Factors associated with intensive care admission in patients with lung cancer: a population-based observational study of 26, 731 patients

We conducted a retrospective, observational, multi-centre, population-based study through an analysis of secondary data comprising linked cancer registration, hospital discharge, intensive care and mortality records. Full details are described elsewhere [2]. Patients resident in the West of Scotland who had a diagnosis of lung cancer (ICD-10 codes; C33, C34.0, C34.1, C34.2, C34.3, C34.8 and C34.9) on the Scottish Cancer Registry between 1st January 2000 and 31st December 2009 were included in this study. Small Cell Lung Cancers (SCLC) were classified as ICD-O morphology M 8041/3 to M 8045/9; all others were classified Non-Small Cell Lung Cancers (NSCLC). We investigated whether they had been admitted to one of the 16 general ICUs located in the region within 2 years of the date of cancer incidence up to 31/12/2011. The first admission to an ICU within 2 years following a diagnosis of lung cancer was identified and linked to the appropriate individual episode of hospital care. We used death and hospital discharge records to identify whether patients died during their hospital stay.

The study used four linked data sets: the Scottish Cancer Registry, Scottish Morbidity Record 01, national death records and the Scottish Intensive Society Audit Group WardWatcher ICU database. WardWatcher collects data on patient demongraphics, admitting specialty, admission diagnosis, the Acute Physiology and Chronic Health Evaluation (APACHE) II scoring system, and the type of organ support. Organ support was defined as receipt of invasive mechanical ventilation, vasoactive drugs to provide cardiovascular support, or renal replacement therapy.

The SMR01 hospital discharge records were reviewed for all primary and secondary diagnoses in the 5 years preceding the date of lung cancer incidence. These ICD-9 or ICD-10 coded diagnoses were computed to determine the presence or absence of co-morbid disease according to the Charleson Comorbidity Index (CCI) through previously published coding algorithms [9, 10].

This study was approved by the West of Scotland Research and Ethics Committee. Approvals to use the data were obtained from the West of Scotland Critical Care Research Network, SICSAG, and the West of Scotland Cancer Surveillance Unit.

Patients who were admitted to an ICU were compared to the remainder of the incident lung cancer cohort according to demographic, clinical and cancer-related variables. Summary statistics including mean and standard deviation or median and inter-quartile range were determined for continuous exposure variables accordingly. Categorical variables were summarised according to frequencies and proportions. Non-ordered categorical variables were analysed by a chi-squared test of association whilst ordinal variables were compared between the two groups by use of a chi-squared test for trend.

Factors associated with admission to ICU were estimated using a multivariable logistic regression model. The odds of survival were computed for each exposure variable using univariable analyses with 95% confidence intervals and P values. Variables found to have a potentially statistically significant alteration in the odds of survival (P < 0.3) in univariable analyses were included in an adjusted multivariable model. The cancer treatment modalities (surgery, radiotherapy, chemotherapy) were not included in any multivariable model due to potential collinearity with treatment intent. A P value of < 0.05 was used to identify statistically significant associations in multivariable modelling for ICU admission.

The authors do not have permission to share the data used for this article, however, the data can be accessed via application to the Information Services Division of NHS Scotland.

Reasons for ICU admission were predominantly related to respiratory conditions with lower respiratory tract infections and complications from the malignancy accounting for 27.4 and 20.3% of all admissions respectively. However, the ICU admission was entirely unrelated to the lung cancer diagnosis/ treatment in one in four ICU lung cancer patients. The majority of ICU patients with lung cancer required organ support, the most common modality being invasive ventilation (70.9%) reflecting the high incidence of respiratory disorders.

For patients admitted to ICU, mortality in ICU, hospital and six-months post admission were 41.5, 58.0 and 68.8%, respectively. Mortality in our study was relatively high when compared with previous multi-centre studies [11‐13]. The largest study of this nature was performed on the American surveillance, epidemiology and end results- medicare registry (SEER), including nearly 50,000 patients. Hospital mortality was reported as 24%, significantly lower than that seen in this study. However, mortality at six-months was similar at 65% [13]. This registry-based study did not detail any of the features of the critical illness which has been demonstrated to be the largest determinant of short-term mortality [3]. Soares et al conducted the only prospective, multi-centre cohort study and reported ICU, in-hospital and six-month mortality of 28, 39 and 55% respectively. However, the proportion of subjects requiring mechanical ventilation was only 53% compared with the 79% observed in our study, suggesting that the burden of critical illness was less.(11).

While the majority of patients with lung cancer were treated with palliative intent, a disproportionate number of patients in the ICU group had received curative treatment with the majority of patients receiving surgical intervention. This observation likely reflects clinician behaviour whereby patients with curative treatment intent are preferentially selected for ICU during an acute critical illness. However, this could also be partially attributable to post-operative complications in patients undergoing curative surgical procedures as the majority of ICU admissions were surgical in nature (66%) and ICU admission was directly related to cancer surgery in 15%.

Surgical treatment of cancer was found to be the strongest predictor of ICU admission with a seven-fold increase in odds of ICU admission compared with those not treated with surgery. This may be due to critical illness occurring at the time of surgery with one in eight ICU admissions were directly attributable to the surgical treatment. While it is possible that patients having received curative surgery were being considered better candidates for ICU, treatment intent was not associated with ICU admission on the multivariable model. Other treatment interventions influenced ICU admission by reducing the odds by approximately half in those who were treated with chemotherapy or radiotherapy compared with those who were not. As these patients are being admitted early after diagnosis they may not have had the opportunity to receive these treatment interventions prior to ICU admission.

Age of incidence was found to be a strong negative predictor of ICU admission with a reduction in the odds of admission with increasing age above 75 years that reduces further for those aged over 80 years. Previous work by Azoulay et al. demonstrated that the largest influence on a refusal for ICU admission in critically ill patients with cancer was an age of over 65 years [14]. Within the general ICU population, increasing age has been associated with poorer survival both in the short and longer term, and that this was particularly pronounced for patients aged over 75 years [15]. It seems unlikely that critical illness is less common with increasing age, particularly given the increased incidence of comorbidities encountered in an elderly population, rather that concerns about poor outcomes in the elderly patients swings the balance of benefit versus harm away from ICU admission.

Comorbidity was not associated with an increase in odds of ICU admission in the multivariable model. The study by Slatore et al. [16] demonstrated that comorbidity has been associated with poorer outcome after ICU admission in patients with lung cancer, however, our study suggests that this does not necessarily influence admission patterns.

Females had a statistically significant reduction in the odds of ICU admission compared to males (OR 0.75, 95% CI 0.61–0.92, P = 0.006). An increased incidence of critical illness in men has been described for patients with severe sepsis and trauma [17, 18]. The cause of this is not clear but may reflect differences in lifestyle choices, health behaviours or the impact of hormones on stress responses. Furthermore, it is possible that the gender difference may not reflect difference in rates of critical illness but a difference in preferences for end of life care. It is known from other studies that survival from lung cancer is poorer in men than women overall. Among those with a critical illness that we report, survival was also poorer in men than women. Given that a minority of lung cancer patients is admitted to an ICU (1.5%), we would conclude that both non-critically ill as well as critically ill female lung cancer patients have better survival than men.

Our study has a number of strengths. It is the first study to compare lung cancer patients admitted to an ICU with those who were not over a ten-year period. It utilised a large sample size and was conducted across multiple general rather than oncological ICUs, which may improve its generalisability. We used datasets with high levels of case ascertainment for cancer incidence, hospitalisation and deaths. Measurement bias was reduced through the use of objective scoring systems to capture acute illness and past medical history. The use of the SIMD provided a robust, multi-dimensional measure of socioeconomic status rather than relying on a single domain as a proxy for affluence as in previous studies [12, 13, 19]. Exposure variables from the ICU admission were collected and determined prospectively thereby reducing the likelihood of misclassification.

This study has several limitations. Large proportions of missing data precluded the use of information on ethnicity, cancer stage and pathological sub-division which could confound the observed associations. Furthermore, while the date of cancer diagnosis in all cases was prior to ICU admission, we could not determine the certainty of diagnosis at the point of entry to critical care which may affect decisions about whether to admit a patient. We accept that approaches to managing critical illness are likely to differ in patients receiving palliative care and while we adjusted for treatment objective (curative, palliative or unknown), we did not have information on palliative care involvement within the data sources utilised in this study.