WHAT IS ALREADY KNOWN ON THIS TOPIC

Early palliative care can improve the quality of life for patients with non-communicable diseases in advanced stages.

However, it is difficult to identify patients with palliative care needs early and at the right time for reasonable palliative care.

WHAT THIS STUDY ADDS

Seven screening instruments were identified, but none comprehensively assessed the patient’s physical, psychological and spiritual distress, and their need for social support.

The overall methodological quality of the evidence ranged from very low to moderate.

In hospitals, Supportive and Palliative Care Indicators Tool has better clinical performance than Necesidades Paliativas (Palliative Needs) and Taiwanese version Palliative Care Screening Tool, with a pooled sensitivity of 79.8% (95% CI 72.6% to 85.5%).

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

Future studies should focus on validating screening instruments against relevant criteria for referring patients to palliative care rather than solely against their ability to predict mortality.

Additional efforts are required to enhance the comprehensiveness of existing instruments by encompassing a broader spectrum of concerns, including those related to the physical, psychological, social and spiritual dimensions

The applicability of these instruments in different healthcare settings should be validated to improve their clinical performance.

Background

More than 50 million people die from non-communicable diseases (eg, cancer, cardiovascular diseases, and respiratory diseases) globally each year—equivalent to 71% of all deaths.1 2 People with life-limiting illnesses may experience a significant degree of physical, psychological and spiritual distress.3 For them, palliative care is an approach to preventing and alleviating symptoms and improving quality of life through early identification and appropriate management of symptoms.4 5 The need for palliative care continues to grow as a result of the ageing population and the incidence of non-communicable diseases.

Early palliative care is associated with improved symptom control, satisfaction with healthcare, quality of life and survival.6 7 It may also reduce hospital stays, the use of aggressive treatments close to death and healthcare expenditure.8 9 However, most people still only receive palliative care in their last few days or weeks of life. Delayed access to palliative care can contribute to negative outcomes including ineffective medical interventions, higher healthcare costs,10 inappropriate use of treatment modalities,11 and insufficient support for patients and their families.12

The early detection of individuals who require palliative care is essential for the timely initiation of palliative care services.13 Several screening instruments have been developed to identify individuals who would benefit from palliative care so that healthcare professionals (HCPs) can tailor palliative care to the specific needs and preferences of patients.14–19 Five reviews have been conducted to evaluate these instruments but were limited to narrative summaries of the main characteristics or psychometric properties of screening instruments. They did not conduct a meta-analysis to pool the clinical performance of the instruments.19–23 Additionally, existing reviews only included instruments used within a single type of setting, such as primary care settings or hospital settings.19–21 23 Therefore, it is unclear how screening instruments were used across different settings, and whether variations exist in their clinical performance across these instruments and different settings.

To address the gap, this systematic review aimed to (1) Identify existing screening instruments for early identification of individuals who are in need of palliative care, irrespective of setting of care; (2) Describe the key characteristics and psychometric properties of screening instruments, including validity and reliability; and (3) Pool and compare their clinical performance in identifying patients with palliative care needs early.

Methods

This systematic review was registered in PROSPERO (CRD42022335942) and followed the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses statement.24

Search methods

We conducted two rounds of literature searches. First, we performed a systematic search using four English databases (PubMed, CINAHL, Embase and Scopus) and two Chinese databases (CNKI and Wanfang), from inception to May 2023. After identifying the instruments, we used their name to search for additional relevant studies. A detailed search strategy is provided in online supplemental appendix S1.

Study selection process

All retrieved articles were imported into the software Covidence to facilitate citation management. After removing duplicates, two researchers independently screened by titles and abstracts of the included citations. Full-text reading was then performed by the same researchers against the pre-established criteria. After that, the researchers completed quality evaluation, data extraction and analysis. Discrepancies between researchers were resolved by a third author when necessary.

Inclusion/exclusion criteriaTarget population

We included instruments targeting: (1) Adults aged 18 years or older; and (2) Patients with any advanced non-communicable diseases such as cancer, heart disease, dementia, organ failure and so on.

Instruments

Screening instruments were used to identify patients with palliative care needs early. We excluded screening instruments designed for a specific disease because our objective was to identify generic instruments for multiple diseases and settings. We excluded screening instruments specifically for intensive care units and emergency departments. It was because the rapid change in patient conditions in these settings makes long-term palliative care needs prediction challenging.25 We excluded screening instruments that were primarily used in the last few days or weeks of life because these instruments are not suitable for the early identification of palliative care patients.

Outcomes

The outcomes of interest were the development, validation and clinical performance of screening instruments. For development of the instruments, we examined aims and methodology. With regards to key characteristics and psychometric properties of screening instruments, we explored the number of items, responses format, time frame, scope of population, settings for use and content. The clinical performance indicators included sensitivity, specificity, positive likelihood ratio (LR+), negative likelihood ratio (LR–), positive predictive values (PPV), negative predictive values (NPV) and diagnostic OR (DOR).

Types of studies

We included studies published in peer-reviewed journals in either the English or Chinese language because the research team members are fluent in both languages. We excluded protocols, conference abstracts, reviews, commentaries, letters to the editor, oral presentations, case reports and studies for which full texts were not available.

We divided the included studies in two groups. Group 1 was the studies describing the development of the included instruments (development and validation studies); group 2 referred to the studies aiming at examining the clinical performance of the included instruments (clinical performance studies).

Assessment of methodological quality of included instruments

We used COnsensus-based Standards for the Selection of Health Measurement INstruments (COSMIN) to assess the methodological quality of the development of screening instruments.26 COSMIN assesses design quality, test quality, content validity and construct validity, using a 4-point rating scale (1=very good, 2=sufficient, 3=doubtful, 4=inadequate). Results from all studies were qualitatively summarised to determine the overall relevance, comprehensiveness, understandability and content validity of the screening instrument, with each aspect rated as adequate (+), insufficient (−), inconsistent (±) or inconclusive (?).27 Finally, the quality of the evidence was graded as ‘high’, ‘moderate’, ‘low’ or ‘very low’, using the revised Grading of Recommendations, Assessment, Development and Evaluation methodology.27

Risk of bias of the included studies

We assessed the risk of bias using the Newcastle-Ottawa Scale for observational studies and Cochrane Collaboration’s risk of bias assessment tool for randomised clinical trials (RCTs). The Newcastle-Ottawa Scale rates bias on three dimensions: selection, comparability and outcome, resulting in low (less than 5 points), medium (6–7 points) and high (8–9 points).28 The Cochrane Collaboration’s risk of bias assessment tool assesses seven possible sources of bias, including selection bias, performance bias, detection bias, attrition bias, reporting bias and other bias, and has three responses (ie, ‘High risk’, ‘Low risk’ and ‘Unclear’) for each item.29

Data extraction

Three predefined data extraction forms were used to extract information. The first table extracted the basic features of screening instruments from development and validation studies (eg, year of development, country, language, target populations, users of the instrument, numbers of items, format of responses, psychometric properties, survival time predicted by the instruments, settings for use and contents). The second table extracted information about the clinical performance studies, including study design, instrument used, setting and number of subjects. For each clinical performance study, we also developed a third table, the 2×2 table, to calculate sensitivity, specificity, LR+, LR–, PPV, NPV and DOR. The sensitivity is the ratio of correctly identified deaths (true positive results) to the total number of actual deaths (positive results). The specificity is the ratio of correctly identified non-deaths (true negative results) to the total number of non-deaths (negative results). The DOR, the ratio of disease positive rate to the non-disease positive rate (LR+ divided by LR−), ranges from 0 to infinity, with higher values indicating better discriminatory test performance.30 We also recorded the area under the curve for each instrument. If necessary, researchers contacted the authors for more data to calculate clinical performance. Data were extracted by one researcher and double-checked for accuracy by a second researcher.

Data synthesis and analysis

The key features of screening instruments were described in a narrative synthesis. We performed a meta-analysis to pool the sensitivity, specificity, LR+, LR− and DOR using Meta-Disc 2.0 and Stata 17.0. The threshold heterogeneity in clinical performance among studies for each instrument was determined by the Spearman correlation coefficient between the logarithm of sensitivity and 1−specificity. If p<0.05 of the Spearman correlation coefficients reveals heterogeneity of threshold effect, the summary receiver operating characteristics (SROC) curve was used to combine the statistics from which the specificity and sensitivity of the group of studies were obtained. The SROC curve is an integrated receiver operating characteristics curve based on the weighting of the diagnostic advantage ratio in individual diagnostic tests.31 We used I2 measures to report non-threshold effect heterogeneity in the summary estimates of diagnostic performance (I2<50% for low, 50%≤I2<75% for moderate, and I2≥75% for high).32 If significant heterogeneity existed (I2>50%), the random effects model was used; otherwise, the fixed effects model was applied. Subgroup analyses were performed by the age groups of included populations or healthcare settings (p<0.05). Publication bias was assessed using the asymmetry regression test described by Deek’s, with a symmetrical plot indicating no publication bias.33 A narrative synthesis was completed if data associated with the clinical performance of screening instruments was insufficient for meta-analysis.

Results

As shown in figure 1, a total of 11 894 articles were initially retrieved, and 4962 articles remained after removing duplicates. After title and abstract screening, 128 articles remained. Following a full-text review, 31 articles were included, reporting seven screening instruments: the Gold Standards Framework Prognostic Indicator Guidance (GSF-PIG), the RADboud indicators for PAlliative Care needs (RADPAC), the Taiwanese version Palliative Care Screening Tool (TW-PCST), the Necesidades Paliativas (Palliative Needs) (NECPAL), the Supportive and Palliative Care Indicators Tool (SPICT), Rainone and AnticiPal. Out of the 31 articles, 13 reported the development and validation of these instruments,17 18 34–44 while the other 18 articles focused on the clinical performance of the instruments.14 15 45–60 The content validity of these seven screening instruments was doubtful or inadequate because of very low to moderate quality evidence. The quality of the clinical performance articles ranged from 6.5 to 8 (online supplemental appendix S2). There was no evidence for publication bias in the funnel plot (online supplemental appendix S3).

Figure 1

Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) flow chart of the study selection.

As indicated in table 1, the majority of the studies (17 studies, 94%) were observational. Follow-up times varied across the studies. No studies on the clinical performance of AnticiPal were retrieved.

Table 1

Characteristics of the clinical performance articles

We identified six studies that reported the clinical performance of SPICT in either hospital or primary care settings with follow-up periods of 6 months or 12 months. Among studies on SPICT, one was an RCT, and the remaining five were observational studies.14 56–60 SPICT was used by a range of HCPs, including specialists, general practitioners, nurses and doctors.

Four observational studies evaluated the clinical performance of NECPAL in multiple settings (eg, hospital, nursing home and primary care centre) with follow-up periods of 12 months or 24 months. NECPAL was used by either physicians or nurses.49–51

Four observational studies reported the clinical performance of TW-PCST.52–55 These studies were conducted in hospitals settings, with follow-up periods ranging from 3 months to 12 months. TW-PCST was used by nurses in these studies.

Three observational studies used GSF-PIG to predict the12-month prognosis of patients in hospitals, with the evaluators being specialists and clinicians.45–47

A prospective observational study on Rainone was conducted in primary care settings, and patients were evaluated by family medical staff. The study did not report follow-up times.15

Development process of the screening instruments

Among the seven identified instruments, five were in the format of traditional paper-based screening instruments (SPICT, NECPAL, RADPAC, GSF-PIG and TW-PCST),16–18 39 52 and two were in the format of electronic screening instruments (AncitiPal and Rainone).15 61 Development of the traditional paper-based screening instruments involved a combination of literature review, expert consultation, focus group interviews and clinical trials. SPICT was developed in the UK through a literature review, peer expert review and prospective case finding. It is the most widely used instrument and is now available in 15 languages (eg, Italian, Thai, German, Spanish, Swedish, Danish and Indonesian).34–38 40 42 43 62 63 The development of NECPAL was based on GSF-PIG and SPICT. NECPAL has been translated into Portuguese and Chilean.41 44 RADPAC was developed through a process that included literature review, focus group discussions and a modified Rand Delphi approach. GSF-PIG was originally developed in English and translated into Italian in 2014.64 We did not find detailed development information about GSF-PIG and TW-PCST. As for the electronic instruments, AncitiPal used an iterative software development life-cycle approach and analysed retrospective cases to create a computer software algorithm for the automatic identification of individuals with palliative care needs. Rainone used electronic medical records to identify the most common factors affecting inpatient mortality and built identification criteria on that basis, but detailed development information was not reported.

Characteristics of the screening instruments

The detailed characteristics of the seven screening instruments are shown in table 2. More detail information about the instruments can be found in online supplemental appendix S4. Except for RADPAC, all instruments include both general and disease-specific palliative care indicators.15 16 18 39 52 61 All instruments are appropriate for patients with cancer, chronic obstructive pulmonary disease and congestive heart failure. All the included instruments except RADPAC are also appropriate for patients with cardiac, respiratory, neurological, liver and kidney disorders, and frailty. Three instruments (GSF-PIG, NECPAL and Rainone) include the surprise question (SQ) which asks the assessing clinician to consider likely death within a specified timeframe.15 16 18 Four instruments (GSF-PIG, NECPAL, TW-PCST and AnticiPal) include indicators related to comorbidity.16 18 52 61 Three instruments include psychological indicators (NECPAL, TW-PCST and Rainone).15 18 52 GSF-PIG and NECPAL assess the occurrence of adverse events.16 18 NECPAL considers request of palliative care service from both the patients and their families.18 GSF-PIG is the only instrument that evaluates the financial conditions of patients.16 SPICT provides recommendations for future care plans based on the results of assessment.39

Table 2

Main features of the identified screening instruments

Psychometric properties of the included instruments

The quality of evidence related to the development process and content validity was generally poor (table 3). Specific information regarding the quality of evidence can be found in online supplemental appendix S5. Only the articles of SPICT and NECPAL provided data on reliability or content validity of the instruments. Five studies reported the reliability of different versions of SPICT, with a range of 0.35–0.97 on the Kuder-Richardson formula 20,37 38 42 a Kappa range of 0.66–0.9836 38 42 and a Cronbach’s alpha of 0.84.63 The Content Validity Index (CVI) was only reported for the Italian-SPICT, which was 0.86.36 One study reported a Kappa-adjusted CVI of 0.96 for the Israeli-NECPAL.44 The reliability and content validity of TW-PCST, GSF-PIG, RADPAC and Rainone was not reported. According to the COSMIN methodology, no studies reported construct validity, cross-cultural validity assessment, measurement error and criterion validity of the identified instruments.

Table 3

Summary of psychometric properties of the identified screening instruments

Clinical performance of the included instruments

Eighteen studies reported clinical performance for five instruments (GSF-PIG, SPICT, NECPAL, TW-PCST and Rainone). Detailed information on the clinical performance of each instrument can be found in online supplemental appendix S6. We found that different studies used different positive cut-off values (scores ≥2 or ≥ 4) for TW-PCST. In the meta-analysis, the Spearman correlation coefficient of TW-PCST was not statistically significant (ρ=0.8, p=0.200>0.05), indicating that there was no heterogeneity of threshold effect in the studies for TW-PCST.

A meta-analysis of pooled sensitivity (Se), pooled specificity (Sp), LR+, LR−, area under the receiver operating characteristics curve and DOR for each instrument revealed high heterogeneity (table 4). To identify the source of heterogeneity, we conducted a subgroup analysis of sensitivity, specificity, LR+, LR− and DOR, taking into account the setting and age of the population (table 5). The results indicated a significant difference in the sensitivity and specificity of the SPICT-setting subgroup (p=0.004 < 0.05), but significant heterogeneity was not presented for the age of the population. The clinical performance of SPICT was found to be better than other instruments (NECAPAL and TW-PCST) in hospital settings. In hospital settings, SPICT demonstrated a Se of 79.8% (95% CI 72.6% to 85.5%), a Sp of 59.1% (95% CI 52.2% to 65.1%), and a pooled DOR of 5.70 (95% CI 3.51 to 9.28). In contrast, when used in non-hospital settings, SPICT exhibited a Se of 45.3% (95% CI 30.2% to 61.4%), a Sp of 97.0% (95% CI 95.7% to 97.9%) and a pooled DOR of 26.75 (95% CI 12.44 to 57.41). However, due to the limited number of studies, we were not able to identify the source of heterogeneity for NECPAL and TW-PCST.

Table 4

Details of clinical performance of the identified screening instruments

Table 5

Clinical performance of SPICT across setting and age of population

The clinical performances of GSF-PIG and Rainone were narratively summarised because of the small number of studies (table 4). The sensitivity of GSF-PIG ranged from 62.6% to 83%, specificity ranged from 22% to 91.9%, and DOR was between 1.38 and 18.85, which showed a wide variation in clinical performance. In the case of Rainone, an observational study reported a sensitivity of 94.0%, specificity of 97.0%, LR+of 31.33, LR− of 2.00 and DOR of 15.67.15

Discussion

The systematic review identified and assessed psychometric properties and the clinical performance of seven screening instruments for early identification of patients with palliative care needs. The overall methodological quality of evidence related to the reliability and validity of screening instruments ranged from very low to moderate. Notably, the quality of evidence about NECPAL was rated as low, which is consistent with findings of previous reviews.20 22 The quality of evidence regarding the content validity of SPICT was rated as ‘moderate’ in this review, but was rated as very low by Teike’s review.20 One possible reason for this discrepancy is that we included the validation studies of all versions of SPICT, but Teike’s review only included the development study of the original version of SPICT. Additionally, we found that none of the screening instruments demonstrated high clinical performance, with a Se ranging from 60.0% (poor) to 73.8% (moderate).

This review found that none of the included instruments assessed all of the physical, psychological, social and spiritual domains of care. Palliative care is described as a holistic approach to care which addresses the needs of the whole individual.65 According to the biopsychosocial-spiritual model, individuals’ relationships should be considered when assessing holistically. Illness poses a disruption to the biological relationships, which in turn impacts all the other relational aspects of a person. Genuinely holistic healthcare will address the totality of the patient’s relational existence—physical, psychological, social and spiritual concerns.65 66 Palliative care screening instruments should identify the breadth of these concerns to effectively address the corresponding needs of patients. Second, psychological distress is multifaceted, but is also reflective of the interactions of the biopsychosocial-spiritual model. Many patients with advanced non-communicable diseases experience physical suffering leading to high levels of psychological distress.67 68 Psychological distress may also stem from uncertainty about the future and fear of death.69 Third, spiritual distress including deep internal questioning and struggles is prevalent in this group of patients,70 which may frustrate attempts to treat physical and psychological symptoms and adversely affect the quality of life.67 71 Finally, the social needs of an individual are crucial as the end of life approaches.72 The need for support can affect the patient’s physical functioning, quality of life, and psychological and spiritual status throughout the course of the disease. However, the need for social and spiritual support, and to a lesser extent psychological support are often overlooked in the evaluated instruments due to a focus on physical symptoms.73 If a screening instrument only assesses one or two of these domains of care, patients may miss the optimal time to receive palliative care. To ensure timely access to palliative care that is responsive to the biopsychosocial-spiritual model, we recommend that screening instruments comprehensively assess physical, psychological, social and spiritual concerns. To ensure the integrity of the content, it is equally important to select an appropriate theoretical framework such as the biopsychosocial-spiritual model to guide the development of instruments for palliative care.

In general, screening instruments are meant to have high sensitivity and high specificity. The SPICT used in hospital settings was the best performing screening instrument. In the included articles, SPICT and NECPAL were evaluated by physicians and nurses, whereas TW-PCST was solely evaluated by nurses. However, it is unclear whether physicians are more accurate than nurses in identifying patients in need of palliative care. The sensitivity of SPICT was better than that of NECPAL in hospitals maybe because SPICT does not include the SQ. The SQ relies on HCPs’ subjective intuition and is largely influenced by patients’ disease trajectory and depends on the HCP’s skill in prognostication.74 The palliative care population involves patients with cancer and non-cancer patients, including those with frailty and dementia.5 75 The trajectories of both cancer and non-cancerous diseases are highly variable, and the prognosis is often difficult to estimate.76 77 A previous meta-analysis claimed that the ability to identify the palliative care population SQ was slightly better among patients with cancer compared with patients not diagnosed with cancer.74 Another reason is the unreliability of HCPs’ subjective judgements. The judgements are often based on clinical experience,56 but some HCPs do not have sufficient clinical experience to make reliable judgements. The use of SQ increases ‘false positives’, suggesting that a large number of patients who do not necessarily need palliative care are identified as positive.78

The clinical performance of the included instruments was assessed based on mortality/prognosis prediction. However, identifying those individuals who would benefit from palliative care should be focused on recognising unmet needs in a holistic needs-based assessment rather than on predicting the rate of physical deterioration or projected survival.23 When prognosis is used as a contributing characteristic to identify the need for palliative care, the complex and diverse disease trajectories pose difficulties for HCPs in identifying those who could potentially benefit from palliative care needs. On the other hand, use of a mortality/prognosis prediction model may encourage clinicians to focus on when to start ‘planning for death’ and may lead to delayed reviews of unmet needs and care of goals until the very last stages of a patient’s life.39 76 79 Furthermore, use of mortality/prediction as an indicator of the clinical performance could also mistakenly screen out some patients with palliative care needs.

This review emphasises that the screening instrument is intended to provide a framework for raising HCPs’ awareness of the increasing disease burden in patients and to identify patients with potential palliative care needs early. This approach motivates both patients and clinicians to consider palliative care as an option and to conduct holistic assessments at any point the need arises. It ensures that patients receive the appropriate care at the right time, according to their wishes and preferences.

This review focuses on the content, reliability, validity and clinical performance of existing screening instruments for the early identification of people in need of palliative care. Our study has identified instruments which can be applied in various settings, including hospitals, communities and homes. Because there is no uniform reference standard to evaluate whether screening instruments have the ability to identify people with genuine palliative care needs, we used sensitivity and specificity to evaluate the clinical performance of these instruments. Our study showed that the sensitivity of SPICT is the highest performing of the extant screening instruments when used within hospital settings. Palliative care has long been practised in non-hospital settings (eg, community and home), however, only a small number of studies of SPICT have reported its application in these settings.

Strengths and weaknesses

To our knowledge, this is the first comprehensive systematic review focusing on the clinical performance of screening instruments for the early identification of patients with palliative care needs. We used a comprehensive strategy to identify relevant research, including a secondary search using instrument names. We used sensitivity as a reference to compare the clinical performance of the screening instruments.

However, our study has some limitations. First, this review only included studies published in peer-reviewed journals in English or Chinese. Therefore, instruments and studies published in other languages were omitted. Second, the meta-analysis only included a small number of high-quality studies. Therefore, this result should be interpreted with caution. Subgroup analysis was only performed for SPICT based on different care settings and age groups. It is unclear whether there are other factors that can influence the clinical performance of the instrument. More studies are clearly needed to validate these instruments for the early identification of patients with palliative care needs.

Conclusion

The seven included instruments in this study have a low to moderate clinical performance and none comprehensively assess physical, psychological, social and spiritual problems. SPICT is the most commonly used instrument and has a relatively better ability to identify patients’ palliative care needs early in the hospital setting compared with other settings and other instruments. To better support early identification of palliative care patients, further work is needed to refine the content of the existing instruments to more systematically and accurately assess patients. Moreover, early identification of palliative care patients should shift from estimating when a patient will die to identifying the unmet palliative care needs.

Data availability statement

No data are available.

Ethics statementsPatient consent for publication

Not applicable.

Ethics approval

Not applicable.

Acknowledgments

The authors thank Claire E Johnson for providing guidance on the language of the article.