The predictive role of NLR and PLR for solid non-AIDS defining cancer incidence in HIV-infected subjects: a MASTER cohort study

We conducted a retrospective cohort study of HIV-infected patients of the MASTER cohort
in follow-up from January 2000 to December 2012, either ART naive or experienced.
Inclusion criteria were: age of 18 years and over, and no solid NADC diagnosis before
baseline. The baseline was 1st January 2001 for subjects enrolled in MASTER cohort
before that date, and the date of enrollment in the cohort for patients who entered
in the cohort after that date. The characteristics of the MASTER cohort and the procedures
of cancer data collection have been described elsewhere 17]. The following data were retrieved from the electronic database: gender, age, country
of origin, HIV exposure group, date of enrolment in the cohort, viral hepatitis C
or B co-infection, ART, AIDS event and cancer occurrence. Moreover, the following
parameters, measured within 6 months from the diagnosis of cancer, were retrieved:
HIV-RNA, CD4 cell count, CD8 cell count, neutrophil, lymphocyte and platelet count.

The inflammatory factors evaluated were NLR and PLR, considered as continuous and
dichotomized according to their median.

The primary outcome was the incidence of NADCs, which were coded according to the
international classification of diseases (ICD), 9th and 10th revisions 18]. Hematological cancers (ICD-10 code, from 81 to 96) and solid ADCs (Kaposi sarcoma
and invasive cervical carcinoma, ICD-10 code 46 and 53, respectively) were excluded
from the analysis.

The study was conducted in accordance with the guidelines of the Declaration of Helsinki
and the principles of Good Clinical Practice. The study protocol was approved by the
local ethics committees. Informed consent was obtained by all patients enrolled.

Statistical analysis

Observation time was calculated from study inclusion until cancer occurence, death,
last follow-up visit or 31st December 2012.

The differences in demographic, clinical and pathological features between losses
to follow-up and non-losses to follow-up were tested using common statistical methods
for median and proportion comparisons.

The age- and gender-adjusted NADC incidence rates were calculated dividing the number
of observed cases by the corresponding person-years at risk, using the direct method
of standardization, truncated at 65 years-old, with the European population as the
standard, according to calendar period 19]. All the rates were expressed per 1000 person-years.

The associations of NLR, PLR with cancer incidence were evaluated by univariate and
multivariate analysis using both time independent and time dependent Cox proportional
hazard models, which provided estimates of hazard ratios (HRs), their 95 % confidence
intervals (95 % CIs) and p-values. In time dependent regression models, the study
period was divided into intervals of 1-year duration. Gender, age at enrolment, intravenous
drug use and hepatitis C or B virus co-infection were included in the model as fixed
covariates. NLR, PLR and CD4 cell count were included as time-dependent covariates.

To evaluate whether the associations of NLR and PLR with risk of cancer were not linear,
we also fitted time dependent Cox models with a cubic-spline for NLR and PLR, respectively.
We used the Akaike’s information criterion to assess fitting of the models with linear
and non-linear terms and to choose the number of spline knots.

Finally, we conducted sensitivity analyses. Particularly, we assessed the relations
between NRL and PLR and cancer incidence i) using competing risk regression models
with death from all causes as a competing event, ii) applying inverse probability
weighted methods in order to adjust for selection bias due to losses to follow-up,
iii) applying inverse probability weighted methods in order to adjust for selection
bias due to missing values, iv) limiting the analysis to subjects enrolled from 1st
January 2000, v) excluding the first year of follow-up; subjects who had diagnosis
of NADC within 1 year were not included in such analysis, while the follow-up started
1 year later for everyone else, vi) using 1:1 nested case—control design matched on
age, gender, date of enrolment and late presentation. The proportional hazards assumption
was investigated for each covariate and globally by analyzing Schoenfeld residuals.
We first produced the graphical plots and then carried out formal statistical tests
of their independence over the rank transformation of time, but no departures from
this assumption were found.

For statistical tests, P values lower than 0.05 were considered significant in two-tailed
tests. All the computations were carried out using the Stata program for personal
computer, version 12.0 (StataCorp, College Station, TX, USA).