Clinical spectrum and prognostic impact of cancer in critically ill patients with HIV: a multicentre cohort study

Background Both AIDS-defining and non-AIDS-defining cancers (ADC/NADC) predispose people living with HIV (PLHIV) to critical illnesses. The objective of this multicentre study was to investigate the prognostic impact of ADC and NADC in PLHIV admitted to the intensive care unit (ICU). Methods All PLHIV admitted over the 2015–2020 period in 12 university-affiliated ICUs in France were included in the study cohort. The effect of ADC and NADC on in-hospital mortality (primary study endpoint) was measured through logistic regression with augmented backward elimination of potential independent variables. The association between ADC/NADC and treatment limitation decision (TLD) during the ICU stay (secondary study endpoint) was analysed. One-year mortality in patients discharged alive from the index hospital admission (exploratory study endpoint) was compared between those with ADC, NADC or no cancer. Results Amongst the 939 included PLHIV (median age, 52 [43–59] years; combination antiretroviral therapy, 74.4%), 97 (10.3%) and 106 (11.3%) presented with an active NADC (mostly lung and intestinal neoplasms) and an active ADC (predominantly AIDS-defining non-Hodgkin lymphoma), respectively. Inaugural admissions were common. Bacterial sepsis and non-infectious neoplasm-related complications accounted for most of admissions in these subgroups. Hospital mortality was 12.4% in patients without cancer, 30.2% in ADC patients and 45.4% in NADC patients (P < 0.0001). NADC (adjusted odds ratio [aOR], 7.00; 95% confidence interval [CI], 4.07–12.05) and ADC (aOR, 3.11; 95% CI 1.76–5.51) were independently associated with in-hospital death after adjustment on severity and frailty markers. The prevalence of TLD was 8.0% in patients without cancer, 17.9% in ADC patients and 33.0% in NADC patients (P < 0.0001)—organ failures and non-neoplastic comorbidities were less often considered in patients with cancer. One-year mortality in survivors of the index hospital admission was 7.8% in patients without cancer, 17.0% in ADC patients and 33.3% in NADC patients (P < 0.0001). Conclusions NADC and ADC are equally prevalent, stand as a leading argument for TLD, and strongly predict in-hospital death in the current population of PLHIV requiring ICU admission. Supplementary Information The online version contains supplementary material available at 10.1186/s13613-023-01171-4.


Background
The epidemiology of cancer in people living with HIV (PLHIV) has markedly evolved over the past decades owing to augmented access to combination antiretroviral therapy (cART).The incidence and attributable mortality of AIDS-defining cancers (ADC) (namely, high-grade B-cell non-Hodgkin lymphoma [NHL], Kaposi sarcoma, and carcinoma of the cervix) have globally declined though PLHIV remain at higher hazard for these neoplasms than seronegative individuals, especially in case of undiagnosed HIV infection or social issues that jeopardise sustainable cART adherence [1][2][3][4].Conversely, extended life expectancy, lifestyle risk factors, certain viral co-infections and chronic systemic inflammation ensuing from silent HIV replication may all contribute to the growing prevalence of non-AIDS-defining cancers (NADC) which nowadays prevail and stand as a leading cause of death in the overall population of cART-treated PLHIV [5][6][7][8].
Both ADC and NADC predispose PLHIV to life-threatening events resulting from sepsis, metabolic disorders, mechanical complications, or chemotherapy-related toxicity [9,10].Yet, granular data are lacking to appraise the current impact of cancers on the clinical features and outcomes of critical illnesses in these patients [11].Next, palliative approaches now constitute an essential component of care for PLHIV ageing with severe comorbidities and altered performance status [12]; however, treatment limitation decisions (TLD) remain under-investigated in those admitted to the intensive care unit (ICU) and how cancers affect TLD in this setting is unknown [11].These elements appear pivotal to improve the collaborative management of PLHIV by intensivists, infectious diseases specialists and oncologists, define the appropriate level of care, and help rationalising the utilisation of ICU resources in this changing patient population.
The primary objective of this multicentre cohort study was to measure the effect of NADC and ADC on hospital mortality in a contemporary population of PLHIV admitted to the ICU.The secondary objective was to appraise the impact of cancer on TLD in these patients.The association between cancer status and one-year mortality in patients alive at hospital discharge was investigated as an exploratory objective.

Study design and patient inclusion
This study was conducted in 12 medical ICUs located in university-affiliated hospitals in France and contributing to the GRRROH (Groupe de Recherche Respiratoire en Réanimation Onco-Hématologique, www.grrroh.fr) and CRICS-TRIGGERSEP (Clinical Research in Intensive Care and Sepsis-Trial Group for Global Evaluation and Research in Sepsis, www.crics-trigg ersep.org) networks for research in intensive care (see the Additional file 1 for details on centre recruitment and characteristics).All adult PLHIV admitted to these ICUs over a 5.5 year period (from January 2015 to June 2020) were retrospectively identified by local investigators through hospital coding databases using the International Classification of Diseases-10th revision items related to HIV infection (that is B20-24, Z21, and R75) and included in the study cohort provided that there was no missing information regarding cART use at ICU admission and the vital status at hospital discharge.Only the first ICU stay was analysed in patients with multiple admissions over the inclusion period.The study protocol was approved by the ethical committee of the French Intensive Care Society on August 3, 2020 (report n°CE-SRLF-20-70); the requirement for informed consent was waived owing to the observational design.Results are reported according to the STROBE guidelines [13].

Data collection and definitions
The patient medical records from the ICU and the downstream unit were anonymised by local investigators and centralised before data extraction with a standardised form.Vital status at last follow-up visit in the participating centre was also collected, when available.
cART was defined according to the current International Antiviral Society guidelines [14].When available, CD4 cell count and HIV viral load were collected within the 6 months preceding ICU admission for patients with previously known HIV infection and at ICU admission for those with newly diagnosed seropositivity.Missing values are indicated in Additional file 1: Table S1.
AIDS-defining conditions were defined according to the Centers for Diseases Control and Prevention classification, with primary brain lymphoma, Burkitt lymphoma, other high-grade B-cell NHL, Kaposi sarcoma and cervix cancer being identified as ADC [15].Other solid and haematological neoplasms were classified as NADC.NADC and ADC were considered active in patients with either inaugural admission, response to first-line therapy, relapse, or refractory disease.Neoplasms with proven remission after first-line therapy were classified as inactive.
The primary diagnoses of the ICU stay were ventilated as cancer-related non-infectious complications, noninfectious complications of chronic conditions other than cancer, bacterial sepsis, non-AIDS-defining non-bacterial infections, AIDS-defining opportunistic infections, cART-related toxicity, and others.
The primary study endpoint was in-hospital mortality.The secondary study endpoint was the existence of a TLD (either withholding or withdrawal of organ support) during the ICU stay, as stated in the medical record file.Written arguments for TLD were collected.One-year mortality in patients discharged alive from the index hospital admission was investigated as an exploratory endpoint.

Statistical analyses
Data are expressed as median (interquartile range) for continuous variables and number (percentage) for categorical variables, unless otherwise indicated.The analysis of variance test and the Kruskal-Wallis test were used to compare continuous variables with normal and non-normal distributions, respectively.The Fisher's exact test or the χ 2 test, as appropriate, was used to compare categorical variables.
To assess the patient characteristics that could have influenced the occurrence of the primary study endpoint, we used a logistic regression model with in-hospital death as the dependent variable.As independent variables, we first entered in a global starting model a number of patient characteristics that could have been linked to this endpoint.To reduce the number of potential independent variables per event, we then proceeded to augmented backward elimination (ABE) that combines the standardised change-in-estimate criterion with significance-based backward elimination, with liberal criteria to keep independent variables in the model (threshold values, P < 0.35 and change-in-estimate < 35%) as to minimise the risk of eliminating important explanatory variables [16,17].Passive variables that were considered systematically associated with in-hospital death (namely age, sex, and the SOFA score value at ICU admission) were kept in the models.Other active variables that could have been linked to in-hospital mortality or modified the influence of passive variables were also introduced in the starting model submitted to ABE.This procedure was repeated on 1000 bootstrap samples (with replacement) of the study population.Potentially explanatory variables (exposed with their adjusted odd ratio [aOR] and 95% confidence interval [CI]) were retained in the final model if they were selected in more than 50% of the bootstrap samples, with a root mean square difference ratio < 1.5 and an absolute relative conditional bias of less than 50%.Discriminative ability of the model was assessed through area under the receiver operating characteristics curve (AUROC).
The cumulative survival at one year in patients discharged alive from the hospital was compared between the three groups (active ADC, active NADC, and no active cancer) through Kaplan-Meier analyses and the log-rank test, with right-censoring at the date of last follow-up information.
All analyses were conducted using the R software version 3.5.1 (http:// www.R-proje ct.org).Two-tailed P values < 0.05 were considered statistically significant.

Prevalence and spectrum of ADC and NADC
Overall, 106 (11.3%) and 97 (10.3%) patients presented with an active ADC and an active NADC, respectively.Solid cancers-mostly lung and intestinal neoplasmsaccounted for 62 (63.9%) of NADC, with a metastatic stage in half of cases (Table 2).Hodgkin disease, non-AIDS-defining NHL and acute leukaemia were equally distributed amongst haematological NADC.Burkitt lymphoma and other AIDS-defining NHL represented 86 (81.1%) of ADC whilst only 19 cases of Kaposi sarcoma were identified (Table 2).The diagnosis of neoplasm was made during the index hospitalisation in 30 (48.3%), 21 (60.0%) and 65 (75.6%) of patients with solid NADC, haematological NADC and haematological ADC, respectively (P = 0.005).HIV infection and cancer were co-diagnosed during the index hospitalisation in 18 (17.0%)patients with ADC.Patients with NADC were older, received cART more frequently, had more non-neoplastic chronic conditions and presented with a worst performance status than those with ADC (Table 1).None of the solid organ transplant recipients had an active cancer at ICU admission.

Clinical presentation and organ support in patients with and without cancer
Acute respiratory failure, sepsis and non-toxic coma were the leading reasons for ICU admission, whatever the cancer status (Table 1).The extent of organ failures at admission, as reflected by the SOFA score, did not differ between patients with and without neoplasm.Bacterial sepsis and cancer-related non-infectious complications were the most common definite diagnoses of the ICU stay in patients with NADC and ADC.Notably, NHLrelated tumour lysis syndrome accounted for one-fourth of admissions in patients with ADC.Invasive mechanical ventilation was similarly implemented in patients with and without cancer.Patients with ADC were more often treated with vasopressors and renal replacement therapy than those with NADC or no cancer.Half of patients with ADC received anti-neoplastic chemotherapy in the ICU.

In-hospital mortality
A total of 167 patients (17.8%) died during their hospital stay (Additional file 1: Table S1).Overall hospital mortality was 12.4% in patients without cancer, 30.2% in patients with ADC and 45.4% in patients with NADC (P < 0.0001) (Fig. 1).Around half of deaths in patients with cancer occurred after ICU discharge, which is in the downstream ward, and were related to cancer in most of cases (Table 1).For admissions with bacterial sepsis as the main diagnosis, hospital mortality was 19.4% in patients without cancer, 28.0% in those with ADC and 43.8% in those with NADC (P = 0.008).
Arguments for TLD are exposed in Table 4. Cancer impacted this decision in 47 patients (92.2%) with ADC or NADC, without significant variation according to the classification or the stage of the disease (Additional file 1: Table S4).Organ failures and non-neoplastic comorbidities were less often considered in patients with cancer than in those without.Only 10 patients (9.1%) in whom a TLD was made had advance directives at ICU admission, a proportion that was not higher in those with neoplasm (Table 4).

One-year mortality for patients alive at hospital discharge
At least one follow-up hospital visit was available for 643 (83.3%) out of the 772 patients alive at hospital discharge.Figure 2 shows the cumulative survival at one year in these patients according to the cancer status

P-value
Organ failures 61 ( Chronic condition other than cancer 24 ( (right-censoring at the time of last available information), the lowest survival being observed in those with ADC (log-rank test, P < 0.0001).Vital status at one year was available for 577 (74.7%) out of these 772 survivors of the index hospital admission (Additional file 1: Table S5): 1 year mortality in these patients was 38/488 (7.8%) in those without cancer, 9/53 (17.0%) in those with ADC (including 5/9 cancer-related deaths) and 12/36 (33.3%) in those with NADC (including 10/12 cancer-related deaths) (P < 0.0001).

Discussion
One-fifth of the 939 critically ill PLHIV included in this multicentre study suffered from an active neoplasm, with a similar prevalence of NADC and ADC.Bacterial sepsis and non-infectious cancer-related events accounted for most of admissions in these patients.Cancer was a predominant argument for TLD and strongly predicted in-hospital death after adjustment on other frailty and severity markers.The prevalence of NADC was high in our cohort, exceeding rates reported in critically ill PLHIV over the early cART era [9,[18][19][20][21][22].This trend correlates with evidence that PLVIH, including those with sustained viral control under cART, are at increased hazard for certain infection-related NADC (e.g.EBV-promoted Hodgkin lymphoma) and infection-unrelated NADC (e.g.lung cancer) when compared to age-and sex-matched seronegative individuals [8,23].The volume of PLHIV with cancer referred for ICU admission could continue to rise through the coming decade as the burden of NADC is expected to amplify, notably amongst patients aged 65 years or older [4].Besides, Burkitt lymphoma and diffuse large B cell lymphoma represented most of ADC cases.Those high-grade presentations, which generally prevail in PLHIV [24], are associated with a marked propensity for life-threatening complications entailing ICU admission in patients with NHL, regardless of their HIV status [25,26].This may have contributed to the over-representation of ADC in our cohort whilst NADC now distinctly predominate in the overall population of PLHIV [27,28].
The deleterious effect of cancer on hospital survival has several potential explanations.Most of patients with solid NADC presented with indicators of advanced disease, such as altered performance status, metastatic Fig. 2 Cumulative survival at one-year according to the cancer status in patients discharged alive from the index hospital admission.Kaplan-Meier curves (with right-censoring at the date of last follow-up information) were compared using the log-rank test.ADC, AIDS-defining cancer; NADC, non-AIDS-defining cancer stage, or admission for a complication directly related to the underlying tumour.The noteworthy prevalence of inaugural admissions may also reflect aggressive phenotypes or late diagnoses.This could be especially relevant for haematological malignancies, with a large subset of patients with tumour lysis syndrome requiring renal replacement therapy or treated with chemotherapy during their ICU stay, as observed in other cohorts [29].Moreover, patients with cancer were mainly transferred from wards.Early admission from the emergency department, before the worsening of organ failures, has been shown to improve survival in critically ill individuals with neoplasm [30][31][32].Next, the additional immune deficiency resulting from cancer and antineoplastic chemotherapies may intrinsically impair survival in PLHIV at the acute phase of critical illnesses.Indeed, amongst patients admitted for bacterial sepsis, in-hospital mortality was greater in those with cancer, as reported in studies not focussed on PLHIV [33,34].Lastly, in-hospital fatalities occurring after ICU discharge were far more common in patients with cancer -in this later subgroup, most of deaths occurring in the downstream ward were related to disease progression or new cancer-related complications.
A key question is whether HIV infection impairs per se the outcomes of patients with cancer requiring ICU admission.Data regarding the prognostic impact of HIV infection on the outcome of critically ill patients are conflicting.A large cohort study conducted in the early cART era and including both PLHIV and seronegative controls (matched on age, gender, and medical versus surgical reasons for ICU admission) reported higher unadjusted short-term mortality rates in PLHIV [35].However, most of studies from the same period found no independent association between HIV infection and survival in patients admitted to the ICU for common conditions such as resuscitated cardiac arrest or the acute respiratory distress syndrome [36,37].Also, the severity of HIVinduced immune deficiency, as reflected by the CD4 cell count, does not correlate with short-term outcomes in PLHIV admitted for acute respiratory failure [38], neurological failure [39], or sepsis [40,41].Overall, recent case series of critically ill PLHIV showed similar outcomes that those currently reported in seronegative individuals with comparable demographics, comorbidities, and level of organ dysfunctions [11,21].Next, to the best of our knowledge, the prognosis of critically ill patients with a given cancer type as never been studied according to the HIV status.Nevertheless, the hospital mortality rates that we observed in our cohort for patients with the most common cancer presentations-namely NHL and lung cancer-were similar to those observed in critically ill seronegative patients with these neoplasms [25,26,29,42,43].Altogether, these data suggest that HIV infection has no major impact on the prognosis of critically ill patients with cancer though further studies are warranted to confirm this assumption.
TLD during the ICU stay and in-hospital deaths preceded by such decisions were as frequent in our cohort as in the general population of critically ill patients [44,45].Presenting with an active cancer acted as a major determinant for TLD, along with poor baseline health status and the extent of organ failures.HIV infection was only rarely taken in account, contrasting with studies conducted in the early cART era [46,47].Overall, these results and the evidence discussed above suggest that, in a context of extended access to cART and improved life expectancy, TLD in PLHIV admitted to the ICU now rests upon a similar making-process than for seronegative individuals, with a substantial impact of cancer prognosis.Strikingly, advance directives were unavailable for most of patients with cancer.Though the frequency of inaugural admissions may partly explain this observation, advance directive completion appears essential for better defining the goals of care in an ageing population of PLHIV in whom the prevalence of malignancies is increasing [48][49][50].
This work has limitations.First, ADC and NADC were pooled for analyses due to the restricted number of patients with each cancer subtype.Dedicated studies are necessary to appraise the specific prognostic impact of common neoplasms such as lung cancer or AIDSdefining NHL in critically ill PLHIV.Second, owing to the retrospective design, some relevant predictors of outcome might have been omitted, notably the time elapsed from the first symptoms of critical illness to ICU admission.Transient stress in bed availability or reluctance for ICU admission of patients with cancer could have led to delayed referral and impaired prognosis.Along this line, policies for both admission and do-not-resuscitate decisions may differ from one ICU to another, which could affect the external validity of our results.Next, we did not collect information on the course of organ failures throughout the ICU stay.Nevertheless, life-sustaining therapies were entered as active variables in the model for mortality prediction and whether organ failures influenced TLD was extracted from the patient report files.Lastly, as follow-up ended at hospital discharge, we did not explore the effect of critical illness on disease progression, functional outcomes and long-term survival in PLHIV with cancer.

Conclusions
NADC and ADC are equally prevalent, stand as a leading argument for TLD, and strongly predict in-hospital death in the current population of PLHIV requiring critical care.Enhanced screening procedures for earlier cancer diagnosis, routine policy for prompt ICU admission and promotion of advance directive completion in case of documented neoplasm may constitute important axes for improving the quality of care in these patients.

Table 1
Characteristics of the study population according to the cancer status

Table 2
Features of active cancers and association with in-hospital mortality in the study population

Table 3
Independent predictors of in-hospital death: results of the final model aOR, adjusted odds ratio; CI, confidence interval; NADC, non-AIDS-defining cancer; ICU, intensive care unit; ADC, AIDS-defining cancer; SOFA, sepsis-related organ failure assessment; WHO, World Health Organization

Table 4
Written arguments for treatment limitation decision in the intensive care unit