Clinical Indicators That Identify Risk of Deteriorating and Dying in People with a Hematological Malignancy: A Case–Control Study with Multivariable Analysis

Abstract

Background:

Identifying people who are at risk of deteriorating and dying is essential to inform goals of care, appropriate treatment decisions, patient autonomy, and effective end-of-life care. Limited literature exists on predicting survival near the end of life for people with a hematological malignancy.

Objective:

To identify the key clinical indicators that signal a person with a hematological malignancy is at high risk of deteriorating and dying.

Design, Setting, Participants:

Eleven clinical indicators identified in a Delphi approach were tested via a retrospective case–control study. Each indicator was assessed for at each in-patient admission between living (n = 236) and deceased (n = 120) people with a hematological malignancy who were admitted to a large tertiary hospital between 1st July 2014 and 31st December 2015.

Results:

Six clinical indicators were independently associated with mortality in the final three months of life: declining performance status (Odds Ratio [OR] 7.153, 95% Confidence Intervals [CI] 3.281–15.597, p = < 0.001); treatment limitations of the hematological malignancy (OR 7.855, 95% CI 3.528–17.489, p = < 0.001); relapse, refractory or persistent disease (OR 3.749, 95% CI 1.749–8.039, p = 0.001); presence of two or more comorbidities (OR 2.991, 95% CI 1.319–6.781, p = 0.009); invasive fungal infections (OR 4.887, 95% CI 1.197–19.949, p = 0.027); and persistent infections (OR 6.072, 95% CI 2.551–14.457, p = < 0.001).

Conclusions:

This study has identified six clinical indicators that signal a person with a hematological malignancy is at high risk of deteriorating and dying and may benefit from an assessment of palliative needs and proactive planning, along-side appropriate treatment.

Background

Identifying when a person is likely to be nearing the end of life is vital to facilitate appropriate treatment decisions, patient autonomy, resuscitation planning, advance care planning, and best practice care at the end of life.¹ Accurate identification of risk can allow clinicians to have honest sensitive conversations with patients and their families around the potential for deterioration and death, whilst continuing current appropriate care.^1,2 “The right conversation with the right people at the right time can enable patients and their loved ones to make the best use of the time that is left and prepare for what lies ahead”.^{3 (p. 653)} Timely recognition of risk of dying allows time for the involvement of specialized healthcare staff, such as palliative care teams or those knowledgeable in advance care planning.³

An international body of literature highlights that there are challenges providing palliative care and transitioning to end-of-life care for people with a hematological malignancy.^4,5 This is largely due to difficulties predicting survival as people with a hematological malignancy often experience a fluctuating and unpredictable illness trajectory.^4–7 Several authors have highlighted the need for greater research into prognostic variables and clinical tools to help clinicians predict survival times near the end of life for people with a hematological malignancy.^4,8

The body of literature concerning prognosticating at the end of life for people with a hematological malignancy has largely focused on people admitted to the intensive care unit or those who are still receiving aggressive treatment.⁹ To date, few studies have developed or tested prognostic models at the end of life and identifying risk of deteriorating and dying for people with a hematological malignancy for the purposes of informing palliative care provision and transitioning to end-of-life care.^10–14 There is an urgent need for more work in this area.

Objectives

This study aimed to identify clinical indicators associated with deteriorating and dying for people with a hematological malignancy and identify when these indicators first present before death. The overall aim was to identify the most appropriate time to engage in honest, sensitive discussion with patients and their families regarding possible risk of deteriorating and dying and enable greater patient autonomy and smoother transitioning to the end of life.

This study did not aim to identify when any form of palliative care should be integrated, as palliative care is recommended to be integrated at any time in the illness trajectory on a needs basis.¹⁵ Rather, this study aimed to identify when a person with a hematological malignancy is at high risk of deteriorating and dying and requires an assessment of current and potential palliative care needs in the context of possible imminent deterioration and death. Ideally, palliative care (including referral to specialist palliative care service) should occur early in the disease trajectory (including upon diagnosis) for people with serious illness, as this has been shown to be beneficial for patients and carers.^16–18

Terminology

In this study, the primary outcome was death, not the process of deteriorating and dying. However, the term “deteriorating and dying” was used to align with internationally used language around risk of death for the purposes of palliative care integration.¹ In addition, this terminology was appropriate as it is difficult to separate deterioration and death without the benefit of hindsight, particularly for people with a hematological malignancy who often experience episodes of sudden deterioration in their clinical condition, followed by periods of recovery and stability.⁷ People who experience a clinical deterioration are likely to have physical, psychosocial, or spiritual palliative care needs, regardless of the outcome of the deterioration. For these reasons, the terms deterioration and dying have been used.

Design

Study setting, design, and population

This study performed multivariable testing of 11 clinical indicators that achieved consensus by an international expert panel of hematology and palliative care clinicians as being associated with deteriorating and dying in a three-step modified Delphi approach (conducted by the authors).¹⁹ The Delphi approach was informed by a systematic review of prognostic factors present in the final three months of life for people with a hematological malignancy,⁹ and allowed for clinician expertise to inform the research topic using rigorous methodology.²⁰ The mixed method and iterative approach of the Delphi was required as limited evidence existed on this complex topic.²¹ The expert Delphi panel achieved consensus on 11 clinician-assessed indicators that were a combination of objective and subjective markers.¹⁹ These were “bedside” indicators, meaning that they were easily identifiable in every day practice by a range of healthcare professionals. The expert panel felt that the use of subjective and objective indicators was appropriate to harness the expertise of clinicians and allowed for indicators to be clinically and contextually relevant, applicable in a range of different settings, and across various disease types.¹⁹

Patients with any type of hematological malignancy were the focus of the Delphi approach, and were included in this study to optimize the application of results and not limit the scope of findings to one particular cohort. This approach is in line with other work focused on identifying risk of deteriorating and dying that includes multiple types of chronic illness.¹ The clinical indicators identified in the Delphi were developed to account for the heterogeneity in the hematological malignancies and between individuals.¹⁹ For example, the indicators were intended to cover the different illness trajectories of someone with high-risk versus low-risk disease, the elderly population with comorbidities and treatment limitations and younger people treated aggressively who may die from infection of graft versus host disease (GVHD) post stem cell transplant. This study provides preliminary testing of the indicators for people with all types of malignancies.

A retrospective case–control design was utilized as it is an efficient and effective method of testing association between an exposure and an outcome and is widely used in prognostic study designs.²² This was a single-center, hospital-based study of people with a hematological malignancy admitted to a large tertiary hospital (929 beds) in Brisbane, Australia. This cancer center cares for people with all types of hematological malignancies at any stage in the illness trajectory, including those receiving allogeneic stem cell transplantation.

Cases were selected historically (sequentially) as the last 120 deceased people who met the following criteria: (1) had been diagnosed with a hematological malignancy; (2) were ≥18 years of age at the time of death; and (3) experienced an in-patient admission at any time in the six months before study entry between 1st July 2014 and 31st December 2015. Dates of deaths of cases ranged from 28th November 2014 to 12th December 2016. Each case was individually matched with two living controls by gender and disease type. Controls were identified and matched historically (sequentially) using the same selection criteria and in the same time period (1st July 2014–31st December 2015). Cases were matched with controls by the closest time period of hospitalization. To ensure that there was clear distinction between cases and controls, only patients who were alive at least six months after the study inclusion time period were selected as living controls for the deceased cases. Disease type was classified using the World Health Classification of Hematological Neoplasms²³ and the Revised European–American classification of lymphoid neoplasms.^24,25 Classification manual available as Supplementary Table S1 (Supplementary Data are available online at www.liebertpub.com/jpm). For cases, the date of study entry was the date of death, and for controls it was the most recent admission historically before 31st December 2015. Data were collected backwards in time.

A sample size of 360 patients was determined to be feasible in consideration of time, resources, and availability of the sample.²⁶ In the absence of any reliable data to perform sample size calculations, it was arbitrarily assumed that a minimum of approximately 10% of the study sample would have the presence of any particular clinical indicator to be tested as recommended in the absence of available data.²⁷ Based on a minimum odds ratio of 2.5, type one error set at 0.05, and case/control ratio of 1:2, a sample size of 360 was estimated to result in a power of 83.68%.

Ethical approval

Low and negligible risk ethical approval was obtained from the Royal Brisbane and Women's Hospital Human Research Ethics Committee (HREC/15/QRBW/289). A waiver of consent was approved as this retrospective study was considered to be of negligible risk.

Study variables

Data were collected on every in-patient admission for the preceding six months before study entry for cases (study entry–date of death) and controls (study entry–most recent admission historically before 31st December 2015). Demographic characteristics as well as admission and discharge dates were recorded. The outcome variable was mortality (yes/no). The explanatory variables tested were demographic characteristics and the presence of 11 clinical indicators (yes/no) as identified in a modified Delphi approach study performed by the investigators.¹⁹ See Table 1 for description of clinical indicators tested. The clinical indicators were a combination of objective and subjective variables as is common in clinical tools that identify risk of dying.²⁸ As discussed in the results of the Delphi, many of the indicators overlapped slightly with each other or were interlinked,¹⁹ for example, age and comorbidities and disease status and treatment limitations. This was deemed by the expert panel to be unavoidable and clinically appropriate due to the interconnected nature of illness.¹⁹ Statistical tests were performed to ensure that the clinical indicators were independent variables. To assess for multicollinearity between the explanatory variables, the variables in the logistic regression model were run in a linear regression model with age (continuous) as the dependent variable.

Table 1.

Clinical Indicators Tested

Age Age ≥70	Yes No
Declining performance status Progressive and likely irreversible decline in performance status. Spending more than 50% of time in bed/lying down/sitting, needs help with personal cares, and/or dependence in 2 or more activities of daily living.	Yes No
Comorbidities Concurrent complex conditions (2 or more) which are progressing or advanced and likely irreversible, including: Liver disease Cardiac disease Kidney disease (including patient on renal replacement therapy) Respiratory disease Neurological disease Dementia Cancer (solid tumor diagnosis) Current conditions that significantly reduce reserve (2 or more).	Yes No
Relapse, progressive or treatment refractory disease Single relapse of high-risk/acute disease or multiple relapses of low- to moderate-risk disease (i.e., difficult to treat, rapidly accelerating, and requires aggressive treatment). Persistent, progressive anemia, neutropenia and/or thrombocytopenia, related to disease progression, (likely irreversible rather than short term response to treatment) with associated increase in transfusion or colony-stimulating factor support. Progressive liver, renal, respiratory, or cardiac dysfunction associated with disease progression. If relapsed, postallogeneic stem cell transplant.	Yes No
Persistent infections (bacterial or viral) Persistent, recurrent, and/or drug-resistant infections–including multiresistant organisms that make patient symptomatically unwell.	Yes No
Invasive fungal infections Persistent/recurrent fungal infections–unresponsive to treatment that make patient symptomatically unwell and require ongoing treatment. Particularly in context of immunocompromised patient due to progressive disease, requires ongoing active treatment, or on immunosuppressant medications.	Yes No
Refractory graft versus host disease Acute graft versus host disease that is steroid refractory. Chronic graft versus host disease requiring ongoing immunosuppression.	Yes No
Requiring high-care preadmission Need for ongoing high level of care at home or in nursing home or other high-care facility (excluding short stay for rehabilitation).	Yes No
Signs of frailty in last six months (not related to an acute cause) Incontinence Breakdown in skin integrity (pressure areas, skin tears, incontinence-associated dermatitis) Falls Recurrent infections Persistent dysphagia Delirium.	Yes No
Treatment limitations of hematological malignancy Due to inability to tolerate treatment (poor physiological reserve, age, performance status, infections, comorbidities, organ damage). No appropriate curative or life-prolonging treatment available or effective. Treatment limitations set in place by healthcare professionals or the patient/family.	Yes No
Anorexia/weight loss Anorexia and/or unexplained progressive weight loss, likely irreversible and unanticipated.	Yes No

Steps were taken to reduce risk of bias during data collection and to enhance the rigor of the approach. A data collection manual was developed with descriptions of all possible clinical scenarios and definitions around each of the variables to be collected. Notes were taken by the researcher on the characteristics associated with each clinical indicator variable for record keeping and data cleaning purposes. The data collection tool and manual were piloted on 30 patients to ensure that the study protocol was feasible and to address any unforeseen issues. As no significant changes were made to data collection, the first 30 patients were included in the final sample. Inter-rater and intrareliability testings were performed.

Study protocol

Cases and controls were identified via a list of in-patient admissions by hospital computer systems. All records were cross referenced with individual patient medical records for accuracy. Diagnoses were confirmed via pathology reports and documentation of the treating hematology medical team. Data were extracted from the electronic medical records of each patient and were recorded on a specially designed data collection tool. The presence of each clinical indicator was assessed at two time periods for each in-patient hospital admission: (1) on admission (within the first 24 hours); and (2) during admission (during the remainder of the admission), as far back as six months from study entry. The indicators age and high-care preadmission were only assessed on admission.

Inter-rater and intrarater reliability testing

Inter-rater reliability was assessed by a trained second reviewer independently assessing medical records of a random 5% of the study sample (n = 18). Intra-rater reliability testing was also performed and analysis was observed and checked by the second reviewer and another member of the research team. The researcher recollected data on a random 5% of the study sample medical records at a later point of time. The random samples were generated via the random number sequencing website www.random.org.²⁹ A kappa coefficient (K) ≥0.61 was considered to be acceptable.

Data analysis

A statistical analysis plan was agreed upon a priori. All analyses were performed in the statistical software package IBM SPSS Statistics version 23 (SPSS). A p-value of >0.05 was considered to indicate statistical significance. Descriptive statistics were computed for all study variables to describe the sample and check for missing data. Categorical/dichotomous variables were examined via frequencies and percentages. Central tendency and distribution was calculated for continuous data. Chi-square tests were used to assess differences between cases and controls for demographic characteristics and presence of clinical indicators.

Univariate logistic regression was used to test the effect of a single explanatory variable (demographic characteristics and presence of clinical indicators) on the outcome variable (mortality). Odd ratios (OR), 95% confidence intervals (CI), and p-values were reported to measure the association between each explanatory variable and mortality. A positive association was considered to be present when the odds ratio was greater than one. Conversely, a negative association was represented by an odds ratio of less than one. A large confidence interval reflected a low level of precision of the odds ratio, and a small confidence interval indicated a higher precision.³⁰ Variables with a p-value <0.2–0.05 in univariate logistic regression were included in a multivariable binomial logistic regression model. Multivariable analysis was used to identify independent predictors of mortality and accounted for the effect of multiple explanatory variables on the outcome variable. It was possible to perform multivariable logistic regression as the sample size was large enough to ensure that there were at least 10 events for each explanatory factor tested.^31,32 Backwards logistic regression modeling was manually performed in SPSS. Cox and Snell R square and Nagelkerke R square were used to assess model performance. Indicators that were significant in multivariable analysis were considered to be predictive of mortality. For the clinical indicators and disease risk (low, moderate, and high), survival curves were plotted using the Kaplan–Meier method. Significance was assessed via log rank test p-values, with p < 0.05 indicating a clinical indicator affecting survival.

To identify time periods that clinical indicators were present before death, univariate and multivariable logistic regressions were computed for five different time periods in the study going backwards in time using the same method described above: (1) between 0 and 7 days before study entry; (2) between 8 and 30 days before study entry; (3) between 31 and 90 days before study entry; (4) between 91 and 180 days before study entry; and (5) at or before 181 days before study entry. Each indicator was marked (yes/no) as if it was present at any time during a hospital admission for each of these time periods. If a patient did not have an admission in a time period, this was recorded as missing data.

Results

A total of 356 patients were included in the study. Four cases with very rare or poor prognostic disease were only able to be matched with one control each. Therefore, 120 cases were matched with 236 controls. Fourteen controls (5.9% of controls) were unable to be matched by gender and were matched on disease only. No patients were excluded due to incomplete or missing records. The median age of the sample was 62 (IQR 18, 19–91). Table 2 displays the demographic characteristics of the sample. The median time a patient was followed backwards in time from study entry was 45 days (IQR = 105, 1–219 days). The median number of admissions data collected on was 1 (IQR = 1, 1–7), and the median length of stay for each admission was nine days (IQR = 18, 1–127). There was very little missing data; relationship status and place of birth each had two missing cases (0.6%) as shown in Table 2.

Table 2.

Comparison of Baseline Characteristics and Clinical Indicators

Variables	Case (n)	Control (n)	Total (n,%)	χ²	p-value
Total sample	120	236	356	—	—
Demographic characteristics
Gender
Male	75	135	210, 59%	0.923	0.337
Female	45	101	146, 41%	0.923	0.337
Disease type
Acute leukemia	44	88	132, 37.1%	0.09	1.000
Myeloproliferative neoplasms	4	8	12, 3.4%
Myelodysplastic syndromes	9	18	27, 7.6%
Chronic leukemia	9	16	25, 7.0%
Multiple myeloma	22	44	66, 18.5%
High-risk lymphoma	3	6	9, 2.5%
Mod-risk lymphoma	24	46	70, 19.7%
Low-risk lymphoma	5	10	15, 4.2%
Disease risk
Low risk	27	52	79, 22.2%	0.02	0.991
Moderate risk	46	90	136, 38.2%
High risk	47	94	141, 39.6%
Country of birth
Australia	87	172	259, 66.0%	0.03	0.0865
Outside Australia	31	64	95, 26.7%	0.03	0.0865
Missing	0	2	2, 0.6%
Relationship status
Married/de facto	73	162	235, 66%	2.04	0.153
Divorced/separated/single	46	73	119, 33.5%	2.04	0.153
Missing	0	2	2, 0.6%
Residence classification
Major urban area	54	105	159, 44.7%	0.49	0.781
Urban area	61	117	178, 50.%
Rural area or other	5	14	19, 5.3%
Clinical indicators present
Age ≥70
Yes	48	34	82, 23.%	29.39	<0.001
No	72	202	274, 77.%	29.39	<0.001
Declining performance status
Yes	78	20	98, 27.5%	127.41	<0.001
No	42	216	258, 72.5%	127.41	<0.001
Comorbidities (≥2)
Yes	51	26	77, 21.6%	45.52	<0.001
No	69	210	279, 78.4%	45.52	<0.001
Relapsed, refractory, or progressive disease
Yes	79	45	124, 34.8%	76.65	<0.001
No	41	191	232, 65.2%	76.65	<0.001
Persistent infections (viral or bacterial)
Yes	50	18	68, 19.1%	59.65	<0.001
No	70	218	288, 80.9%	59.65	<0.001
Invasive fungal infections
Yes	16	5	21, 5.9%	18.02	<0.001
No	104	231	335, 94.1%	18.02	<0.001
Refractory GvHD
Yes	22	21	43, 12.1%	6.67	0.010
No	98	215	313, 87.9%	6.67	0.010
High care preadmission
Yes	46	14	60, 16.9%	59.60	<0.001
No	74	222	296, 83.1%	59.60	<0.001
Signs of frailty
Yes	68	50	118, 33.1%	45.20	<0.001
No	52	186	238, 66.9%	45.20	<0.001
Treatment limitations of hematological malignancy
Yes	84	18	102, 28.7%	151.39	<0.001
No	36	218	254, 71.3%	151.39	<0.001
Anorexia/weight loss
Yes	53	35	88, 24.7%	36.79	<0.001
No	67	201	268, 75.3%	36.79	<0.001

Univariate comparisons of variables between cases and controls were by Pearson's Chi-square.

Inter-rater reliability testing

Overall, there was an acceptable level of consistency (K > 0.61) for inter- and intrarater reliability. There was substantial agreement (K 0.61–0.80) or almost perfect agreement (K 0.81–1.0) for inter-rater reliability for nine of the clinical indicators. The indicator fungal infections achieved moderate agreement (K 0.61–0.80). Fungal infection was not a commonly found indicator; therefore, any differences between the two assessors may have caused a disproportionately large change in the K score compared with more frequently observed indicators. There was substantial or almost perfect agreement for intrarater reliability for 10 of the indicators. There were not enough observations to calculate an inter-rater reliability score for graft versus host disease (GvHD) and intrarater reliability score for fungal infection.

Clinician indicators that identify risk of deteriorating and dying

None of the demographic variables was statistically different between cases and controls in Chi-square tests. All clinical indicators were significantly different between cases and controls in Chi-square tests (Table 2). None of the demographic characteristics was significantly associated with mortality in univariate logistic regression. However, the variable relationship status achieved a p-value that met criteria to be included in multivariable logistic regression (OR 1.40, 95% CI 0.88–2.22, p = 0.154). All the clinical indicators were significantly associated with mortality in univariate logistic regression. The clinical indicators that were most predictive of death in univariate analysis were treatment limitations (OR 28.26, 95% CI 15.21–52.49, p = < 0.001) and declining performance status (OR 20.06, 95% CI 11.095–36.26, p = < 0.001). See Table 3 for results of univariate logistic regression for clinical indicators.

Table 3.

Results of Univariate Logistic Regression for Baseline Characteristics and Clinical Indicators

Baseline variables	OR	95% CI	p	Overall p
Gender
Female	—	—	0.337	0.337
Disease type
Acute leukemia	—	—	1.00	1.00
Myeloproliferative neoplasms	—	—	1.00
Myelodysplastic syndromes	—	—	1.00
Chronic leukemia	—	—	0.796
Multiple myeloma	—	—	1.00
High-risk lymphoma	—	—	1.00
Mod-risk lymphoma	—	—	0.892
Low-risk lymphoma	—	—	1.00
Disease risk
Low risk	—	—	0.991	0.991
Moderate risk	—	—	0.931
High risk	—	—	0.899
Country of birth				0.865
Other than Australia	—	—	0.865
Relationship status
Divorced/separated/single	1.40	0.88–2.22	0.154	0.154
Residence classification
Major urban area	—	—	0.783	0.773
Urban area	—	—	0.953
Rural area or other	—	—	0.505

Clinical indicator variables	OR	95% CI	P
Treatment limitations of hematological malignancy
Yes	28.259	15.213–52.493	<0.001
Declining performance status
Yes	20.057	11.095–36.259	<0.001
High-care preadmission
Yes	9.857	5.128–18.949	<0.001
Persistent infections (bacterial or viral)
Yes	8.651	4.737–15.799	<0.001
Relapsed, refractory, progressive disease
Yes	8.178	4.971–13.454	<0.001
Invasive fungal infections
Yes	7.108	2.536–19.919	<0.001
Comorbidities ≥2
Yes	5.970	3.462–10.296	<0.001
Signs of frailty
Yes	4.865	3.018–7.841	<0.001
Weight loss
Yes	4.543	2.732–7.555	<0.001
Age 70 years or over
Yes	3.961	2.366–6.630	<0.001
Refractory GvHD
Yes	2.298	1.207–4.376	0.011

OR, odds ratio; CI, confidence interval.

In multivariable logistic regression analysis, the following variables were independent predictors of mortality if they were present at any time in the six-month study period: declining performance status; presence of two or more comorbidities; relapse, refractory, or persistent disease; persistent infections (bacterial or viral); invasive fungal infections; and treatment limitations. See Table 4 for results of multivariable analysis. The remaining variables tested were excluded from the model as they were not significant when accounting for the other factors. The model containing the six predictive clinical indicators was statistically significant (n = 356, χ² = 245.265, df = 6, p = < 0.001). The model as a whole explained between 50% (Cox and Snell R square 0.498) to 69% (Nagelkerke R square 0.690) of the variance between living and deceased patients and correctly classified 88.5% of cases. These variables described 49%–68% of the variation between deceased cases and living controls. Collinearity statistics were computed and were satisfactory for all the included variables as they were all under the value of three.³⁰

Table 4.

Results of Multivariable Logistic Regression

	OR	95% CI	p-value
Variables in the model
Treatment limitations of hematological malignancy
Yes	7.855	3.528–17.489	<0.001
Declining performance status
Yes	7.153	3.281–15.597	<0.001
Persistent infections (bacterial and viral)
Yes	6.072	2.551–14.457	<0.001
Invasive fungal infections
Yes	4.887	1.197–19.949	0.027
Relapsed, refractory, progressive disease
Yes	3.749	1.749–8.039	0.001
Comorbidities ≥2
Yes	2.991	1.319–6.781	0.009
Variables excluded from the model
Relationship status
Yes	—	—	0.338
Age 70 or over
Yes	—	—	0.104
Refractory GvHD
Yes	—	—	0.157
Signs of frailty
Yes	—	—	0.111
High care
Yes	—	—	0.462
Weight loss
Yes	—	—	0.548

Kaplan–Meier analysis using the log rank test identified that there was a statistically significant association between survival and the following: (1) declining performance (log rank test p < 0.001); (2) two or more comorbidities (log rank test p < 0.001); (3) relapsed, refractory or progressive disease (log rank test p < 0.001); (4) persistent infections (log rank test p < 0.001); (5) treatment limitations (log rank test p < 0.001; (6) high care (log rank test p < 0.001); (7) signs of frailty (log rank test p < 0.001); (8) refractory graft versus host disease (log rank test p < 0.005); and (9) weight loss (log tank test p < 0.027). There was no significant difference in the time to survival between people with and without invasive fungal infections (log rank test p = 0.574). Survival curves are displayed in the Supplementary Figures S1–S10. Kaplan–Meier analysis using the log rank test indicated that there was no difference in time to survival between people with different risk categories of disease (p = 0.063): low, moderate, and high. See Supplementary Figure S11.

Due to the relatively small sample size, we were unable to perform multivariable logistic regression modeling for each disease type or category of risk. To assess if there were differences between clinical indicators present among patients with low-, moderate-, and high-risk diseases, Chi-square tests were calculated for the six clinical indicators that were included in the multivariable logistic regression model, stratified according to disease risk. Declining performance was significantly different in the cases in the high-risk group, compared with the moderate- and low-risk groups (44.5%, 76.1%, 81.5%, respectively, χ² = 127.40, p = < 0.001). There was no clinically significant difference among low-, moderate-, and high-risk diseases for the presence of relapsed, refractory disease status, comorbidities, and infections in the deceased cases. There were greater fungal infections present in deceased cases with high-risk disease versus those with moderate- and low-risk disease (19.1%, 8.7%, 11.1%, respectively, χ² = 18.02, p = < 0.001). There were also fewer treatment limitations in the deceased cases with high-risk disease versus those with moderate- or low-risk disease (66%, 73.9%, 70.4%, respectively, χ² = 151.39, p = < 0.001).

When clinical indicators are first present before death

Descriptive statistics for when the clinical indicators were first present are shown in Table 5, including if they were present for the first time (before study entry) on or during an admission, and the median time before death for cases. All the clinical indicators were more commonly identified for the first time (before study entry) on admission with the exception of persistent infections and invasive fungal infections. The indicator high-care needs preadmission was only measured on admission. When univariate and logistic regression was computed for if a clinical indicator was present in cases and controls (yes/no) in five different time periods, five clinical indicators were consistently significant in multivariate logistic regression if they were present in the 0–7 days, 8–30 days, and 31–90 days before study entry, these included the following: (1) declining performance status; (2) treatment limitations; (3) relapsed, refractory, or persistent disease; (4) persistent infections; and (5) two or more comorbidities. Fungal infection was present in the overall model (any time in six months before study entry) and one of the time-dependent models (0–7 days before study entry). As the models went further back in time from study entry, they accounted for less variance between deceased cases and living controls. All the indicators in the first three time models (0–7, 8–30, and 31–90 days before study entry) had considerably high odds ratios, indicating a good magnitude of effect on the dependent variable. Table 6 provides a summary of logistic regression models for all the time periods, including if clinical indicators were present at any time in the six months before study entry.

Table 5.

Time Frames When Clinical Indicators Were First Present

Clinical indicator	Not present at all n (%)	First present on admission n (%)	First present during admission n (%)	Median days present before death ^a (min–max)
Declining performance status	258 (72.5)	60 (16.9)	30 (10.7)	25 (0–198)
Comorbidities ≥2	279 (78.4)	70 (19.7)	7 (2)	49 (1–206)
Relapsed, refractory, progressive disease	232 (65.2)	104 (29.2)	20 (5.6)	48 (1–219)
Persistent infections	288 (80.9)	12 (3.4)	56 (15.7)	21 (0–188)
Invasive fungal infections	335 (94.1)	2 (0.6)	19 (5.3)	24 (0–187)
Refractory GvHD	313 (87.9)	33 (9.3)	10 (2.8)	20 (1–157)
Signs of frailty	238 (66.9)	87 (24.4)	31 (8.7)	41 (0–206)
High care (on admission only)	296 (83.1)	60 (16.9)	N/A	26 (1–198)
Treatment limitations	254 (71.3)	55 (15.4)	47 (13.2)	26 (0–188)
Weight loss	268 (75.3)	62 (17.4)	26 (7.3)	18 (1–183)

Time clinical indicator present for cases only.

Table 6.

Summary of Logistic Regression Models for Different Time Periods

Time periods	Clinical indicators (present in time period) significant in logistic regression model	OR	95% CI	p-value	Variation explained by model		Overall p-value
Time periods		OR	95% CI	p-value	Cox and Snell R squared	Nagelk-erke R-squared	Overall p-value
0–7 days before study entry	Invasive fungal infections	11.566	2.810–47.757	0.001	49%	68%	<0.001
	Declining performance status	8.155	3.675–18.094	<0.001
	Treatment limitations	7.589	3.443–16.724	<0.001
	Relapsed, refractory, or persistent disease	4.418	2.069–9.432	<0.001
	Persistent infections	3.182	1.292–7.837	0.012
	Comorbidities ≥2	2.935	1.297–6.644	0.010
8–30 days before study entry	Treatment limitations	7.384	3.165–17.226	<0.001	42%	58%	<0.001
	Declining performance status	4.200	1.850–9.537	0.001
	Comorbidities ≥2	3.241	1.514–6.938	0.002
	Relapsed, refractory or persistent disease	3.113	1.5214–6.355	0.002
	Persistent infections	2.743	1.146–6.569	0.024
	Weight loss	2.333	1.024–5.315	0.044
31–90 days before study entry	Treatment limitations	5.719	2.144–15.256	<0.001	22%	31%	<0.001
	Persistent infections	3.316	1.196–9.196	0.021
	Declining performance status	3.315	1.339–8.206	0.010
	Comorbidities ≥2	2.593	1.115–6.027	0.027
	Relapsed, refractory, or persistent disease	2.229	1.097–4.526	0.027
91–180 days before study entry	Treatment limitations	7.002	1.419–34.558	0.017	14%	29%	<0.001
	Comorbidities ≥2	5.157	2.002–13.283	0.001
	Relapsed, refractory, or persistent disease	4.027	1.719–9.438	0.001
181 or more days before study entry	Comorbidities ≥2	10.217	1.180–88.471	0.035	18%	25%	0.011
Any time in six months before study entry	Treatment limitations	7.855	3.528–17.489	<0.001	50%	69%	<0.001
	Declining performance status	7.153	3.281–15.597	<0.001
	Persistent infections	6.072	2.551–14.457	<0.001
	Invasive fungal infections	4.887	1.197–19.949	0.027
	Relapsed, refractory, progressive disease	3.749	1.749–8.039	0.001
	Comorbidities ≥2	2.991	1.319–6.781	0.009

Cox and Snell R squared and Nagelkerke R-squared.

N.B. For cases, the date of study entry was the date of death, and for controls, it was the most recent admission historically before 31st December 2015. Data were collected backwards in time.

The assumptions of logistic regression were met for this study. There were 11 predictors tested and 120 cases, meaning there were enough events per predictor tested.^33,34 Alternatively, 15 cases are needed for each variable tested, which was also achieved in this study.³⁰ The sample had no outlying cases as the variables were categorical. There were no issues with collinearity between variables in any of the models. All the included variables had a variance inflation factor reading of under three.³⁰

Discussion

Six clinical indicators were found to independently predict death in the final three months of life for people with a hematological malignancy. Treatment limitations and declining performance status were the strongest indicators of mortality. The prognostic value of performance status is substantiated by the few studies that have explored predicting survival at the end of life for people with a hematological malignancy. Chou et al.¹³ reported that the Palliative Prognostic Index (PPI) accurately predicted survival in terminally ill people with a hematological malignancy known to a specialist palliative care service (median survival of 16 days, IQR 4–4.75 days). Ohno et al.¹² reported the PPI was predictive of death in people with a hematological malignancy admitted to an in-patient hematology unit (median survival of 6.4 weeks). Although declining performance status was a strong predictor of mortality in this study, only 27.5% (n = 98) of patients had this indicator present at some time during the six months study period. Similarly, only 23.6% (n = 77) of the sample had comorbidities ≥2. This may be due to the study sample being in-patients in a large tertiary hospital. It is possible that many patients were receiving aggressive treatments and died suddenly of complications. In addition, patients with poor performance or comorbidities, who are not eligible for aggressive treatment, may remain in rural or regional areas and may be underrepresented in the study sample.

The indicator treatment limitations (of the underlying malignancy) was found to be a strong predictor of mortality and was defined as: (1) inability to tolerate treatment; (2) no appropriate curative or life-prolonging treatment available or effective; or (3) limitations set in place by the healthcare team or family. The appropriateness of these indicators seems apparent considering that disease progression is often the leading cause of death for people with a hematological malignancy. This subjective definition is that this indicator is clinically relevant in the age of targeted therapies and advances in available treatments.

The core components of the PPI include performance status, dyspnea, delirium, oral intake, and edema.¹³ Ohno et al.¹² also found that a prognostic model developed by Kripp et al.,¹¹ which included low performance status, requiring opioid analgesia, low platelet count, low albumin levels, and high lactate dehydrogenase levels, was predictive of death for people with a hematological malignancy on a palliative care unit (median survival of 58 days). With the exception of performance status, the range of prognostic factors mentioned above were not assessed in our study as clinical indicators that achieved consensus in the Delphi approach (that informed this study) were focused on earlier markers of deteriorating (ref–under review). This was to facilitate providing proactive palliative care planning before acute deterioration. Differences in our results and the Chou et al.,¹³ Ohno et al.,¹² and Kripp et al.¹¹ are likely to be due to these studies having shorter time frames of survival in their populations than our study. To the best of our knowledge, the remaining indicators identified in our study have not been assessed in any other prognostic study designed to improve palliative care provision and transitioning to end-of-life care. This study significantly adds to the body of knowledge as it is the first of its kind to identify clinical indicators that signal risk of deteriorating and dying up to three months before death.

The findings of this study indicate that identifying risk of deteriorating and dying with a core set of clinical indicators is most feasible within the final three months of life for people with a hematological malignancy. This is a shorter time frame than used in most clinical tools that identify risk of deteriorating and dying within 6–12 months in other populations.²⁸ This is not surprising considering the rapid speed, in which people with a hematological malignancy can deteriorate.⁵ While most indicators were present first on admission, persistent infections and invasive fungal infections were more commonly found for the first time during admission. Therefore, an assessment of the clinical indicators should be performed on admission and repeated at other clinically relevant time points.

This study had several limitations, including a relatively small sample size, leading to wide confidence intervals. The magnitude of the effect, statistical significance, and clinical significance of results were considered in the interpretation of findings to counteract this limitation. The study was set in a single center, thus limiting external validity. However, the hospital sample is likely to share many characteristics with other tertiary centers, increasing generalizability of results. A strength associated with collecting data at a single site is the uniformity of medical records, which enhanced data collection procedures. The retrospective nature of this study was a limitation as medical records are often not completely accurate or detailed. The researcher took due diligence to cross-reference all medical records for accuracy and completeness where possible. Prospective testing of the clinical indicators is required in a larger multi-site study, with exploration into different disease types (i.e., low-risk versus high-risk disease) to assess if there are differences in clinical indicators for different patient groups. Despite these limitations, this study has provided valuable preliminary data and adequately addressed the study aims.

Findings of this study can help health professionals internationally identify the most appropriate time to engage in honest, sensitive discussion with patients and their families regarding possible risk of deteriorating and dying. This will enable a collaborative reassessment of the goals of care and planning for potential need. Ultimately, the findings of this study will help to give patients more control over their death and the time they have remaining.

Conclusion

This study identified six clinical indicators associated with deteriorating and dying for people with a hematological malignancy in a multivariable logistic regression model. Future research is needed to test these findings further. However, this study has begun to address a gap in knowledge and informed palliative care provision and transitioning to the end of life for people with a hematological malignancy.

Footnotes

Acknowledgments

The lead author was supported by the following organizations: (1) National Health and Medical Research Council, Centre for Research Excellence in End-of-Life Care, Australia; (2) Queensland University of Technology, Australia; and (3) Royal Brisbane and Women's Hospital Foundation, Australia and the Centaur Memorial Fund for Nurses. These organizations were not involved in the conduct of the study or preparing the article.

Author Disclosure Statement

No competing financial interests exist.

References

Highet

, Crawford

, Murray

, et al.: Development and evaluation of the Supportive and Palliative Care Indicators Tool (SPICT): A mixed method study. Support Palliat Care, 2013; 1:1–6.

Dalgaard

, Thorsell

, Delmar

: Identifying transitions in terminal illness trajectories: A critical factor in hospital-based palliative care. Int J Palliat Med, 2010; 16:87–92.

Barclay

, Maher

: Having the difficult conversations. Br Med J, 2010; 341:653–655.

Moreno-Alonso

, Porta-Sales , Monforte-Royo

, et al.: Palliative care in patients with haematological neoplasms: An integrative systematic review. Palliat Med, 2018; 32:79–105.

Manitta

, Philip

, Cole-Sinclair

: Palliative care and the hemato-oncological patient: Can we live together? A review of the literature. J Palliat Med, 2010; 13:1021–1025.

Hung

Y-S

, Wu

J-H

, Chang

, et al.: Characteristics of patients with hematologic malignancies who received palliative care consultation services in a medical center. Am J Hosp Palliat Med, 2013; 30:773–780.

McGrath

, Holewa

: Special considerations for haematology patients in relation to end-of-life care: Australian findings. Eur J Cancer Care, 2007; 16:164–171.

Auret

, Bulsara

, Joske

: Australasian haematologist referral patterns to palliative care: Lack of consensus on when and why. Intern Med J, 2003; 33:566–571.

Button

, Chan

, Chambers

, et al.: A systmatic review of prognostic factors in the final three months of life for popele with a hematological malignancy. BMC Cancer, 2017; 17:213.

10.

Corbett

, Johnstone

, McCracken Trauer

, et al.: Palliative care and hematological malignancies: Increased referrals at a comprehensive cancer centre. J Palliat Med, 2013; 16:537–542.

11.

Kripp

, Willer

, Schmidt

, et al.: Patients with malignant hematological disorders treated on a palliative care unit: Prognostic impact of clinical factors. Ann Hematol, 2014; 93:317–325.

12.

Ohno

, Abe

, Sasaki

, et al.: Validation of 2 prognostic models in hospitalized patients with advanced hematological malignancies in Japan. Am J Hosp Palliat Med, 2017; 34:258–262.

13.

Chou

W-C

, Kao

C-Y

, Wang

P-N

, et al.: The application of the palliative prognostic index, charlson comorbidity index, and Glasgow prognostic score in predicting the life expectancy of patients with hematologic malignancies under palliative care. BMC Palliat Care, 2015; 14:1–7.

14.

Button

, Chan

, Chambers

, et al.: Lack of palliative mortality prediction tools for people with a hematological malignancy: Call to action. J Palliat Med, 2017; 20:802–803.

15.

NICE:. Improving Supportive and Palliative Care for Adults Living with Cancer. The Manual. London: National Institute for Clinical Excellence, 2004.

16.

Thienprayoon

, LeBlanc

: Early integration of palliative care into the care of patients with cancer. ASH Educ Program Book, 2015; 2015:479–483.

17.

Temel

, Greer

, Admane

, et al.: Longitudinal perceptions of prognosis and goals of therapy in patients with metastatic non–small-cell lung cancer: Results of a randomized study of early palliative care. J Clin Oncol, 2011; 29:2319–2326.

18.

Temel

, Greer

, Muzikansky

, et al.: Early palliative care for patients with metastatic non–small-cell lung cancer. N Engl J Med, 2010; 363:733–742.

19.

Button

, Gavin

, Chan

, et al.: Harnessing the power of clinician judgement regarding risk of deteriorating and dying people with a hematological malignancy: A Delphi study. J Advanced Nurs, 2018; (in press).

20.

Hasson

, Keeney

, McKenna

: Research guidelines for the Delphi survey technique. J Adv Nurs, 2000; 32:1008–1015.

21.

Keeney

, Hasson

, McKenna

: A critical review of the Delphi technique as a research methodology for nursing. Int J Nurs Stud, 2001; 38:195–200.

22.

Gordis

: Epidemiology. 5th Edition ed. Philadelphia: Elsevier Saunders, 2014.

23.

Swerdlow

, Campo

, Harris

, et al.: WHO Classification of Tumours of the Haematopoietic and Lymphoid Tissues. Lyon: IARC, 2008.

24.

Hiddemann

, Longo

, Coiffier

, et al.: Lymphoma classification-the gap between biology and clinical management is closing. Blood, 1996; 88:4085–4089.

25.

Harris

, Jaffe

, Diebold

, et al.: Lymphoma classification–from controversy to consensus: The REAL and WHO Classification of lymphoid neoplasms. Ann Oncol, 2000; 11:S3–S10.

26.

Hertzog

: Considerations in determining sample size for pilot studies. Res Nurs Health, 2008; 31:180–191.

27.

Noordzij

, Tripepi

, Dekker

, et al.: Sample size calculations: Basic principles and common pitfalls. Nephrol Dial Transplant, 2010; 25:1388–1393.

28.

Maas

, Murray

, Engels

, et al.: What tools are available to identify patietns with palliative care needs in primary care: A systematic literature review and survey of European practice. Support Palliat Care, 2013; 3:444–451.

29.

RANDOM.ORG: True random number service, www.random.org/ (Last accessed June, 2016).

30.

Field

: Discovering Statistics Using IBM SPSS Statistics: And Sex and Drugs and Rock 'n' Roll. London: Sage Publications, 2013.

31.

Hauser

, Stockler

, Tattersall

: Prognostic factors in patients with recently diagnosed incurable cancer: A systematic review. Support Care Cancer, 2006; 14:999–1011.

32.

Pallant

: SPSS Survival Manual. United Kingdom: McGraw-Hill Education, 2013.

33.

Harrell

, Lee

, Califf

, et al.: Regression modelling strategies for improved prognostic prediction. Stat Med, 1984; 3:143–152.

34.

Hulley

, Cummings

, Browner

, et al.: Designing Clinical Research. Lippincott Williams & Wilkins, 2013.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.31 MB

0.00 MB