Non-AIDS defining malignancies in the combination ART era: immunological and socio-behavioral risk factors

The Swiss HIV Cohort Study

The SHCS is an ongoing observational study that promotes clinical and public health research on HIV in Switzerland^17,18. The SHCS is a collaboration between various hospitals, clinics, laboratories, and private practices across the country. Prospective enrollment started in 1988, and involves PLWHIV aged 16 years or above. Detailed socio-demographic and behavioral parameters for each patient are collected at enrollment. Laboratory results such as CD4+ and CD8+ cell count, HIV-1 RNA viral load or indicators of exposure to hepatitis B or C virus are recorded at enrollment and during follow-up visits. Detailed information on ART and other treatments, and of AIDS defining events are also collected at regular intervals. While NADM are recorded in the study, their ascertainment is incomplete.

Linkage to cancer registries

Incident cancers in the SHCS were identified through linkage of SHCS records with the records of the Basel, Geneva, and Zurich cantonal cancer registries. At the time of the linkage, the data covered the period up to 2012 for Geneva and Zurich, and 2011 for Basel. Only SHCS participants registered in the areas covered by the three aforementioned cancer registries were included, representing about 65% of all participants. The linkage was done using privacy-preserving probabilistic record linkage^19,20. Briefly, we employed a three-party protocol, with the SHCS and cancer registries collecting confidential individual-level patient data and the Institute of Social and Preventive Medicine at University of Bern serving as the independent linkage center. The linkage center supplied purpose-written encryption software²¹ that was used at SHCS centers and cancer registries to irreversibly encrypt personal identifying data (names, date of birth) using Bloom filters and calculate the similarity coefficients, which were then used to identify matching records at ISPM.

Cancers included in analyses

We supplemented the cancer cases reported in the SHCS by the cases identified through data linkage. We analyzed NADM classified according to the 10^th revision of the International Classification of Diseases (ICD-10) that were diagnosed in about 40 or more patients: malignant neoplasm of anus and anal canal (C21), malignant neoplasm of liver and intrahepatic bile ducts (C22, excludes malignant neoplasm of other and unspecified parts of biliary tract), malignant neoplasm of trachea, bronchus and lung (C33–34), and prostate cancer (C61). We did not include Hodgkin lymphoma because most cases had been diagnosed in the early years of cART, when CD8 counts and viral load measurements were often missing.

Standardized incidence ratios

We calculated standardized incidence ratios (SIRs) for each identified NADM by comparing the observed number of cancers among people with HIV with the expected number of occurrences, based on age-, sex- and period-specific rates for the Swiss population provided by the National Institute for Cancer Epidemiology and Registration (indirect standardization)²². We produced 95% confidence intervals (CI) for the SIRs assuming a Poisson distribution²³.

Baseline and time-updated exposure variables

We assessed the effect of baseline and time-updated exposure variables in Cox regression models. Time-updated variables included CD4 cell count (cells/µL), CD8 cell count (cells/µL), CD4/CD8 ratio and HIV viral load (copies/mL). We updated the immunological and virological factors at the start of each month. Months with no CD4 cell count, CD8 cell count, or HIV viral load were interpolated by carrying forward or carrying back the closest measured value within 12 months, provided there was no change in cART and no diagnosis of cancer in the interval. Time-updated bacterial pneumonia was additionally considered for lung cancer: history of bacterial pneumonia changed from no to yes when pneumonia was diagnosed. Other variables included age, likely HIV transmission group (men having sex with men [MSM], intravenous drug users [IDU], other men, other women), current smoking at enrollment (yes, no), highest completed level of education (compulsory, secondary, tertiary), chronic hepatitis B and exposure to hepatitis C (yes or no, as a time-updated variable), and calendar period (time-updated; before 2002, 2002–2007, 2008 and later). Patients were defined as having chronic hepatitis B if they had previously been tested positive for HBs-antigen or HBV-DNA, and as being exposed to hepatitis C during follow-up if they had tested positive for HCV-RNA.

Statistical analyses

We used Cox regression models for time to event analyses, measuring time from enrollment to the diagnosis of a NADM, the last clinic visit in the study period, or death, whichever came first. Follow-up time in patients enrolled before 1996, the year cART was introduced, was left truncated at 31 December 1995. Patients diagnosed with a NADM before 1 January 1996 were excluded from the analysis for that specific NADM.

In a first step, we modeled the association of the time-updated immunological and virological exposure variables (CD4 cell count, CD8 cell count, CD4/CD8 ratio, and HIV viral load) with each of the four cancers separately, adjusting for the other variables. To reduce the risk of detecting associations that reflect reverse causality, we lagged these variables for 12 months or more, and captured cumulative exposure using simple moving averages (SMA). We entered the following representations of each variable: lagged by 12, 24 or 36 months; SMA over 12 months - lagged by 12 or 24 months; SMA over 24 months - lagged by 12 months (see Figure 1). In a next step we selected variables and representations of variables for inclusion into a final, malignancy-specific model. We selected one representation of each immunological and virological variable based on the goodness of fit (indexed by the Akaike information criteria [AIC]) and carried over variables significantly associated with cancer incidence (P <0.05) when modeled separately. The intermediary and final models were adjusted for the variables age, transmission group, smoking, education, chronic hepatitis B, exposure to hepatitis C, and calendar period. The time-updated variables for bacterial pneumonia (used only in the lung cancer analysis), hepatitis B, and hepatitis C were lagged by 12 months.

Figure 1. Schematic representation of simple moving averages (SMAs), lagged over different periods.

We assessed proportionality assumptions of the Cox model using the Schoenfeld residuals. Results are shown as hazard ratios (HR) with 95% confidence intervals (95% CI). We imputed missing values for smoking using multivariate imputation by chained equations and pooled results of five imputed dataset. We present the analyses with imputation since results were similar to complete-case analyses. All analyses were done in R Project for Statistical Computing (version 3.6.0) software²⁴ using packages survival (version 2.42), mice (version 3.30) and mitools (version 2.3). R code and explanatory documents are available as extended data²⁵.

Incidence and patient characteristics

A total of 563 incident NADM were identified in the two data sources (SHCS and cancer registries). The SHCS recorded 60.9% of all cases, and the registries 76.2%. For the cancers analyzed in this study, the completeness of recording of cases in the SHCS was 54.3% for anal cancer, 66.7% for liver cancer, 71.4% for lung cancer, and 72.7% for prostate cancer. The corresponding percentages for the cancer registries were 85.7%, 80.6%, 85.7% and 93.2%, respectively (Table 1).

Anal cancer was diagnosed in 70 patients over 71,592 person-years for an incidence rate of 9.78 per 10,000 person-years (95% confidence interval [CI] 7.62-12.35). Lung cancer was diagnosed in 49 patients over 71,888 person-years (6.82 per 10,000; 95% CI 5.04-9.01), prostate cancer in 44 men over 50,322 person-years (8.74 per 10,000; 95% CI 6.35-11.74), and liver cancer in 36 patients over 71,911 person-years (5.01 per 10,000; 95% CI 3.51-6.93).

Compared to the Swiss general population, anal, lung, and liver cancer all exhibited higher rates in the SHCS, whereas the rate of prostate cancer was similar (Figure 2). The difference was largest for anal cancer: the incidence was 76 times higher in people with HIV (SIR 76.1, 95% CI 60.2-96.2) than in the general Swiss population.

Figure 2. The number of observed and expected incident cancer cases and standardized incidence ratios for non-AIDS defining malignancies (NADM) diagnosed among participants of the Swiss HIV Cohort Study.

Characteristics of SHCS participants who developed cancer and participants who did not are shown in Table 2. Compared to those who remained free of cancer, HIV-positive individuals, who developed cancer were slightly older, more likely to be smokers, and at a more advanced clinical stage of HIV infection according to the Centers for Disease Control and Prevention classification²⁶. The distribution across HIV transmission groups also varied between cancers: the majority of patients who developed anal or prostate cancer were MSM, while those with lung or liver cancer were predominantly patients with a history of IDU.

Immunological and socio-behavioral risk factors

We selected one representation of each exposure variable based on goodness of fit models. The AIC values for the different models are shown in Table 3, the results of the separate models for each immunological and virological variable in Table 4 and the results of the mutually adjusted models in Table 5. For anal cancer, CD4 count, CD8 count and CD4/CD8 ratio were all associated with cancer risk in the separate models (Table 4). In the mutually adjusted model, a higher CD4 cell count (lagged 12 months) and a CD8 count below 1000 cells/µl (lagged by 24 months) continued to be significantly associated (P=0.002) with a reduced risk of anal cancer (Table 5). For example, in patients with a 12-month lagged CD4 count above 500 cells/µl, risk was reduced to about a fourth compared to individuals with a CD4 count below 200 cells/µl (hazard ratio [HR] 0.24; 95% CI 0.10-0.61). The risk of anal cancer was substantially higher in MSM (HR 6.86, 95% CI 1.64-28.6) than other men and higher in smokers (HR 2.07, 95% CI 1.07-3.98) compared to non-smokers.

For lung cancer and prostate cancer, Cox regression models were stratified by age to preserve proportional hazards. In the case of lung cancer, CD8 cell count (lagged by 36 months) was the only significant predictor (P=0.04) among the immunological and virological variables (Table 2): participants with CD8 cell counts equal to or above 1000 cells/µl had a greater risk of developing lung cancer, compared to those with counts below 1000 cells/µl (HR 1.89, 95% CI 1.03-3.48). A history of bacterial pneumonia was associated with lung cancer in the separate model (Table 4), but not after adjusting for CD8 cell count (Table 5). Smoking at enrollment was strongly associated with lung cancer (HR 6.50, 95% CI 1.36-31.2, Table 5). For prostate cancer, none of the immunological or virological were associated with a risk of cancer (Table 4), though there was some evidence of smoking being associated with higher rates (HR 1.92, 95% CI 0.96-3.83, Table 5).

For liver cancer, exposure to lower CD4 counts (SMA over 24 months, lagged by 12 months) and the CD4/CD8 ratio (lagged by 24 months) were associated with cancer incidence in the separate models (Table 4). In the final model (Table 5), only the CD4 cell count was predictive, with patients with average counts of 500 cells/µl or above at much lower risk than patients with counts below 200 cells/µL (HR 0.07, 95% CI 0.01-0.59). Chronic hepatitis B infection (HR 7.86, 95% CI 3.78-16.4) and hepatitis C infection (HR 4.77, 95% CI 1.80-12.7) were both strongly associated with liver cancer (Table 5).

Ethics statement

The SHCS was approved by the Ethics committees of the participating institutions. At enrolment, SHCS participants provide written informed consent for the use of biological and clinical data. For the present study, we obtained additional ethical approval for privacy-preserving probabilistic record linkage from the Ethics Committee of the Canton of Bern.

Source data

The National Institute for Cancer Epidemiology and Registration (NICER) is happy to share the anonymized cancer registry data with eligible partners. If you wish to receive specific data from the NICER database, please complete the respective form at https://www.nicer.org/en/data/request-data/.

Underlying data

The data analyzed in this study are sensitive, having been contributed by people living with HIV. Access to de-identified data is restricted to collaborative projects that have been submitted to and approved by the Scientific Board of the SHCS. Please contact the SHCS at http://www.shcs.ch/contact if you are interested in gaining access to the study data. Sharing or linkage of data may require additional ethics approval.

Extended data

Files with the R code used in the analysis are provided as extended data, along with an explanatory document, on Open Science Framework

Open Science Framework: Extended data for Non-AIDS defining malignancies in the combination ART era: immunological and socio-behavioral risk factors (Ruffieux et al. Faculty1000research 2019). https://doi.org/10.17605/OSF.IO/GY5VM²⁵

This project contains the following extended data:

Data are available under the terms of the Creative Commons Zero "No rights reserved" data waiver (CC0 1.0 Public domain dedication).