Results

Adolescent Samples

Descriptives

The descriptive statistics of the adolescent samples (i.e. KLS, GFT, and F&H) are described in Table 1. Sex distribution was relatively balanced across the samples: Females constituted 52.7% of the KLS sample, 54.3% of GFT, and 46.7% of F&H. The mean age at T1 was between 12.6 and 13.8 years across samples. The KLS sample consisted of only students from higher general secondary/pre-university education. The GFT and F&H samples also consisted mostly of adolescents in higher educational levels. The mean loneliness is relatively low, as scores vary between 0.35 and 0.56 across all samples on a 0- to 3-point scale. Regarding smoking, it should be noted that 4.1%, 4.2%, and 35.7% smoked at the baseline of the KLS, GFT, and F&H study, respectively, and were excluded from the analyses.

 | KLS (N = 600) | GFT (N = 551) | F&H (N = 261)
Gender
Male | 47.3% | 45.7% | 53.3%
Female | 52.7% | 54.3% | 46.7%
Age T1 | 12.59 (0.40) | 12.81 (0.61) | 13.80 (0.39)
Educational type T1
Pre-vocational | n.a. | 32.5% | 26.4%
Higher general secondary | n.a. | 9.2% | 41.3%
Pre-university | n.a. | 58.3% | 29.9%
Loneliness
T1 | 0.47 (0.49) | 0.38 (0.45) | 0.53 (0.53
T2 | 0.43 (0.50) | 0.35 (0.50) | 0.54 (0.53)
T3 | 0.38 (0.45) | 0.52 (0.54) | 0.55 (0.57)
T4 | 0.44 (0.52) | 0.45 (0.50) | 0.56 (0.58)
T5 | n.a. | 0.55 (0.48) | 0.51 (0.58)

Discrete-Time Survival Analyses: How Is Loneliness Longitudinally Associated with Smoking Initiation?

The life table of smoking for the three adolescent samples is shown in Supplementary Material VI. As the hazard ratios were not significant in any of the three adolescent samples, they were not further interpreted (the exact values can be found in this section and in Supplemental Material VI).

Sample 1: KLS

The results from the main analysis suggest that loneliness did not predict smoking initiation (b = − 0.17, SE = 0.11, p = 0.137, hazard ratio (HR) = 0.85); see Table 2. This association did not differ between time intervals; see Supplementary Material VII. The likelihood to smoke was found to be statistically significantly higher during period 3 (i.e. from grades 9 to 10) than the mean likelihood to start smoking across all time periods; the likelihood to start smoking during periods 1 (i.e. from grades 7 to 8) and 2 (i.e. from grades 8 to 9) did not differ from the mean. Sex and age were not significantly associated with smoking initiation.

 | KLS (N = 600) | GFT (N = 551)a | F&H (N = 261)
b | SE | p | HR | 95% CI | b | SE | p | HR | 95% CI | b | SE | p | HR | 95% CI
Intercept | − 2.49 | 0.10 | 0.001 | 0.08 | [0.07, 0.10] | − 3.06 | 0.28 | 0.001 | 0.04 | [0.03, 0.08] | − 1.96 | 0.11 | 0.001 | 0.14 | [0.11, 0.17]
Time |  |  |  |  |  |  |  |  |  |  |  |  |  |  | 
Period 2 | − 0.07 | 0.13 | 0.581 | 0.93 | [0.71, 1.21] | 0.71 | 0.32 | 0.001 | 2.03 | [1.08, 3.80] | − 0.04 | 0.19 | 0.797 | 0.95 | [0.66, 1.37]
Period 3 | 0.47 | 0.14 | < 0.001 | 1.61 | [1.22, 2.12] | − 1.01 | 0.52 | 0.052 | 0.36 | [0.13, 1.01] | 0.09 | 0.20 | 0.649 | 1.10 | [0.74, 1.62]
Period 4 | n.a. | n.a. | n.a. | n.a. | n.a. | 0.12 | 0.52 | 0.823 | 1.12 | [0.41, 3.11] | 0.05 | 0.22 | 0.819 | 1.05 | [0.68, 1.62]
Period 5 | n.a. | n.a. | n.a. | n.a. | n.a. | − 0.59 | 0.82 | .474 | 0.55 | [0.11, 2.78] | − 0.38 | 0.27 | 0.179 | 0.69 | [0.40, 1.19]
Gender |  |  |  |  |  |  |  |  |  |  |  |  |  |  | 
Female | 0.10 | 0.10 | .339 | 1.10 | [0.91, 1.34] | 0.06 | 0.11 | 0.596 | 1.06 | [0.85, 1.32] | − 0.07 | 0.09 | 0.439 | 0.93 | [0.77, 1.12]
Age | 0.11 | 0.10 | 0.252 | 1.12 | [0.92, 1.35] | 0.42 | 0.11 | < 0.001 | 1.50 | [1.23, 1.83] | 0.04 | 0.09 | 0.667 | 1.04 | [0.87, 1.24]
Educational type |  |  |  |  |  |  |  |  |  |  |  |  |  |  | 
Higher general secondary | n.a. | n.a. | n.a. | n.a. | n.a. | − 0.12 | 0.26 | 0.640 | 0.88 | [0.53, 1.48] | − 0.06 | 0.14 | .668 | 0.94 | [0.72, 1.24]
Pre−university | n.a. | n.a. | n.a. | n.a. | n.a. | −0.46 | 0.18 | 0.011 | 0.63 | [0.44, 0.90] | − 0.09 | 0.15 | .534 | 0.91 | [0.68, 1.22]
Loneliness | − 0.17 | 0.11 | 0.137 | 0.85 | [0.68, 1.05] | − 0.21 | 0.13 | 0.116 | 0.81 | [0.63, 1.05] | − 0.18 | 0.10 | 0.085 | 0.84 | [0.69, 1.02]

Sample 2: GFT

The results from the main analysis suggest that loneliness did not predict smoking initiation (b = − 0.21, SE = 0.13, p = 0.116, HR = 0.81); see Table 2. This association did not differ between time intervals; see supplemental Table VII. The likelihood of initiating smoking was higher in the second period, compared to the mean of all time periods, but this can be explained by the fact that this period had the longest time interval, with time periods being shorter towards the end of the study. Sex was not significantly associated with smoking initiation. Age was significantly associated with smoking initiation, with older adolescents being more likely to initiate smoking (b = 0.42, SE = 0.11, p = < 0.001, HR = 1.50). Furthermore, pre-university students were less likely to initiate smoking compared to the average likelihood across all educational types (b = − 0.46, SE = 0.18, p = 0.011, HR = 0.63).

Sample 3: F&H

The results from the main analysis suggest that loneliness did not predict smoking initiation (b = − 0.18, SE = 0.10, p = 0.085, HR = 0.84); see Table 2. This association did not differ between time intervals; see Supplementary Material VII. The likelihood of initiating smoking did not significantly differ across the time intervals. Other factors, such as sex, age, and education, were not significantly associated with smoking initiation.

Overall, these results show that higher levels of loneliness do not predict a higher or lower chance of starting to smoke.

University Student Samples

Descriptives

In both samples, most participants were female (HSL, 71%; al-RISCO, 66%) (see Table 3). The mean age was 21.73 (HSL) and 22.55 (al-RISCO) years old. The HSL sample consists of only research university students (‘WO’), and the al-RISCO sample consists of 84% research university (‘WO’) and 16% applied university students (‘HBO’). The mean score for loneliness was 1.2 (HSL, range 0–4) and 1.7 (al-RISCO, range 1–3). Across both samples, most participants were non-smokers. In the HSL sample, at T1 to T3, occasional and regular smokers comprised 4.3% and 5.3%, 4.9% and 4.6%, and 3.6% and 3.5% of participants, respectively. For the al-RISCO sample, these figures were 5.9% and 7.9% at T1, 6.2% and 8.4% at T2, and 6.7% and 6.2% at T3, respectively. Smoking quantity (measured in smokers only) in the HSL sample varied between 18.1 and 21.9 cigarettes per week between T1 and T3.

 | HSL (N = 2,636) | al-RISCO (N = 404)
Gender T1
Male | 576 (23.3%) | 137 (33.9%)
Female | 1,859 (75.2%) | 265 (65.6%)
Other | 38 (1.5%) | 2 (0.5%)
Age T1 | 21.73 (2.44); n=2636 | 22.55 (1.86); n = 404
Education type T1a
Applied University | n.a. | 340 (84.2%)
Research University | 2,636 (100%) | 64 (15.8%)
Loneliness T1b | 1.19 (0.73); n = 2471 | 1.72 (0.51); n = 404
Smoking status T1 |  | 
Non-smoker | 1929 (88.5%) | 348 (86.1%)
Occasional smoker | 113 (5.2%) | 24 (5.9%)
Regular smoker | 139 (6.4%) | 32 (7.9%)
Smoking status T2
Non-smoker | 1751 (87.5%) | 318 (84.4%)
Occasional smoker | 128 (6.4%) | 25 (6.6%)
Regular smoker | 122 (6.1%) | 34 (9.0%)
Smoking status T3
Non-smoker | 1327 (87.7%) | 274 (84.1%)
Occasional smoker | 96 (6.3%) | 27 (8.3%)
Regular smoker | 91 (6.0%) | 25 (7.7%)
Smoking quantity (cig/week) in occasional and regular smokers (min = 0; max = 210)
T1 | 21.86 (33.78);n = 252 | n.a.
T2 | 18.12 (30.00); n = 250 | n.a.
T3 | 19.00 (32.01); n = 186 | n.a.

SEM Models: Does Baseline Loneliness Predict Smoking Status at Follow-Up?

Sample 4: HSL

The results from the main analysis suggest that baseline loneliness does not predict smoking status at T2 (β = 0.05, SE = 0.06, p = 0.067) but does significantly predict smoking status at T3 (β = 0.07, SE = 0.06, p = 0.035); see Table 4. Smoking status at baseline was consistently positively associated with smoking status at T2 (β = 0.67, SE = 0.05, p < 0.001) and T3 (β = 0.60, SE = 0.06, p < 0.001). Participants who identified as gender other were less likely to smoke frequently at T2 compared to males/females (β = − 0.06, SE = 0.30, p = 0.032). No additional effects were found. In the sensitivity analyses with complete cases/smoking quantity (see Supplementary Material IX.1 and IX.2), no significant effects of baseline loneliness on smoking status/quantity at T3 were found. However, similar trends were observed. The effect of baseline smoking status on smoking status at follow-up remained consistent across the sensitivity analyses. The exploratory analyses (nominal regression) showed that baseline loneliness was specifically associated with being a regular smoker at T3 (OR = 1.62, p = 0.033), not an occasional smoker (OR = 1.02, p = 0.935, see Supplemental Material VIII). Attrition analyses (Supplemental Material X) revealed that younger participants, participants who felt less lonely, and occasional smokers at baseline were more likely to miss data on smoking status at T2. Additionally, occasional and regular smokers at baseline were more likely to miss data on smoking status at T3.

T1 variables | HSL (N = 2,636) | al-RISCO (N = 404)
Smoking status T2 | Smoking status T3 | Smoking status T2 | Smoking status T3
b | β | SE | p | b | β | SE | p | b | β | SE | p | b | β | SE | p
Gender a | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | 0.17 | 0.06 | 0.26 | 0.499 | 0.09 | 0.03 | 0.22 | 0.689
Male vs. female + other d | 0.12 | 0.04 | 0.10 | 0.216 | 0.12 | 0.04 | 0.10 | 0.207 | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | n.a.
Other vs. male + female d | − 0.64 | − 0.06 | 0.30 | 0.032 | 0.24 | 0.02 | 0.19 | 0.191 | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | n.a.
Education level b | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | − 0.16 | − 0.04 | 0.25 | 0.528 | 0.27 | 0.08 | 0.27 | − 0.308
Age | − 0.01 | − 0.01 | 0.02 | 0.729 | − 0.02 | − 0.04 | 0.02 | 0.282 | − 0.01 | − 0.02 | 0.05 | 0.767 | 0.00 | 0.00 | 0.05 | − 0.999
Smoking status c | 1.82 | 0.67 | 0.05 | < 0.001 | 1.50 | 0.60 | 0.06 | < 0.001 | 1.51 | 0.65 | 0.12 | < 0.001 | 1.34 | 0.61 | 0.12 | < 0.001
Loneliness | 0.10 | 0.05 | 0.06 | 0.67 | 0.12 | 0.07 | 0.06 | 0.035 | − 0.06 | − 0.02 | 0.24 | 0.812 | 0.17 | 0.07 | 0.20 | 0.400

Sample 5: al-RISCO Lifestyle Project

Within the al-RISCO sample, baseline loneliness did not predict smoking status at T2 (β = − 0.06, SE = − 0.02, p = 0.812) or T3 (β = 0.17, SE = 0.20, p = 0.400), as shown in Table 4. Baseline smoking consistently predicted smoking status at T2 (β = 0.65, SE = 0.12, p < 0.001), and T3 (β = 0.61, SE = 0.12, p < 0.001). No additional effects were found. Sensitivity analyses with complete cases confirmed these results (see Supplementary Material IXI.1). The exploratory analyses also showed no significant associations between baseline loneliness and being an occasional or regular smoker at T2 or T3 (see Supplementary Material VIII). It should be noted that the OR between baseline loneliness and being a regular smoker at T3 was 2.48 (in line with the significant result in the HSL sample), but the result was not statistically significant (p = 0.073). Attrition analyses (Supplemental Material X) revealed that participants who felt less lonely and frequent smokers at baseline were more likely to miss data on smoking status at T3.

Overall, we did not find strong evidence for a relationship between loneliness and smoking initiation in the adolescent samples or between loneliness and smoking status in the young adult samples. The results of the discrete-time survival analyses in all three adolescent samples suggested that baseline loneliness did not predict smoking initiation. We found evidence in one sample that education level predicted smoking initiation, such that adolescents from higher education levels were less likely to initiate smoking. Further, structural equation models in the two young adult samples did not provide support for the predictive role of loneliness on smoking status. However, we found that baseline loneliness predicted smoking status 18 months later in sample 4. Further analyses revealed that baseline loneliness predicted being a regular smoker at follow-up, but not an occasional smoker at follow-up.