Males vs. females		1.40 [1.20, 1.65]	1.43 [1.22, 1.68]	1.84 [1.46, 2.32]
Distance in childhood		1.02 [0.76, 1.37]	1.06 [0.79, 1.42]	1.05 [0.72, 1.55]
SAMS density		1.18 [1.15, 1.22]	1.17 [1.14, 1.21]	1.18 [1.06, 1.32]
Age at registration			0.97 [0.96, 0.99]	-
fgrsAUD			1.24 [1.18, 1.30]	1.26 [1.18, 1.36]
Grades			0.56 [0.51, 0.60]	0.53 [0.45, 0.61]

Notes: AUD = alcohol use disorder; PRDA = Propinquity-of-Rearing Defined Acquaintances; HR = hazard ratio; CI = confidence interval; SAMS = Small Areas for Market Statistics; FGRS = familial genetic risk score.

The log curve and its 95% CIs describing the relationship between distance and risk in PRDA2 is seen in Figure 1. It depicts a steep drop in risk for transmission of AUD over the first 25 km of distance between PRDA1 and PRDA2 and then begins to level out and becomes, by 250 km, nearly flat. The predicted risk for AUD in PRDA2 residing at five representative distances from PRDA1 were, respectively, as follows: 10 km, 0.73 [0.66, 0.82]; 25 km, 0.66 [0.57, 0.76]; 50 km, 0.60 [0.51, 0.72]; 100 km, 0.55 [0.45, 0.68]; and 200 km, 0.50 [0.40, 0.64].

Figure 1.

Risk for alcohol use disorder (AUD) as a function of distance in Model B for the PRDA analyses. The association of distance between PRDA1 and PRDA2 at the time of PRDA1's first registration for AUD and the hazard ratio, from a Cox regression model, for a first registration for AUD in PRDA2, with a value of 1 equaling the risk for PRDA2 living “next door” to PRDA1. The solid black line represents the results of Model B for PRDA pairs in Table 2, which depicts a log model of distance controlling for sex, mean distance between PRDA1 and PRDA2 in childhood, and the community density of AUD in the community in which PRDA2 resided at the time of PRDA1's onset for AUD. The dotted black lines represent the 95% confidence intervals around these estimates. PRDA= Propinquity-of-Rearing Defined Acquaintance.

Potential moderation of contagious effects

Models of contagion often contain indices of susceptibility (or “immunity”) of potentially exposed individuals. We here explore three: age, genetic risk, and academic achievement. We first examine the main effects of these predictors in Model C in Table 3. As predicted, younger age, higher genetic risk, and lower academic achievement all increased risk for AUD. Next, we asked the more important question for these analyses—are these susceptibility factors for the impact of contagion? That is, do they moderate the impact of contagious exposure as indexed by distance?

As seen in Appendix Table 3, we explored this by examining interactions, on an additive regression scale, for the impact of these risk factors and distance between PRDA1 and PRDA2. We see that each of these risk factors interact in the expected direction with the geographical proximity of PRDA1 and PRDA2. That is, the predisposing effects of young age (p = .008), high genetic risk (p = .003), and poor academic achievement (p < .0001) on AUD risk were stronger in PRDA2 individuals who lived close to versus relatively distant from PRDA1. We illustrate these effects in Figures 2a–2c. PRDA2 individuals under the median age of 27 had appreciable and significant decreases in AUD risk as a function of their proximity to the PRDA1 affected individual (Figure 2a). By contrast, for PRDA2 individuals at the 75th and 95th percentiles of age (ages 32 and 39, respectively), proximity to PRDA1 had no significant effect on risk for AUD. Risk for AUD in PRDA individuals with low genetic liability were similarly unrelated to proximity to PRDA1, whereas a quite different picture was seen for those at high risk, especially those in the highest 5% (Figure 2b). The largest effect was seen for academic achievement (Figure 2c), where PRDA subjects with particularly low performances were quite sensitive to proximity effects whereas these effects were quite modest to nil in those with median or higher levels of attainment.

Figure 2A.

Interaction between log of distance and age at registration. A Cox model incorporating the log of distance predicting risk of alcohol use disorder (AUD) registration in PRDA2 including age of PRDA2 at the first AUD registration of PRDA1 (multiple colored lines), distance between PRDA1 and PRDA2 at time of registration in PRDA1 (x-axis), and their interaction in the prediction of the risk for AUD in PRDA2 (y-axis). The model controlled for sex, mean distance between PRDA1 and PRDA2 in childhood, and the community density of AUD in the community in which PRDA2 resided at the time of PRDA1's onset for AUD. Depicted are the estimated curves from the 5th to the 95th by the percentile of age distribution in PRDA2 with their 95% confidence intervals and the mean age to which that is equivalent. PRDA = Propinquity-of-Rearing Defined Acquaintance; p-tile = percentile.

Figure 2C.

Interaction between log of distance and academic achievement. A Cox model incorporating the log of distance predicting risk of alcohol use disorder (AUD) registration in PRDA2 including academic achievement for PRDA2 at age 16, distance between PRDA1 and PRDA2 at time of registration in PRDA1 (x-axis), and their interaction in the prediction of the risk for AUD in PRDA2 (y-axis). The model controlled for sex, mean distance between PRDA1 and PRDA2 in childhood, and the community density of AUD in the community in which PRDA2 resided at the time of PRDA1's onset for AUD. Depicted are the estimated curves from the 5th to the 95th by the percentile of age distribution in PRDA2 with their 95% confidence intervals and the mean age to which that is equivalent. PRDA = Propinquity-of-Rearing Defined Acquaintance.

Figure 2B.

Interaction between log of distance and familial genetic risk score for alcohol use disorder (AUD). A Cox model incorporating the log of distance predicting risk of AUD registration in PRDA2 including genetic risk for AUD for PRDA2, distance between PRDA1 and PRDA2 at time of registration in PRDA1 (x-axis), and their interaction in the prediction of the risk for AUD in PRDA2 (y-axis). The model controlled for sex, mean distance between PRDA1 and PRDA2 in childhood, and the community density of AUD in the community in which PRDA2 resided at the time of PRDA1's onset for AUD. Depicted are the estimated curves from the 5th to the 95th by the percentile of age distribution in PRDA2 with their 95% confidence intervals and the mean age to which that is equivalent. PRDA = Propinquity-of-Rearing Defined Acquaintance.

Groups of Propinquity-of-Rearing Defined Acquaintances

Finally, we examined the main effects of proximity in “acquaintanceships” consisting of one PRDA1 individual and multiple PRDA2 individuals. Thus, controlling for all the potential confounding features of PRDA1, we examined whether among multiple PRDA2 individuals, their risk for AUD is influenced by their proximity to PRDA1. These results are presented in the right-hand column of Table 3. Most importantly, the parameter estimates for all of the key variables were very similar for the PRDA pair and acquaintanceship samples. These results validate our method and suggest that potential confounders in PRDA pairs, which would be controlled for within acquaintanceships, are not likely to have a substantial impact on our findings.

Discussion

Our goal was to determine whether, using cohabitation and geographical proximity as two different indices of the potential magnitude of interpersonal contact, we could find further evidence to support a contagion model for AUD transmission in sibling and acquaintances/friends—here termed PRDA pairs. We begin by reviewing our five main findings.

First, in full-sibling pairs, we found strong evidence of a contagion effect of cohabitation. Controlling for a range of background factors, S2 subjects living with an S1 subject with a first onset of AUD were ~20% more likely to develop AUD over the next 3 years than those not cohabiting. We found a nearly identical effect size result when examining sibships of three or greater, providing evidence that this contagious effect is unlikely to result from a range of possible familial confounders.

Second, to our surprise, we found no effect of geographical proximity in noncohabiting siblings on the transmission of AUD. That is, the chance for S2 to develop AUD after the onset of S1 was independent of the distance between them in adulthood.

Third, among PRDAs, a robust effect of proximity was seen on the transmission of AUD after onset in PRDA1. Compared with a PRDA2 living “next door” to a PRDA1 with a first onset of AUD, PRDA2 individuals living 10, 25, and 200 km away had roughly a 25%, 33%, and 50% reduced risk, respectively, of developing AUD over the next 3 years. Importantly, we included as a covariate in our regression models the density of AUD cases in the community in which PRDA2 individuals were residing, thereby controlling for self-selection effects in which vulnerable individuals sought out residence in high-risk communities.

Fourth, we repeated our proximity analyses in acquaintanceship groups with one PRDA1 and multiple PRDA2 individuals. The results were nearly identical to those originally found, which considerably reduces the chances that our findings are driven by background confounders.

Finally, given the robust evidence for proximity-driven contagious effects for AUD among PRDA1–PRDA2 pairs, we examined the impact of potential moderators—that is, indices of susceptibility. Metaphorically, one can imagine these as psychosocial proxies for antibody levels for the transmission of viruses. All three of our putative indices significantly interact with proximity to the PRDA1 case. PRDA2 who have high genetic risk to AUD, low academic achievement, and/or young age were especially sensitive to the contagious effects of AUD communicated to them by living close to an acquaintance with a recent AUD onset. Our findings are also consistent with prior evidence that early drinking and low education attainment favors the eventual development of AUD (Rosoff et al., 2021; Skala & Walter, 2013).

In aggregate, our results provide considerable further evidence in favor of a contagious transmission of substance use disorders in a general population of individuals (siblings or acquaintances) selected for the likelihood of significant interpersonal relations. Most specifically, our findings are consistent with two prior studies of AUD in Sweden showing transmission of AUD between spouses (Kendler et al., 2018) and parent–offspring pairs who reside in the same household versus neighborhood or municipality (Kendler et al., 2020b). Similar contagious effects for DUD have also been demonstrated in Sweden using cohabitation in the same household, neighborhood, or municipality (Kendler et al., 2020a) and proximity measures similar to (but not identical with) those used here in siblings (Kendler et al., 2020a) and in PRDA (Kendler et al., 2019b). Our DUD analyses of proximity in PRDA1–PRDA2 transmission also showed a moderating effect of age, genetic risk, and academic achievement.

Our results are also in line with prior studies which demonstrate that, consistent with the predictions of social learning theory (Bandura, 1986) and substance use disorder and epidemiological theorists (Anthony, 2006; Benedict, 2007; Dishion & Dodge, 2006; Ma et al., 2015; Ramamoorthi & Muthukrishnan, 2021), problematic alcohol use can be transmitted through a process of contagion within networks of relatives, friends, or acquaintances (Christakis & Fowler, 2013). As noted above, social contagion has been previously established for the use of alcohol (Rosenquist et al., 2010; Slomkowski et al., 2009) and tobacco (Christakis & Fowler, 2008) as well as obesity (Christakis & Fowler, 2007), depression (Rosenquist et al., 2011), and happiness (Fowler & Christakis, 2008). The demonstration of such social transmission pathways for the spread of AUD has obvious implications for prevention strategies as does our demonstration that particular individual risk factors (age, genetic risk, and prior academic performance) can have a strong impact on the susceptibility of individuals to contagious transmission of AUD. Although data are not available in the Swedish registries to test the specific mechanisms of contagious transmission of AUD, prior work, largely done within families, suggests two processes: direct modeling (e.g., Kerr et al., 2012; White et al., 2000) and impact on alcohol expectancies (Smit et al., 2018).

Our study has one troubling finding—the absence of significant proximity effects for the transmission of AUD within sibling pairs. The mean distance between S1 − S2 pairs was modestly smaller when S1 developed AUD versus did not (Table 1), but, as our modeling reveals, these effects were not significant. On average, we would expect the social relations to be stronger among siblings than PRDA and we did see a clear cohabitation effect on AUD transmission in the siblings. Although we still judge our results to be consistent with the contagion model for AUD specifically and substance use disorders more broadly, in large part because of the consistency of our results, our negative finding for proximity effects in siblings suggests some caution in the interpretation of our results.

Potential limitations

These results should be interpreted in the context of six potential methodological limitations. First, the validity of our results depends on the accuracy of the registry diagnoses of AUD. Although registry data have important advantages (e.g., absence of refusals and self-report bias), they will not identify the same cases as would interview-based assessments. Our affected subjects were likely, on average, more severely ill than those meeting Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (American Psychiatric Association, 2013), criteria at interview. However, our rates of under-ascertainment may not be very large because our lifetime prevalence of AUD was only moderately lower than those identified by a large interview-based study in nearby Norway (Kringlen et al., 2001). The validity of our definitions was supported by the high rates of concordance observed across our ascertainment methods (Kendler et al., 2015, 2018). Furthermore, the pattern of resemblance in relatives for AUD seen in Sweden was similar to those found in other samples based on personal interviews (Prescott & Kendler, 1999).

Second, we used “crow-fly” distances. We re-analyzed our sibling data using estimates of driving time calculated from Swedish road maps. No appreciable change was observed and the correlation between crow-flies and driving time measures of distance was very high (r = +.987).

Third, some prior evidence suggests that the contagious transmission of substance use disorder might differ by sex. Therefore, we explored the magnitude of the effect in male–male, female–female, male–female, and female–male PRDA1 and PRDA2 pairs (Appendix Table 4). Consistent with some prior expectations, the effects were strongest in male–male pairs, but the CIs were overlapping, suggesting that we were not well powered to examine such effects, especially in females with considerably lower levels of AUD.

Fourth, we sought to validate our PRDA model by examining whether PRDA1 and PRDA2 pairs differing in age might have attenuated the effect of proximity on potential contagion of AUD obtained in the same-age pairs used in our analysis. As seen in Appendix Figure 1, such attenuation was indeed seen. The effect became nonsignificant once the PRDA1–PRDA2 age difference increased to 4 or more years.

Fifth, we did not explore formally the length of the PRDA2's residence in the SAMS in which they resided when the PRDA1 had his/her AUD onset. Therefore, we performed a sensitivity analysis by excluding pairs in which the PRDA2 had resided less than 3 years in the SAMS area at the time of PRDA1 AUD onset (22.7%) (for details, see Appendix Table 5). Comparing the results for Model C, we see that the effect of distance is identical but with slightly broader 95% CIs.

Finally, as noted above, we do not possess self-reported identification of friends in any Swedish registry data. Therefore, our definition of PRDA is only an approximation and could identify individuals who were close friends while growing up and individuals who had little or no contact with each other. Our hope, which we consider supported by our findings, is that the representativeness and size of our sample in part compensates for the lack of this more detailed information.

Conclusions

Because of rapid advances in the technology of DNA genotyping and sequencing, most of the attention of the alcohol research community has recently been focused on the molecular genetic contribution to the well-documented familial aggregation of AUD. In this article, we add further empirical support to a very different mechanism of resemblance for AUD in relatives—contagious transmission. We also present results suggesting that this transmission expands beyond biological relatives to peers—friends and acquaintances. Furthermore, we show important moderators of this contagious transmission. Although the study of social contagion will not lead to insights into the biological causes of the genetic liability to AUD as molecular genetic studies potentially can, our findings are likely much more actionable in the short and medium term and worthy of further research and implementation trials.