Diagnosing serious infections in older adults presenting to ambulatory care: a systematic review

Abstract

Objectives

to summarise all available evidence on the accuracy of clinical features and blood tests for diagnosing serious infections in older patients presenting to ambulatory care.

Methods

systematic review, searching seven databases using a comprehensive search strategy. We included cross-sectional prospective diagnostic studies on (1) clinical features, (2) diagnostic prediction rules based on clinical features alone, (3) blood tests and (4) diagnostic prediction rules combining clinical features and blood tests. Study participants had to be community-dwelling adults aged ≥65 years, in whom a physician suspected an infection. We used QUADAS-2 to assess risk of bias. We calculated measures of diagnostic accuracy and present descriptive statistics.

Results

out of 13,757 unique articles, only six studies with a moderate to high risk of bias were included. There was substantial clinical heterogeneity across these studies. Clinical features had LR− ≥0.61 and LR+ ≤4.94. Twelve prediction rules using clinical features had LR− ≥0.30 and LR+ ≤2.78. There was evidence on four blood tests of which procalcitonin was the most often investigated: levels <0.37 ng/ml (LR− = 0.20; 95%CI 0.10–0.42) were suitable to rule out sepsis in moderately high prevalence situations. Two diagnostic prediction rules combining clinical features and procalcitonin had LR− of ≤0.12 (95%CI 0.05–0.33) and LR+ of maximum 1.39 (95%CI 1.30–1.49).

Conclusions

we found few studies on the diagnostic accuracy of clinical features and blood tests to detect serious infections in older people presenting to ambulatory care. The risk of bias was mostly moderate to high, leading to substantial uncertainty.

biomarkers, prediction rules, diagnostic tests, older adults, signs and symptoms, systematic review

Key points

• Evidence on the diagnosis of serious infections in older people presenting to ambulatory care is scarce.

• Some studies suggest that clinical features alone or combined in a prediction rule cannot safely rule out a serious infection.

• Some blood biomarkers at low thresholds could be useful to rule out sepsis.

Introduction

Acute infections pose a substantial burden on older adults [1]. Compared to people aged 30–50, mortality caused by acute infections increases more than fiftyfold in persons aged 65 years or older (up to 363 deaths/100.000 person-years), and they have a four times higher risk of being admitted to hospital for an acute infection compared to the general population [2,3]. Older persons who survive a serious infection may suffer from functional decline afterwards with a subsequent increase of their level of dependency [4–6]. Pneumonia, for example, leads to an increased long-term morbidity and mortality following hospitalisation, with over 70% of surviving patients being readmitted to hospital at least once within the next 3 years [7], which is much higher than the 17% annual hospitalisation rate in people over 75 years of age who did not suffer from pneumonia [8]. Risk of mortality and morbidity after pneumonia are strongly correlated with severity upon admission, suggesting earlier detection may potentially improve prognosis [7].

Reasons for the increased vulnerability to infections in older adults include, amongst others, the presence of chronic conditions such as diabetes and heart failure, the age-related decline of host defences (immunosenescence), the lower physiologic reserve, a higher prevalence of polypharmacy as a result of multimorbidity and decreased mobility [4,9]. In addition, the severity of an infection is determined by the causative pathogen and by the health status of the patient [4]. This general concept is especially true in older patients. Due to the decline in host defences, any infection may be associated with a high risk of complications or mortality in this population. An influenza infection for example may be benign and self-limiting but it can also result in complications and death or require hospital admission in a more vulnerable older patient.

Especially in ambulatory care, diagnosing a serious infection is hampered by the relative absence of typical symptoms, as these have not yet developed in the earlier stages of the illness. For example, patients with pneumonia may not present with cough, fever, chest discomfort or sputum production [10]. Diagnosis of meningitis may be complicated by the lack of neck stiffness [11], and peritoneal signs may be absent in case of an intra-abdominal infection [12]. This lack of clear signs and symptoms is even more problematic in older persons, who may also present with atypical symptoms such as confusion, falls, delirium or a sudden decline in functional status [4,13]. Overlap with existing symptoms from comorbidities, such as dyspnoea from COPD or heart failure, further complicates the diagnostic assessment.

Furthermore, clinicians in ambulatory care, including general practice, walk-in clinics, geriatric assessment units and emergency departments (EDs), mostly have to rely on clinical features, either separately or in combination, with or without readily available biomarkers such as blood tests. Decisions taken in ambulatory care include referring the patient on to the next level up, starting treatment in ambulatory care or reassurance with appropriate safety netting. Ambulatory care clinicians have to balance the risk of rapidly evolving infections with the risk of hospital admission and unnecessary treatment in a vulnerable population. Ultimately, safely ruling out a serious infection remains the priority in this setting. This systematic review aims to identify and summarise all available evidence on the accuracy of clinical features and blood tests for diagnosing serious infections in older patients.

Methods

The protocol was registered a priori on PROSPERO (CRD42019120442). We report this study in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines (PRISMA, checklist available as Appendix 1).

Search strategy and selection criteria

The MEDLINE, EMBASE, DARE, HTA, CINAHL, Web of Science and Cochrane library databases were searched for articles on the diagnosis of serious infections in older patients. The first search was undertaken from inception to January 2019 and was updated on 14 October 2019. The search strategy, developed in collaboration with a clinical librarian, included MeSH/Emtree terms and free text (Appendix 2). Additionally, we consulted experts in geriatrics to check for any omissions and employed snowballing by checking references of included studies and related reviews, as well as references of relevant institutional guidelines.

Selection of studies was done by five reviewers (TS, HB, AS, JV, AVdB) using selection criteria that were defined a priori. The selection process was piloted on a sample of 50 studies—no changes were required at this stage. Selection was performed in two rounds: first on title and abstract, then on full text. Conflicts were resolved by two independent reviewers (AvdP, JV). We used the Covidence online software for the deduplication and study selection (Covidence systematic review software, Veritas Health Innovation, Melbourne, Australia).

Inclusion criteria

We included cross-sectional prospective diagnostic accuracy studies on clinical features and blood tests for diagnosing serious infections in older adults presenting to an ambulatory care setting. Ambulatory care settings included general practice, primary care out-of-hours service, walk-in clinics, interface clinics, EDs and other settings in which patients were not referred by a primary care professional prior to inclusion. Study participants had to be community-dwelling adults aged 65 years and older, in whom a physician suspected an infection. Studies that included adults of all ages were eligible when at least 50% was 65 years old or above or when they reported results for these older adults separately.

We selected studies examining the following index tests: (1) clinical features (patient characteristics, signs and symptoms), (2) diagnostic prediction rules with clinical features, (3) blood tests and (4) diagnostic prediction rules combining clinical features and blood tests. The following target diseases were considered as serious infections: complicated urinary tract infections requiring hospital admission, influenza infection requiring hospital admission, sepsis, infectious encephalitis or meningitis, osteomyelitis, prosthetic joint infections, septic (infectious) arthritis, infective endocarditis, infectious cholecystitis, infectious spondylitis, appendicitis, cellulitis requiring hospital admission, complicated diverticulitis, pneumonia for which a chest radiography has been taken or of which the patient has died and severe acute exacerbation of chronic obstructive pulmonary disease resulting in hospital admission or visit to the emergency room.

Exclusion criteria

Given the low incidence of some target conditions, we excluded studies reporting a total sample size smaller than 50 participants, because such studies would include only a few subjects with the target condition and would be prone to selection bias [14,15]. Consequently, case reports were excluded as well. Case series, retrospective studies, conference abstracts, systematic reviews (which served as a reference) and case-control studies were also excluded, whereas nested case-control studies in which cases and controls were sampled from the same source population were eligible. Studies including people living in nursing homes were not included in this review, as they investigate a different study population. We also excluded studies investigating non-serious infections or urinary tests. We did not apply language restrictions.

Quality assessment

We used QUADAS-2 to assess risk of bias [16]. Quality assessment was performed by one reviewer (TS) and checked by a second reviewer (AS), using the Review Manager software (RevMan, Version 5.3. Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014). Disagreements were discussed and resolved in a consensus meeting (TS, HB, JV, AVdB).

Data extraction and analysis

Data were extracted by one reviewer (TS) and checked by a second reviewer (AS). Any discrepancies were discussed and corrected. We reconstructed two-by-two tables based on the study data and data provided by authors upon request by e-mail. We calculated measures of diagnostic accuracy, using the RevMan software (Review Manager [Computer program]. Version 5.3. Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014.). Our protocol stated that data would be pooled when at least four studies without important clinical heterogeneity were available for a particular outcome, using a bivariate model and displaying results in a hierarchical summary ROC curve for continuous test results. Given the low number of studies identified, we had to restrict our analyses to descriptive statistics. We calculated the positive and negative likelihood ratio (LR) for each test. Confidence intervals were calculated using the log method described by Altman [17]. Pretest and post-test probabilities of serious infections were calculated and presented in Dumbbell plots and a probability plot, using SAS software, Version 9.4 (SAS Institute Inc., Cary, NC, USA). We considered tests to be useful in ruling out a serious infection in ambulatory care if their negative likelihood ratio (LR−) was lower than 0.20; conversely diagnostic tests were considered useful as ‘red flags’ for serious infections when their positive likelihood ratio (LR+) was 5.0 or higher [18,19].

Results

We identified 13,757 unique studies of which 184 were selected for full-text screening. Ultimately, six studies were included (Figure 1) [20–25]. Study characteristics are shown in Table 1. The target condition was sepsis in five studies and pneumonia in one study. The total number of included patients was 1,545 (number of participants per study ranging from 102 to 581).

Figure 1

Flow diagram for selection of studies.

Table 1

Characteristics of included studies

Study	Design	Population (N; median age; prevalence)	Index tests	Reference standard	Setting; country	Inclusion criteria	Exclusion criteria
Sepsis
Camm et al. [20]	Prosp, consec	316; 75y; 38.0%	Age < 85y, new confusion, known dementia diagnosis, Barthel, SIRS, NICE HR, NICE MR and HR, qSOFA, NEWS	Combination of infection, IV fluids, hospitalisation	ED; UK	Adult residents within Oxfordshire with suspicion of infection	None
de Guadiana Romualdo et al. [21]	Prosp, cx, consec	223; 69y; 31.4%	Previous antibiotics, PCT, CRP presepsin	Sepsis-3 guidelines	ED; Spain	Adult patients (>14 years) with suspected infection and body fluid cultures at admission	Evidence of an immunocompromised state, terminal stage of disease and pregnancy
de Guadiana Romualdo et al. [22]	Prosp, cx, consec	102; 71y; 48.0%	PCT, CRP LBP	Sepsis-3 guidelines	ED; Spain	Adult patients (>14 years) with suspected infection and body fluid cultures at admission	Evidence of an immunocompromised state, terminal stage of disease and pregnancy
Geier et al. [23]	Prosp, cx, consec	151; 68y; 47.7%	Dyspnea, level of consciousness, known dementia diagnosis, diabetes mellitus, ESI, MEWS, MEDS	Sepsis guidelines of surviving sepsis campaign	ED; Germany	Adults with suspected sepsis: two or more SIRS criteria or a working diagnosis that included a potentially systemic infection, a SIRS, sepsis, severe sepsis or septic shock	None
Van Nieuwkoop et al. [25]	Prosp, cx, consec	581; 66y; 22.6%	Age, Body temp>38.6°C, heart rate, shaking chills, altered mental status, diabetes mellitus, PCT	Positive blood culture	Primary care + ED; The Netherlands	Adults with febrile UTI: age ≥ 18 years + fever + at least one symptom of UTI and a positive nitrite dipstick test or leukocyturia	Current treatment for urolithiasis or hydronephrosis, pregnancy, hemo- or peritoneal dialysis, a history of kidney transplantation or known presence of polycystic kidney disease
Pneumonia
Heining et al. [24]	Prosp, cx, consec	172; 71y (mean); 16.9%	PCT	Chest radiograph	ED; Germany	Adults presenting with acute dyspnea	None

Prosp, prospective design; cx, cross-sectional design; consec, consecutive enrolment; ED, emergency department; PCT, procalcitonin; CRP, C-reactive protein; LBP, lipopolysaccharide-binding protein; ESI, Emergency Severity Index, cut-off ≤2; MEWS, Modified Early Warning Score, cut-off ≥5; MEDS, Mortality in Emergency Department Sepsis score, cut-off ≥8; Barthel score ≥ 15; SIRS, systemic inflammatory response syndrome Score, cut-off ≥2; NICE HR, National Institute for Health and Care Excellence: high-risk criteria form NICE guideline NG51, cut-off ≥1; NICE MR + HR, moderate-risk and high-risk criteria form NICE guideline NG51, cut-off ≥1; qSOFA, quick Sequential Organ Failure Assessment score, cut-off ≥1; NEWS, National Early Warning Score, cut-off value of >4; UTI, urinary tract infection; IV, intravenous.

Study population

All studies were performed at the ED; one study was performed both at the ED and in primary care clinics [25]. Median age across studies ranged from 66 to 75 years, and 45.8% were men (the proportion of men per study ranged from 37.7% [25] to 59.8% [22]). Prevalence of serious infections differed substantially between studies, from 16.9% [24] to 48% [22], which was mainly caused by differences in inclusion criteria. For example, one study with a pneumonia prevalence of 16.9% included ‘adults presenting with acute dyspnea’ [24], whilst another study used more selective inclusion criteria—‘adults with suspected sepsis: two or more SIRS criteria or a working diagnosis that included a potentially systemic infection, a SIRS, sepsis, severe sepsis or septic shock’—leading to a sepsis prevalence of 48% [23]. More information on the inclusion criteria of each study is presented in Table 1.

Risk of bias

An overview of the risk of bias assessment is presented in Figure 2. The overall risk of bias was moderate to high. Three studies did not pre-specify the thresholds for all biomarkers under investigation [21,22,25]. In one of these studies, it was also not clear whether the index test was interpreted without knowledge of the result of the reference standard [22]. Vice versa, two studies interpreted the reference standard with knowledge of the index tests [20,23], and one other study did not clearly specify how blinding for the index test results was ensured [24]. Finally, not all patients received the same reference standard in one study [20].

Figure 2

Risk of bias summary: QUADAS-2 risk of bias and applicability.

Diagnostic test accuracies

Figure 3 is a dumbbell plot presenting shifts in infection probability by individual clinical features, diagnostic prediction rules combining clinical features, blood tests, and diagnostic predictions rules with clinical features and blood tests combined.

Figure 3

Likelihood ratios and pre- and post-test disease probabilities (%) for (a) clinical features (patient characteristics, signs and symptoms), (b) diagnostic prediction rules with clinical features, (c) blood tests and (d) diagnostic prediction rules combining clinical features and blood tests. 95%CI, 95% confidence interval; SIRS, systemic inflammatory response syndrome criteria for sepsis alert; qSOFA, quick Sequential Organ Failure Assessment; NEWS, National Early Warning Score; MEWS, Modified Early Warning Score; MEDS, Mortality in Emergency Department Sepsis score; ESI, Emergency Severity Index; CRP, C-reactive protein; LBP, lipopolysaccharide-binding protein; PCT, procalcitonin.

Individual clinical features

Patient characteristics and clinical features such as age, gender, confusion of recent onset or dyspnoea perform poorly to rule out a serious infection (minimum LR− of 0.61 (95%CI 0.48–0.80) for body temperature > 38.6°C). The abilities to rule in were equally moderate (maximum LR+ 4.94 (95%CI 1.10–22.10) for altered level of consciousness), suggesting that they are also less useful as red flags, even though the associated specificities were high (86%–94%, Appendix 3) and altered level of consciousness did come close to our pre-specified criterion of LR+ 5 for ruling in [20,23,25].

Diagnostic prediction rules combining clinical features

One study developed a new diagnostic prediction rule (age > 65 years, body temperature > 38.6°C, heart rate > 100 beats/minute) and reported the lowest LR− (0.30, 95%CI 0.16–0.57) [25]. A negative result on this rule shifted the probability of sepsis from 22.5% prior to testing to 8.0%. In addition, two studies evaluated previously developed prediction scores (Barthel score, SIRS score, NICE score, qSOFA, NEWS, MEWS, MEDS and ESI). The LR− for these prediction rules ranged from 0.39 (95%CI 0.23–0.66) to 1.05 (95%CI 0.81–1.37).

None of the clinical prediction rules combining clinical features were useful as a red flag for sepsis, with LR+ ranging from 0.97 (qSOFA score, 95%CI 0.79–1.17) to 2.78 (MEDS score, 95%CI 1.75–4.40).

Blood tests

Procalcitonin (PCT) was the most commonly investigated biomarker, in three studies for sepsis and one study for pneumonia. None of the cut-offs achieved our predefined rule-in criterion, whereas a few reached the rule-out criterion, i.e. a level below 0.10 ng/ml results in a LR− of 0.03 (95%CI 0.00–0.23), a level below 0.25 ng/ml in a LR− of 0.11 (95%CI 0.05–0.22) and a level below 0.37 ng/ml in a LR− of 0.20 (95%CI 0.10–0.42) for sepsis. In order to further explore a possible threshold effect for procalcitonin by prevalence, we plotted the change in probability for each cut-off (Figure 4). The plot suggests that procalcitonin levels <0.37 ng/ml in higher prevalence situations are suitable for ruling out sepsis. The same does not apply to pneumonia, for which the only available study found that levels <0.25 ng/ml lower the disease probability only from 16.9% to 11.6%.

Figure 4

Probability of serious infection in older patients, by procalcitonin threshold (ng/ml). Triangles represent the probability of infection after a negative test. Rectangles represent the probability of infection after a positive test.

Four studies investigated blood biomarkers [21,22,24,25]. C-reactive protein (CRP) was assessed for the diagnosis of sepsis at high cut-offs (220 mg/L and 280 mg/L) in two studies, but neither reached a sufficient rule-in (LR+ 3.03 at 280 mg/L (95%CI 1.57–5.83)) or rule-out value (LR− 0.53 at 220 mg/L (95%CI 0.37–0.75)). This also applies to lipopolysaccharide-binding protein (LBP), which was assessed by one study for the diagnosis of sepsis using a cut-off value of 33.9 μg/dl. The use of presepsin as a biomarker might be useful to rule out sepsis at a cut-off of 312 ng/L (LR− 0.17; 95%CI 0.04–0.69) but not at a cut-off of 849 ng/L, based on one study [21].

None of the biomarkers reported in this review proved useful as a red flag at the investigated cut-offs, although at higher cut-off levels, some came close to our pre-specified criterion of LR+ 5 for ruling in a serious infection (e.g. PCT at 10.0 ng/ml, LR+ 4.43 (95%CI 2.60–7.24)).

Diagnostic predictions rules with clinical features and blood tests combined

One study developed two new prediction rules for the diagnosis of sepsis, using various combinations of age, body temperature, heart rate, diabetes and procalcitonin [25]. The lowest LR− was achieved by combining procalcitonin >0.25 ng/ml with body temperature > 38.6°C (LR− 0.10; 95%CI 0.04–0.27). The second-best rule further added diabetes mellitus as comorbidity to these (LR− 0.12; 95%CI 0.05–0.33). In this study, these rules lowered the probability of sepsis from 22.6% to 2.9% and 3.4%, respectively.

Discussion

We identified six studies on the accuracy of clinical features and biomarkers for the diagnosis of serious infections in older people presenting to ambulatory care. The identified studies differed substantially regarding the included study population, and the risk of bias was moderate to high. Especially the evaluation of biomarkers without a pre-specified cut-off led to these studies being scored as high risk of bias. PCT, for which most evidence was available, proved to be useful for ruling out sepsis at values below 0.37 ng/ml in moderately high prevalence settings. No studies investigated its value in a low prevalence setting. We found only a few studies on CRP, and none evaluated its use for ruling out a serious infection.

Strengths and weaknesses

The main strengths of this study were the elaborate search in combination with a strict focus on prospective studies performed in an older population presenting to ambulatory care. A comprehensive search strategy was needed to identify the limited number of studies fitting our strict inclusion criteria. We deliberately restricted our search based on applicability and quality, which resulted in high applicability scores of QUADAS-2. Studies in hospitalised or institutionalised patients were excluded, and at least 50% of all included participants had to be 65 years or older. We also excluded retrospective studies, as those studies tend to overestimate the diagnostic accuracy of the index tests [26].

The limited number of studies, their moderate to high risk of bias and the heterogeneity of the included population prevent us from making firm conclusions. We also identified only one study that was performed in primary care. Whether the results of studies performed at EDs are transferable to primary care will depend on country specific differences in healthcare systems and the resulting potential for spectrum bias [27]. A subgroup analysis per setting was not feasible since the only study performed in primary care was also partially conducted at EDs. New studies in lower prevalence situations such as primary care are certainly needed, especially since prognosis in older patients is highly dependent on the severity level of infection at the start of treatment, making early diagnosis a priority in this vulnerable population.

Comparison with other studies

Another systematic review, which only focused on the value of biomarkers for diagnosing serious infections in older patients [28] concluded that further evidence from high-quality studies is needed to guide clinical practice, which is in line with our findings. Based on 11 studies of moderate quality, they also found that there is currently not enough evidence to support the use of CRP to safely rule out a serious infection in older persons and that procalcitonin may be useful to rule out bacteraemia at levels <0.38 ng/ml. They state that symptoms and signs remain the mainstay of diagnosis in community-based populations, which is however not fully supported by our findings. Our data suggest that individual signs and symptoms are not sufficient to rule out serious infections. Moreover, only a few signs and symptoms approached our predefined rule in criterion, in order for them to be useful as a red flag. However, we should acknowledge that the evidence is scarce and signs and symptoms that were not investigated may still be useful. In addition, a diagnostic assessment always consists of multiple tests from observation, history, and clinical examination which in combination may bring disease probability to a point where sufficient certainty is achieved for action to be taken. In a second systematic review, Lee et al. investigated the use of procalcitonin for diagnosing bacterial infections in older patients [29]. They identified four studies, with results similar to our systematic review, suggesting that procalcitonin may have added value to the diagnosis of sepsis in older patients, especially as a rule-out tool.

Implications for clinical practice

Presented with an acutely ill older patients, clinicians’ first goal in ambulatory care is to rule out the possibility of a serious infection which might warrant referral or additional testing. Often assessed clinical features may be less helpful to rule out serious infections in older people presenting to ambulatory care as generally thought, although the supporting evidence is not very strong in this setting. Combinations of clinical features may have better rule-out ability (e.g. NICE) but are still not accurate enough to rule out with sufficient safety. For example, an 80-year-old patient suspected of sepsis presenting to an ED has a median risk of sepsis of 38.0%. This probability decreases to 36.4% if this patient has no sign or symptom of the NICE medium- or high-risk sepsis criteria, which does not allow to safely rule out sepsis. A positive test, with any sign or symptom of these criteria present, only increases the probability to 38.4%. The use of these scores in primary care may therefore lead to increased referral rates to hospital due to high false-positive rates. This may put even more pressure on hospital emergency services, thereby further delaying care for actual septic patients.

PCT may safely rule out sepsis at 0.37 ng/ml or lower in moderately high prevalence settings. Its use in a newly developed prediction rule proved promising as well, although these rules should not be implemented without prior external validation. We could not identify any evidence on its value in low prevalence settings. CRP was only evaluated at higher thresholds, meaning that the uncertainty regarding the optimal rule-out threshold remains.

We could not identify any sign, symptom, blood test or rule that reached our predefined rule-in criterion of LR+ 5.0. Nevertheless, signs such as an altered level of consciousness reached a LR+ of 4.94 and could therefore still be useful as a red flag. Procalcitonin levels >2.0 ng/ml might also be useful to rule in sepsis in moderate to high prevalence settings. We considered tests to be useful in ruling out a serious infection in ambulatory care if their LR− was <0.20 and useful as ‘red flags’ for serious infections when their LR+ was 5.0 or higher. These pre-set levels of clinical significance are not stringent and serve only as a suggested guide for interpretation of the results. Other levels may be more appropriate in some situations, depending on the target disease, patient population and setting of diagnosis or treatment.

Implications for research

Our systematic review clearly shows that evidence is scare and heterogeneous. New diagnostic studies are needed in these older patients, especially in primary care, but they should be more streamlined concerning the patient population and target disease in order to increase the comparability and applicability of the findings. A clear definition of what a serious infection is in an older person presenting to ambulatory care should help to delineate the design of data collection of future diagnostic studies.

CRP is already adopted in routine practice as a point-of-care test in some countries, for example, in patients with lower respiratory tract infections [30,31]. Point-of-care tests for procalcitonin are available as well, although they are currently less fit for purpose because of their longer turn-around-time (i.e. ≥20 minutes) [32]. New studies that are aligned with routine practice are needed to show the incremental value of CRP or other biomarkers over and above the clinical assessment, especially in terms of ruling out serious infections, and their impact on patient outcomes and healthcare processes.

Conclusions and implications

Individual signs and symptoms have limited ability to safely rule out a serious infection in this population—ruling out a serious infection will require a combination of tests. A few signs and symptoms such as altered consciousness moderately raise the probability of a serious infection. Procalcitonin may have potential as a biomarker in ruling out sepsis in older persons presenting to ambulatory care, but the existing evidence is too scarce and too heterogeneous to draw firm conclusions. Its diagnostic potential in low prevalence settings needs to be evaluated. New diagnostic studies in this setting are clearly needed, especially in primary care, but they should be more streamlined to improve the applicability of the findings into practice.

Acknowledgements

We would like to thank Dr. Krizia Tuand, biomedical information specialist at the KU Leuven Libraries—2Bergen—Désiré Collen Learning Centre (Leuven, Belgium), for her help with the development of the search strategy.

Declaration of Conflicts of Interest

None.

Declaration of Sources of Funding

This research was funded by the Research Foundation Flanders (Fonds Wetenschappelijk Onderzoek (FWO), Odysseus program) [grant number G0H8518N]. The financial sponsor played no role in the design, execution, analysis and interpretation of data, nor in the writing of the study.