Introduction

In the past decades, we are experiencing a rapid ageing in Western countries. People live longer but with more chronic diseases and care dependence at the end of life. As a result, an increasing number of older patients are hospitalised.1 In 2014, 25% of all hospitalised patients in Belgium were 75 years old and over. As in all Western countries, it is expected to rise. Some of these patients are admitted on acute geriatric units (AGUs), known for its focus on comprehensive geriatric assessment, early rehabilitation, early discharge planning and person-centred care; others are admitted on non-geriatric wards in which there is a more single-disease approach.2 In Belgium, each acute care hospital has a geriatric care programme where frail older patients are mainly admitted to acute beds managed by the geriatricians. When older people are admitted on non-geriatric wards, the geriatric liaison team offers comprehensive geriatric assessment to older patients screened as frail or on request of the treating team. Each acute care hospital also has a palliative care liaison team where experts in palliative care can be consulted on request of the treating team.

As most people will use hospital services in their final year of life, hospitalisation may be important to initiate advance care planning.3 4 Early integration of a palliative care approach including advance care planning has proven to be beneficial for patients and families at the end of life: there is better quality of life in patients and less mental stress in family members5–8 without shortening of life.7 9 In particular, older hospitalised persons are at increased risk of being deprived from palliative care.10 11 One of the main causes is prognostic uncertainty in the older patient, causing physicians to be less confident in making the switch from cure to care.12 13

The Supportive and Palliative Care Indicators Tool (SPICT) was developed for a more timely identification of patients with palliative care needs for first-line medicine.14 Besides self-reported palliative care needs, prognostication is essential in medical decision making.15 16 Prognostication is essential for many reasons such as supporting patient-centred communication and advance care planning and avoiding the burden of intense level of care in patients with poor prognosis.15 16 Until now, SPICT has insufficiently been validated in older persons neither in the acute care hospital nor for prognostication. To the best of the authors’ knowledge, there is only one study that used SPICT for prognostication in AGU which gave promising results but was limited by its retrospective design.17

The aim of this study was to assess prospectively the prognostic value of SPICT for risk of death within 1 year in older persons admitted to the hospital. As the consequences of a false-negative result (deny patients a palliative care approach including advance care planning) outweigh the consequences of a false-positive result (possibly starting palliative care principles too early),12 18 the authors value sensitivity as being more important than specificity in the assessment of the prognostic accuracy. In each of the research questions, comparison is made between older people admitted at the AGU versus a disease-specific ward as we know that the people admitted on the AGU are more frail than people admitted in disease-specific wards.2

MethodsDesign, participants and settings

Patients aged 75 years and older were prospectively recruited from the AGU and cardiology unit (CU) of four hospitals in Belgium. The CU was chosen as an example for a non-geriatric ward because cardiovascular disease is one of the most prevalent comorbidities in older patients. CUs manage younger and older patients presenting with acute cardiovascular disease without need for intensive care. All older patients with a length of stay of more than 48 hours were eligible for the study (in order to be able to do a thorough assessment). We excluded patients who were transferred from other wards (premorbid SPICT is more difficult to assess when patients had already been admitted elsewhere), and patients who were already included earlier in the study.

Data collection and measures

From January to July 2018, junior doctors trained in geriatric medicine and experienced members of the geriatric team assessed the premorbid palliative care needs by means of the first part of SPICT, asking the patient/family to think about the situation 2 weeks before admittance, blinded from the treating physician. SPICT combines 6 general and 23 disease-specific indicators regarding deteriorating health.14 Different versions of SPICT have been published, and the French and Dutch 2017 version is used in this study (www.spict.org.uk); permission was obtained. There is SPICT identification when at least one general and one disease-specific indicator is present.

Sociodemographic information, basic and instrumental activities of daily living (iADLs) assessed by using Katz Scale19 and Lawton Scale,20 respectively, length of stay and comorbidity measured by Charlson Comorbidity Index21 were collected from the medical record after patient discharge (these data are obtained in routine care).

One year after admission, the patient and family were contacted by a junior doctor or data nurse to collect survival status and timing of death as main outcome measures.

Statistical analysis

Descriptive statistics were computed using IBM SPSS Statistics software V.25. For continuous data, range, median and IQR were computed and the Mann-Whitney U test was used for comparisons; categorical data were compared using Pearson’s χ2 tests. The exact p values are reported, with statistical significance defined as p≤0.05. A multivariable Cox’s regression was used to assess the association between SPICT and 1-year mortality adjusted for age, gender and type of unit (fixed affects) and using hospital as random effect.

To assess diagnostic accuracy, summary receiver operating characteristic (SROC) curves were constructed to obtain the pooled values of sensitivity and specificity. Bivariate modelling for sensitivity and specificity together was performed in R V.3.5.2, using the mada package V.0.5.822 (https://cran.r-project.org/web/packages/mada/vignettes/mada.pdf). Differences in mean sensitivity and specificity were assessed by means of a metaregression. As partial area under the curve (AUC) is restricted to clinically sensible thresholds, we put this measure forward as the measure of our preference.23

ResultsParticipants

Of the 2136 older patients admitted during the study period, 634 were excluded mostly because of length of stay of less than 48 hours (most on CU) or staying more than 48 hours on another ward before coming to AGU or CU. Of 1502 patients eligible for the study, 933 were not included in a period in which the junior physician who asked for informed consent was on holiday or had too much clinical work. Furthermore, 44 patients refused to participate in the study (no informed consent) and 69 incompetent patients did not have a legal representative, leaving 458 patients for the study (figure 1).

Figure 1

Flowchart of recruitment.IC: informed consent.

Older patients admitted at the AGU were older, female and more dependent, according to ADL and iADL scores (table 1).

Table 1

Characteristics of the study population according the type of unit (acute geriatric vs cardiology)

SPICT characteristics

All general indicators of SPICT were more prevalent in AGU versus CU, except for ‘persistent symptoms’ which was present in one in five older patients both on AGU and CU (p=0.823). Lowest prevalence was the indicator concerning patient or family request for palliative care (5.7% of patients on AGU, 1.2% on CU, p=0.007). The median AGU patient had two general indicators present vs one in CU (p<0.001).

For the disease-specific indicators, the median in AGU was 1 (mostly frailty/dementia) vs 0 in the CU (when present, severe heart disease was most prevalent) (<0.001).

In total, 59.3% (124/209) was SPICT identified on AGUs vs 40.6% (101/249) on CUs (p<0.001). SPICT-identified patients in CU reported more functional needs and more symptoms compared with SPICT non-identified patients on CU. On AGU, SPICT-identified patients reported more functional needs only (table 2).

Table 2

Comparison of premorbid care needs† between SPICT-identified and SPICT non-identified patients

1-Year outcome and survival time analysis

Of 458 patients, 2.8% died in the hospital. One-year mortality was known in 202 out of 209 AGU and 242 out of 249 CU patients; 1 year-mortality was 24.3% (n=49/202) vs 21.5% (n=52/242), respectively (p=0.488). One-year mortality differed between SPICT-identified and SPICT non-identified patients on both units (p<0.001). Positive predictive value (PPV) of SPICT was comparable on both units: 33.9% on AGUs and 33.7% on CUs (p=0.972). The negative predictive value (NPV) was 89.3% and 86.8%, respectively (p=0.582).

A first Cox regression model showed no significant interaction between ward (AGU vs CU) and SPICT. In the second Cox regression model without the interaction included, the HR for SPICT was 2.864 (95% CI 1.808 to 4.538) (p<0.001)after adjusting for age, gender and type of unit (figure 2).

Figure 2

Time to death according to SPICT identification and type of unitLegend: the presented HR is based on a Cox regression model after adjustment for age, gender and respecting strata of four hospitals. SPICT non-identified patients admitted at the geriatric unit (black dotted line); SPICT non-identified patients admitted at the CU (grey dotted line); SPICT-identified patients admitted at the geriatric unit (full black line), SPICT-identified patients admitted at the CU (full grey line). AGU, acute geriatric unit; CU, cardiology unit; SPICT, supportive and palliative care indicators tool.

Diagnostic test accuracy

The SROC curves are shown in figure 3. The likelihood ratio tests (metaregression) showed that there was a significant difference for ward (p sensitivity: 0.001, p specificity: <0.001), meaning that the diagnostic accuracy differed significantly between AGU and CU (better on AGU). The pooled average sensitivity, specificity, AUC and partial AUC are presented in table 3. Partial AUC was 0.822 on AGU vs 0.651 on CU (p<0.001).

Figure 3

Summary ROC curvesLegend: the pooled estimate and SROC curve based on bivariate analysis (n=458) are visualised in a scatter plot of the false-positive rate (1 - specificity) and the sensitivity for the individual hospitals, including 95% confidence regions for the pooled estimates. The results of the sampled wards are visualised by a triangle; the pooled estimate per type of ward (cardiology in light blue, acute geriatric unit in darker blue) is indicated by a circle. SROC, summary receiver operating characteristic.

Table 3

Diagnostic accuracies of SPICT (bivariate analysis)

DiscussionMain findings

In a cohort of 458 older hospitalised patients, 60% of older patients admitted at AGUs and 40% of older people on CUs were SPICT identified. SPICT-identified patients in CU reported more functional needs and more symptoms compared with SPICT non-identified patients on CU. On AGU, SPICT-identified patients reported more functional needs only.

In SPICT-identified patients, the hazard of mortality was three times higher compared with SPICT non-identified patients. After 1 year, one in three had died vs 12% in those who were non-identified by SPICT. SPICT performed better as a prognostic tool in AGUs compared with CUs; the pooled sensitivity was 0.82 on AGU compared with 0.69 on cardiology. Specificity was 0.49 vs 0.67, respectively, and partial AUC was 0.82 vs 0.65.

Discussion of results

SPICT had good sensitivity and moderate specificity for predicting 1-year mortality in the older hospitalised patient. This accuracy is similar or better compared with other palliative care tools studied in a general adult hospitalised population as the GSF-PIG (Gold Standards Framework Proactive Identification Guidance) (sensitivity 0.78, specificity 0.72, PPV 38%, NPV 94%),18 NECPAL (Necesidades Paliativas) (sensitivity 0.91, specificity 0.33, PPV 33%, NPV 91%)24 and PALLIAR (AUC 0.73).25

In contrast to the latter studies, our study focused on the older hospitalised person. To the best of our knowledge, we found only two other prospective studies exploring the prognostic accuracy for 1-year mortality specifically in such a population. Ritt et al compared five frailty instruments in two AGUs; the Clinical Frailty Scale26 showed best diagnostic accuracy with an AUC of 0.85.27 They hypothesised that the superiority of the Clinical Frailty Scale26 is due to the physician’s judgement of the remaining life expectancy, which is absent in other frailty instruments27; the added value of clinical appraisal in prognostication is also a point of discussion in palliative care literature.28 29 Pilotto et al found that the Multidimensional Prognostic Index had an AUC of 0.76 for 1-year mortality30 comparable to our findings; however, the Multidimensional Prognostic Index is more extended in comparison to SPICT in which a short patient/family interview is needed, as well as a short medical history review.

This study showed that SPICT was less accurate in older patients hospitalised in the CU. Although the 1-year mortality was the same for both units (one in five), the older persons on cardiology were less frail and had less severe comorbidities. A possible explanation for the difference in accuracy might be that older persons on cardiology die less because of frailty but more because of cardiac death,13 which might be insufficiently captured by SPICT. Indeed, the gold-standard tool for prognostication in heart failure, the Seattle Heart Failure Score, includes age, gender, heart function and creatinine.13 31 Fu and colleagues showed in a cohort of older hospitalised heart failure patients, that the Seattle Heart Failure Score had an AUC of 0.80 which exceeds the AUC of SPICT in this study.32 However, the advantage of SPICT lies in its applicability for a heterogeneous population, which makes hospital-based screening possible.

Limitations and strengths

To the best of the authors’ knowledge, this is the first study in hospitalised older patients comparing the prognostic accuracy of a palliative care tool between older patients treated on AGUs and on a disease-specific ward. Another strength is the prospective multicentre design with few missing data.

However, there are some limitations. First, selection bias cannot be excluded: (1) two-thirds of eligible patients were not included because of time constraints, and we had no patient characteristics of the missing patients, so this was more a convenience sample than a cohort; this was due to working with junior doctors who often had too much clinical work to include patients in the study; finding resources for qualified study personnel is warranted in the future; (2) we excluded patients who were transferred from other wards and patients with short stays, probably excluding the most sick and the fittest older patients. This may give rise to bias, probably underestimation of the discriminative power. Furthermore, we are not sure if our results are generalisable to other disease-specific wards treating the older patient. This study must be viewed a meta-analysis of four samples for AGU and CU. For AGU, the 1-year mortality rate of 24% is in line with other studies performed in different countries,27 underscribing the generalisability for AGUs.

Practice implications

The prognostic accuracy of SPICT for risk of dying within 1 year is similar compared with the widely used Clinical Frailty Scale26 tool and the Multidimensional Prognostic Index.30 These tools might help hospital clinicians to overcome prognostic paralysis and to earlier integrate a palliative care approach including advance care planning conversations. SPICT is a tool meant to identify patients who are at risk of deteriorating and dying. SPICT identification should prompt clinicians to initiate proactive holistic needs assessment, shared decision making about goals of care and anticipatory care planning.14 While basic palliative care should be considered as necessary to offer to SPICT-identified patients, patients with persistent symptoms should be offered specialised palliative care assessment and support when appropriate. Interestingly, patient or family request for palliative care was the lowest-scoring general indicator (5% on AGUs, 1% on CUs). This low percentage is comparable to other studies,18 showing that there is a need not only for a mind shift from cure to care in clinicians33 but also for better acceptance of early palliative care in society.34 35 The advantage of SPICT over geriatric assessment tools such as the Clinical Frailty Scale26 is its focus on care needs as pronounced by the patient and the family, which may give way to introducing a palliative care approach more easy.

However, there is a need for research about how to introduce SPICT as a hospital-based screening tool.36 37 More importantly, we lack evidence on what kind of care model is best to start after identification in acute care hospitals, and more particular in older hospitalised patients. The needs of SPICT-identified older patients on the CU are mainly in the functional domain: one may thus wonder if SPICT should be used to introduce expert geriatric care or expert palliative care or a new care model in which both geriatric and palliative expertise are integrated. It is also under debate if geriatricians and disease-specific specialists should be the ones providing a palliative care approach or if specialised palliative care should come into the wards,38 though there is a preference for combining both.13 39 However, there is a lot of diversity in the studied models of integrated palliative care and thus a call for the development of standardised and conceptually unambiguous strategies.9 40 41

Furthermore, we put forward that we preferred a high sensitivity above specificity in order not to deprive older people from a palliative care approach. However, when considering practical feasibility and costs on a population level, one may choose specificity over sensitivity. Also for treatment limitation decisions, physicians might prefer a higher PPV. When two out of three of SPICT-identified patients are still alive after 1 year, treating physicians do no harm by starting advance care planning and focus on symptom management. However, they must stay precautious not to undertreat older patients and optimally invest in physical revalidation to regain functionality. In short, SPICT can be seen as a good way of starting to think and talk about the right balance between revalidation/life prolongation and comfort in the older patient, always in the perspective of what the patient truly values in life.

Future research

Further studies should assess the accuracy of SPICT in other contexts (emergency department, nursing homes, etc) by different professionals. It would be interesting to make a head-to-head comparison with the Clinical Frailty Scale and to study if clinical appraisal adds to prognostic accuracy.

Future studies should investigate the feasibility, cost and impact of screening for risk of 1-year mortality and palliative care needs in hospitals.

Acknowledgments

The authors thank Maes H, De Bock R, Van de Vyver C, Duyver E, Cornil C, de la Kethulle de Ryhove C and Badin H for the data collection and data entry, and Guédès A for making recruitment possible.