In total, 19.2% of patients in our study had LST limitation (13.0% withheld and 6.2% withdrawn, respectively). While a number of factors were found to be common independent predictors of both withholding and withdrawal of LST, there were additional independent predictors specifically for withholding and for withdrawal of LST. Both withholding and withdrawal of LST were independent predictors of hospital mortality.

International differences in practices of LST limitation are striking [2, 3]. Considering all ICU patients, the proportion with LST limitation in our study (19.2%) was higher than the international average of 11.8% seen in the ETHICUS-2 study which surveyed 199 ICUs in 36 countries in 2015 and 2016 and the international average of 13.2% seen in the Intensive Care Over Nations (ICON) study which surveyed 730 ICUs in 84 countries in 2012 [15, 16]. On the other hand, considering only patients who ultimately died in hospital, the proportion with LST limitation in our study (65.8%) was lower than the international average of 76.6% in the ETHICUS-2 study but higher than the international average of 39.5% in the ICON study. Notably, our study included DNR orders as withholding of LST, just as the ETHICUS-2 study did, while the ICON study did not specify details. Focusing on Asia, the 19.2% with LST limitation in our study was between the average of 10.9% across China, Hong Kong, India, Japan, South Korea, and Thailand in the ETHICUS-2 study and the average of 30% across China, Indonesia, Japan, Malaysia, the Philippines, Singapore, Taiwan, and Thailand in the ICON study. In the Asian Collaboration for Medical Ethics (ACME) study of attitudes towards end-of-life care in the ICU in Asia, ICU physicians in Singapore tended to withhold and withdraw LST more than many other countries, possibly due to a combination of factors including a Western influence on education in a high-income economy [3, 12]. Indeed, decisions to limit LST are less frequently reported in countries with low or lower-to-middle gross national income [16].

Most studies do not distinguish between predictors of withholding and withdrawal of LST. Common independent predictors of both withholding and withdrawal of LST in our study were greater age, no chronic kidney dialysis, greater dependence in activities of daily living, cardiopulmonary resuscitation 24 h before ICU admission, higher APACHE II score, and higher level of care in the first 24 h of ICU admission. Other than absence of chronic kidney dialysis, this is similar to other factors identified in Europe, the United States, and Australia, where patients with LST limitation were older [17,18,19,20,21], had severe comorbidities [8, 22], poorer baseline functional status [16], and were more severely ill [19, 20, 23]. In an international ethics round table consensus statement, factors considered important in deciding to limit LST were an expected survival of less than 1–3 months, multiple organ failure, very severe brain injury, and the health care team's opinion that the patient is receiving non-beneficial therapy or has a nonsurvivable injury [24]. What is surprising in our study is that chronic kidney dialysis was not a predictor for limitation of LST. This may be because chronic kidney dialysis patients are at higher risk of complications and would thus more likely be admitted to a higher acuity setting for post-procedure monitoring. In our local context, LST is typically not limited when patients undergo invasive procedures, so that adequate support can be provided in the event of any post-procedural complications. In addition, the prognosis of chronic kidney disease is uncertain and involves complex decision-making. Prior studies have shown that palliative care is often suboptimal for patients with end stage kidney disease receiving dialysis treatment, with a high rate of intensive care needs towards the end of life [25], and that most nephrologists feel unprepared to lead end-of-life care preferences discussions [26].

In our study, a higher proportion of patients had LST withheld rather than withdrawn. This is consistent with a survey performed in Asian ICUs where physicians reported that they often withheld but seldom withdrew LST at the end of life [12], and may reflect physicians’ views that the two are not the same. Importantly, our findings highlight three differences in additional predictors of withholding versus withdrawal of LST, despite experts’ assertions that the two are ethically similar [6]. First, although it is known that practices of LST limitation are driven by patient demographics and cultural differences, previous studies—which were mostly from the United States and Europe—had generally not distinguished between withholding and withdrawal [18, 24, 27, 28]. In our study, being of a Chinese race and having the religions of Hinduism or Islam independently predicted the withholding but not the withdrawal of LST. While pre-existing data in the literature suggest that the people of China were generally averse to DNR orders [29], and that Hindu and Islamic physicians in Asian ICUs were less likely to withdraw LST [30], such findings are not directly applicable to Singapore’s multicultural society. Second, our findings that malignancy and chronic liver failure independently predicted withholding but not withdrawal of LST are consistent with those of the ICON study and the international SAPS 3 database, although these studies did not differentiate the two forms of LST limitation [16, 31]. Third, lower hospital paying class (meaning more government subsidy) was independently associated with LST withdrawal but not withholding. In Singapore, while greater subsidies for hospital bills are provided for patients’ with lower monthly incomes, out-of-pocket payments are still required [32]. While previous data show that financial considerations play a significant part in decisions to withdraw LST in Asian low-middle income countries [30], our results imply that this may also apply to a high-income nation like Singapore. Taken together, the three afore-mentioned points suggest that withholding of LST is more often affected by baseline characteristics such as race, religion, and comorbidities, while withdrawal of LST is more often considered when costs accumulate as the provision of intensive care continues for longer than anticipated and when the point of medical futility is reached. As depicted in the Kaplan–Meier survival curve (Fig. 1), withholding was more strongly associated with earlier mortality, but withdrawal was more associated with later mortality.

Hospital mortality was 82.1% in our withholding group and 91.8% in our withdrawal group, compared to 71.9% and 88.5%, respectively, in the international ETHICUS-2 study and 86.4% and 92.5%, respectively, in the international SAPS 3 database [15, 31]. The mortality of patients with LST withdrawal was higher than that of patients with LST withholding on univariable analysis. Both were higher than that of patients with no LST limitation. This is unsurprising, since physicians often reserve LST limitation for patients with the gravest prognosis. What is more instructive, however, is that even after accounting for baseline characteristics including severity of illness and level of care in the first 24 h of ICU admission through multivariable GLMM analysis, withholding (odds ratio 13.822) and withdrawal (odds ratio 38.319) were still significantly associated with mortality. While the presence of hidden confounders such as illness severity must be considered, the high odds ratios suggest that hospital survival may have been possible had LST not been limited.

Another interesting finding in our study is that although a minority of patients with LST limitation survived till hospital discharge, a substantial proportion of this minority were discharged home (as opposed to another healthcare facility). 64.7% of hospital survivors who had LST withheld, and 55.6% of hospital survivors who had LST withdrawn, were discharged home. As patients were followed up only until hospital discharge, it is unknown whether these patients passed away soon after, or if a terminal discharge was arranged. In Singapore, terminal discharge is not an uncommon practice [33]. To the best of our knowledge, prior studies on limitation of LST practices report hospital survival but do not have data regarding hospital discharge destination.

Our study has several strengths. First, it was conducted in 21 ICUs across all Singaporean public hospitals over a substantial period of six years, with a large sample size. Second, there were hardly any missing data. Third, it contributes knowledge to a scientific field which is still overwhelmingly dominated by a Western perspective and under-represented by Asian countries. Fourth, it explored an important topic through a unique lens, focusing on the similarities and differences of withholding and withdrawal of LST. Our study also has several limitations. First, it was confined to a single country. Nonetheless, given Singapore’s multicultural nature, it is to our knowledge the first study to compare LST limitation practices in the ICU for several of the world’s major races and religions in one country. Second, it recruited patients over an average of three months annually, rather than across the year. This was done so as to be able to include as many ICUs as possible with the number of data coordinators available. Regardless, the findings remain representative as practice patterns were unlikely to have varied depending on the time of the year. Finally, specific information on when exactly the decisions were made to limit LST was not available, thus preventing a detailed analysis of circumstances around such decisions.

Our study has implications for future research in end-of-life care in the ICU. Investigators aiming to elucidate the predictors of LST limitation in local, regional, or international settings should clearly differentiate withholding and withdrawal, because the two are not the same. Future work should also go beyond exploring the univariable association of LST limitation with mortality to evaluating the true cause-and-effect impact of withholding as well as withdrawal of LST on short and long-term patient-centric outcomes.