2.1 Study population

A consecutive cohort of patients who underwent LT from January 2010 to December 2015 and received tacrolimus-based immunosuppression at 16 transplant institutions composed the eligible cohort. Exclusion criteria were as follows: age <18 years old, re-transplantation, combined organ transplantation, donor with incidental malignancy, human immunodeficiency virus infection, and death within the first year after LT. Patients were followed until December, 2019. Patients developing malignancy after LT, either de novo or recurred, were identified in the eligible cohort and formed the group of cases. Controls were selected among patients who, after an identical follow-up period as cases, had not developed malignancy. The quasi-cohort approach ensured that person-years were comparable between the groups of cases and controls and allowed for computing hazard ratios (HRs).¹⁰ Cases and controls were matched according to a center-specific propensity score based on potential clinical risks factors of cancer (age, gender, smoking habit, alcoholic liver disease, hepatitis B status, hepatitis C status, and hepatocellular carcinoma [HCC] before LT). For each case, one control was selected within the same LT center (ratio 1:1) according to the nearest neighbor approach. This strategy aimed for a homogeneous distribution of potential clinical risk factors of cancer other than immunosuppression between cases and controls.

2.2 Evaluation of immunosuppression and outcomes

The exposure to immunosuppressive drugs within the first 12 months after LT was recorded in selected cases and controls. All available determinations of trough concentrations of tacrolimus were retrieved for each patient. Cumulative exposure to tacrolimus (CET) was calculated by the area under curve of trough concentrations, which has been previously associated with renal impairment after LT.¹¹ Briefly, all measurements of trough concentrations of tacrolimus for a certain patient were plotted in a time-dependent graph. Each determination was joined with the next delineating a curve. Finally, the area under curve of trough concentrations was calculated by the Wagner–Nelson equation. An online calculator of CET is available at Web CET (imibic.org). Categorization of CET was performed according to previously defined thresholds¹¹ in aggressive minimization, minimization, conventional exposure, and high exposure. Thresholds and approximate equivalence between target trough concentrations and CET are shown in Figure 1. The combination of tacrolimus with other immunosuppressive drugs was recorded, including the length of exposure within the first year. All patients received tapering corticosteroids, which were withdrawn between the third and the sixth month after LT, except for patients with autoimmune liver disease, in whom these were kept at the lowest dose tolerated in the long term. The use of induction therapy and the indication of boluses of corticosteroids to treat rejection episodes were registered.

The main outcome of the study was the development of any type of cancer after LT, including recurrence of HCC given its strong prognostic impact. Secondary outcomes were de novo malignancy (excluding HCC recurrence) and internal neoplasms (excluding non-melanoma skin cancer). For those patients experiencing more than one malignancy, only the first tumor diagnosed was computed. Tumor stage at diagnosis, oncologic therapy, and mortality were recorded.

2.3 Sample size calculation

Under these premises, the sample size required would be n = 572, comprising 286 patients with post-LT malignancy and 286 matched controls.

2.4 Statistical analysis

Continuous variables were presented as mean and standard deviation, excepting for those with asymmetrical distribution in which median and interquartile range (IQR) were used. Categorical variables were displayed as percentages in frequency tables. Appropriate hypothesis contrast tests were used according to the type of variables involved in the analysis. Kaplan–Meier curves and log-rank test were used for survival analysis. Risk factors of post-LT malignancy were investigated using univariate and multivariate Cox regression in a two-step process. First, a model composed of demographic and clinical features was computed in the whole eligible cohort. Variables with a p < .40 in the univariate analysis entered the initial multivariate model. Non-significant co-variates were removed in a backward stepwise process. All potential interactions between covariates were tested and kept in the final model if they reached statistical significance. A second model evaluating immunosuppression-related variables on post-LT malignancy was investigated in matched cases and controls (study cohort). In this analysis, clinical predictors contained in the propensity score and immunosuppression-related variables were kept in the final model to control for possible residual bias irrespective of their p value. Clinically meaningful interactions were kept if their removal resulted in a modification of the beta coefficient of the involved variables >20%. The individual effect of CET in each type of post-LT malignancy was evaluated by univariate logistic regression due to the limited incidence of some tumors. All analyses were performed by SPSS 27.0 (IBM) and R v3.6.3 (RStudio Inc.). Every hypothesis tested was two-tailed and considered significant if p < .05.

3.1 Risk factors of cancer in the eligible cohort

The eligible cohort comprised 2495 patients (54.91 ± 9.08 years old, 22.5% women). Major etiologies of liver disease were alcoholic liver disease (52.2%), chronic hepatitis C (35.9%), and chronic hepatitis B (6.8%). The indication for LT was HCC in 1023 patients (41%). Nine hundred forty-one patients (37.7%) were active smokers at inclusion in the waiting list for LT and 1244 patients (61.1%) were past smokers or non-smokers. Data on smoking habit could not be reliably obtained in 30 patients (1.2%). After a median follow-up of 65 months after LT (IQR 49–87), including 13 922 person/years, 491 patients (19.7%) developed cancer, yielding an incidence rate of 3.5/100 person-years for all type of tumors, and 2.6/100 persons-years for de novo malignancy. The incidence of cancer remained unchanged over the study period (Figure S1). Types of malignancy, in order of frequency, were as follows: HCC recurrence (n = 118; 24%), non-melanoma skin cancer (n = 79; 16.1%), lung cancer (n = 67; 13.7%), head and neck tumors (n = 62; 12.6%), prostate cancer (n = 26; 5.3%), lymphoproliferative disorders (n = 24; 4.9%), urinary tract cancer (n = 22; 4.5%), esophageal/gastric cancer (n = 20; 4.1%), colorectal cancer (n = 17; 3.5%), pancreatic cancer (n = 10; 2%), breast cancer (n = 5; 1%), gynecological cancer (n = 4; 0.8%), Kaposi's sarcoma (n = 4; 0.8%), melanoma (n = 4; 0.8%), and sarcomas or metastases of unknown origin (n = 29; 5.9%).

The univariate and multivariate Cox regression analyses (Table 1) identified the following clinical-independent predictors of posttransplant malignancy after controlling for baseline HCC: age at transplantation (HR = 1.06 [95% CI 1.05–1.07]; p < .001), male sex (HR = 1.50 [95% CI 1.14–1.99]; p = .004), active smoking habit at waitlist inclusion (HR = 1.96 [95% CI 1.45–2.66]; p < .001), and alcoholic liver disease (HR = 1.53 [95% CI 1.19–1.97]; p = .001). There was a significant interaction between alcoholic liver disease and hepatitis C, which was included in the final model. Age at LT increased the risk of cancer and this effect was particularly evident for patients over 50 years old (Figure S2). The accumulation of clinical predictors resulted in a progressive increase of the 5-year cumulative incidence of cancer after LT (Figure 2): 0% for patients without risk factors, 12.5% for patients with 1–2 risk factors, and 22.8% for patients with 3–4 risk factors.

3.2 Prognostic impact of cancer in LT patients

The stage of cancer at diagnosis was locally advanced in 22.8% of patients and metastatic in 34.8% of patients. Regarding the initial cancer therapy, 47% of patients received surgical resection with curative intention, 10% of patients underwent local ablation or palliative surgery, 26.1% of patients received chemotherapy, and 16.9% of patients were directly transferred for palliative care. Among 454 deaths registered in the whole cohort after a 13 922 person/years follow-up, 198 (43.6%) were attributed to post-LT malignancy. Lethality of cancer was 40.3% and the median interval between diagnosis of cancer and death was 8.7 months (IQR 3–20.1). Tumor stage at diagnosis and first-intention therapy were main determinants of survival (Figure 3). Regarding tumor stage at diagnosis, 12-month overall survival rates were 92.6% for patients with resectable tumors, 58.3% in patients with locally advanced tumors, and 43.5% for patients with metastatic disease (p < .001). Patients who underwent surgery with curative intention had 90.6% overall survival rates at 12 months, which were higher than that observed in patients receiving local ablative therapies (56.9%), systemic therapy (63.6%) or palliative care (12.5%) (p < .001). Median survival according to the type of tumor is shown in Table 2. Tumors with the worst prognosis were pancreatic cancer, gynecological tumors, melanoma, lung cancer, and HCC recurrence, with median overall survival shorter than 10 months.

3.3 Immunosuppression-related risk factors of cancer

A flowchart depicting the selection of cases and controls to study the relationship between immunosuppression and post-LT malignancy is presented in Figure 4. The analysis of immunosuppression-related risk factors of cancer comprised 850 patients (425 pairs of cases and controls). Baseline features including immunosuppressive therapy of cases and controls are shown in Table 3. Cases and controls were comparable in terms of age, gender, smoking habit, etiology of liver disease, and the presence of HCC before LT. Indeed, the propensity score was identical in both groups (p = .957). The predominant immunosuppression protocol in each patient within the first 12 months after LT was similar in cases and controls (p = .509): tacrolimus alone in 32% of cases versus 30.4% of controls, tacrolimus in combination with mycophenolate in 56.5% of cases versus 60% of controls, and tacrolimus in combination with mTOR inhibitors in 11.5% of cases versus 9.6% of controls. None of the patients received azathioprine within the first 12 months after LT. Individual immunosuppressive drugs and their length of administration were also distributed homogeneously in patients with post-LT cancer and in matched controls (Table 3). The use of boluses of steroids to treat rejection episodes was also identical between groups (p = .911). Patients with post-LT malignancy had increased CET as compared with matched controls, both within the first 3 months (754 ng∙day/ml [IQR 614–920] vs. 695 ng∙day/ml [IQR 580–862]; p = .002) and within the first 12 months (2820 ng∙day/ml [IQR 2413–3334] vs. 2699 ng∙day/ml [IQR 2284–3160]; p = .009). High exposure to tacrolimus according to CET occurred in 36.3% of cases compared to 27.5% of controls at 3 months (p = .005) and in 36.5% of cases compared to 31.3% of controls at 12 months (p = .057). The smoothing splines showed that the increased risk of cancer occurred in patients within the high-exposure strata according to CET, and particularly when this parameter was evaluated at 3 months (Figure 5).

In the multivariate Cox regression analysis, an increased CET was the only independent predictor of post-LT malignancy after controlling for clinical features, baseline HCC, and concomitant immunosuppressive drugs (Table 4). Indeed, the HR of post-LT malignancy for a 20% increase of CET within the first 3 months was 1.11 (95% CI 1.05–1.19; p = .001). Similarly, the HR of post-LT malignancy for a 20% increase of CET within the first 12 months was 1.10 (95% CI 1.03–1.17; p = .004). These results were consistent when considering exclusively de novo malignancy (CET at 3 months adjusted HR = 1.09 [95% CI 1.01–1.18]; p = .020 and CET at 12 months adjusted HR = 1.10 [95% CI 1.02–1.19]; p = .016), and internal neoplasms (CET at 3 months adjusted HR = 1.11 [95% CI 1.04–1.19]; p = .001 and CET at 12 months adjusted HR = 1.09 [95% CI 1.01–1.21]; p = .017). Noteworthy, the use of mTOR inhibitors was associated with increased risk of post-LT malignancy univariately, but it lost statistical significance in the multivariate analysis after controlling the interaction between prescription of mTOR inhibitors and baseline HCC.

The effect of CET was evaluated for each type of cancer univariately (Figure S3). An increase of CET by 20% within the first 3 months was associated with significantly higher risk of colorectal cancer (OR = 2.37 [95% CI 1.20–4.69]; p = .013), lung cancer (OR = 1.31[95% CI 1.03–1.65]; p = .024), and HCC recurrence (OR = 1.26 [95% CI 1.07–1.48]; p = .006). On the other hand, an increase of CET by 20% within the first 12 months, was associated with higher risk of skin cancer (OR = 1.34 [95% CI 1.03–1.74]; p = .030) and colorectal cancer (OR = 2.06 [95% CI 1.05–4.02]; p = .034). A subgroup analysis of patients with HCC recurrence is presented in Supplementary Material.

After diagnosis of cancer, tacrolimus dose was reduced by >25% in 93 patients (21.9%), either in combination with mycophenolate (n = 40, 9.4%) or in combination with mTOR inhibitors (n = 53; 12.5%). Tacrolimus was completely withdrawn in 92 patients (21.6%), either in combination with mycophenolate (n = 9, 2.1%) or in combination with mTOR inhibitors (n = 83, 19.5%). The immunosuppression regimen remained unchanged after the diagnosis of cancer in 56.5% of patients. Modifications in the immunosuppression protocol had no influence on overall survival, neither in the entire cohort (Log Rank p = .347) nor in the HCC recurrence subgroup (Log Rank p = .193).