In hospitalised surgical patients, undernutrition remains widely undiagnosed/underestimated, with a prevalence of 40–64 % depending on the criteria used(Reference Brunn, Bosaeus and Bergstad1–Reference Mourão, Amado and Ravasco4). Undernutrition has been associated with impaired wound healing and higher infection rates, increased postoperative morbidity and mortality, longer hospital length of stay (LOS) and increased overall costs(Reference Norman, Pichard and Lochs5–Reference Leonard-Jones7).
Nutritional screening on admission to detect patients who may require early nutritional intervention is recognised(8); recent evidence advocates that screening and assessment methods may also predict clinical outcomes, such as hospital LOS(Reference Kyle, Kossovsky and Karsegard9–Reference Schiesser, Müller and Kirchhoff12). Moreover, since surgery induces catabolism not always compensated by appropriate nutritional support, surgical patients are more prone to nutritional deterioration throughout their hospital sojourn(Reference Weimann, Braga and Harsanyi13); however, few studies(Reference Brunn, Bosaeus and Bergstad1, Reference McWhirter and Pennington14–Reference Fettes, Davidson and Richardson16) have evaluated such changes. Weight loss and anthropometry may offer a more correct approach to quantify in-hospital nutritional depletion, although current anthropometric thresholds may possibly be deficient in detecting undernutrition in clinical practice(Reference Corish, Flood and Mulligan15, Reference Fettes, Davidson and Richardson16).
The aims of this longitudinal study conducted in hospitalised surgical patients were: (1) to assess the value of nutritional risk (Nutritional Risk Screening-2002 (NRS-2002), Malnutrition Universal Screening Tool (MUST)) and nutritional status methods (recent % weight loss, Subjective Global Assessment (SGA), anthropometry) in predicting hospital LOS; (2) to test the ability of current percentiles thresholds in detecting undernutrition and predicting LOS and (3) to determine possible changes in nutritional status during hospitalisation and their association with LOS.
Material and methods
Study population
This prospective longitudinal study was carried out in a University Hospital large general Surgical Department in Lisbon, Portugal. The study was approved by the Ethics Committee and was conducted in accordance with the Helsinki Declaration of 1975 as revised in 1983 and 2002; included patients gave their written informed consent to participate. During 8 months, all consecutive adult patients ( ≥ 18 years) with a predicted length of stay ≥ 4 d were considered eligible. At admission, 300 patients met the inclusion criteria. Exclusion criteria comprised organ transplantation, coma, bedridden, intensive care patients, or previous surgery, chemo/radiotherapy during the year before hospital admission; patients submitted to surgery before nutritional assessment were also excluded (n 25). Since two patients died before the second evaluation, 298 patients were longitudinally assessed at admission and discharge. Patients' characteristics are summarised in Table 1.
Table 1 Patients' characteristics (Number of patients and percentages)
GI, gastrointestinal.
Study parameters
All nutritional data were collected by a single research dietitian (M. Correia). Nutritional risk and status were assessed before any surgical procedure and within 3 d after hospital admission; weight loss and anthropometric data were again evaluated 24 h before discharge.
Nutritional risk
NRS-2002 screens patients by grading nutritional status and disease severity (low, moderate or severe), with an adjustment for age ≥ 70 years; a patient is at nutritional risk when the final score is ≥ 3(Reference Kondrup, Allison and Elia17). MUST integrates BMI, unintentional recent weight loss and scores the acute disease effect on intake. Nutritional risk is then categorised: 0 = low risk, 1 = medium risk and ≥ 2 = high risk(18). In order to enable comparisons between methods, MUST categories were grouped in: low v. moderate+high risk.
Nutritional status
SGA is based on the patient's clinical history, physical examination, estimated weight loss and changes in diet intake allowing nutritional status categorisation: (a) well nourished, (b) moderately undernourished or suspected of being undernourished, and (c) severely undernourished(Reference Detsky, McLaughlin and Baker19).
Percentage weight loss was calculated and classified at admission as clinically significant, indicative of undernutrition or of nutritional risk, if ≥ 5 % in the previous 6 months(8); % weight loss was determined at discharge and valued by itself or cumulatively.
Anthropometric data included height, weight, triceps skinfold thickness (TSF), mid-arm circumference (MAC) and mid-arm muscle area (MAMA). Height was measured in the upright position using a stadiometer, and weight was determined with a Seca floor scale and registered to the nearest 0·5 kg; both allowed the calculation of BMI (BMI = weight (kg)/height (m2)). BMI was classified using the WHO criteria: underweight if < 18·5 kg/m2, regular 18·5–24·9 kg/m2, overweight 25–29·9 kg/m2 or obese ≥ 30 kg/m2 (20). TSF in mm was measured with a skinfold calliper (John Bull®; Briish Indicators Ltd) at the back of the non-dominant arm, at the midpoint between the tip of the acromial process of the scapula and the olecraneon process of the ulna; the mean of three readings was used. MAC in cm was measured at the same site and position as described for TSF perpendicular to the long axis of the arm, using a non-stretchable flexible tape; care was taken not to pinch or gap the tape and three measurements to the nearest 0·1 cm were taken and their mean was used. MAMA in cm2 was calculated using TSF and MAC values, according to the formula:
$$\begin{eqnarray} MAMA = (MAC\,(cm) - 0\cdot 314\times TSF\,(mm)^{2})/(4\times 3\cdot 14). \end{eqnarray}$$
The calculated muscle area was adjusted for bone area by subtracting 6·5 cm2 in women and 10 cm2 in men. Reference tables, standardised for age and sex, and validated for normal subjects were used for the classification of individual values(Reference Bloch, Shills, Bloch and Shills21); a TSF, MAC or MAMA < 15th percentile is indicative of undernutrition, whereas a TSF >85th percentile is indicative of excess body fat(Reference Bloch, Shills, Bloch and Shills21).
Statistical analysis
Data were analysed using SPSS 16.0 statistical software (SPSS, Inc.). Categorical data were expressed as number and percentage of patients; continuous data with a normal distribution were expressed as means and standard deviations, while continuous data not normally distributed were expressed as medians with their interquartile range. Comparisons were made using non-parametric tests as appropriate: Wilcoxon W for comparing median values between two independent groups; median test for comparing median values between three or more groups and the χ2 test for comparing categorical data. Spearman's correlation coefficient was calculated to verify linear association between two continuous variables. Agreement analysis between methods and LOS was performed using the concordance κ coefficient, the statistics of which were further classified according to Fleiss(Reference Fleiss22). For κ coefficient calculation, two categories of LOS were established by using the median value of its distribution (10 d) as the cut-off value.
Sensitivity, specificity and predictive values were calculated to appraise the value of the nutritional risk and status tools tested in predicting LOS. A prolonged LOS was defined as a stay in hospital above the median value of its distribution. Sensitivity was expressed as the proportion of patients with a longer LOS, which were classified as ‘at nutritional risk’ or ‘undernourished’ or ‘underweight’ by nutritional methods. On the other hand, specificity refers to the proportion of patients with a LOS below the 50th percentile and classified as ‘without nutritional risk’ or ‘well nourished’ or ‘well nourished/overweight/obese’ by nutritional methods. The positive predictive value is defined as the proportion of patients classified as ‘at nutritional risk’ or ‘undernourished’ or ‘underweight’ by nutritional methods, which have a longer LOS. Conversely, the negative predictive value is the proportion of patients identified as ‘without nutritional risk’ or ‘well nourished/overweight/obese’ by nutritional methods, who have a LOS below the 50th percentile.
Statistical significance was determined for P <0·05.
Results
The median hospital LOS was 10 (7–14) d; significantly longer in male (11 (8–14) v. 8 (7–13) d, P< 0·001) and in patients ≥ 65 years (11 (8–15) v. 9 (7–12) d, P< 0·01). According to diagnoses, cancer patients had a longer LOS than non-cancer patients (11 (8–14) v. 9 (7–12·5) d, P< 0·01); among those, gastrointestinal (GI) cancer patients (n 99, 33 %) stayed in hospital for a longer period (12 (9–16) d), when compared with breast or gynaecological cancer patients (n 28, 9 %; 7 (6–8) d, P< 0·001). Regarding indication for surgery, the median LOS of patients admitted for non-elective surgery was significantly longer than that of patients admitted for elective surgery (11 (8–16) v. 8·5 (7–12) d, P< 0·001). LOS was also longer in GI surgery v. non-GI surgery (11 (8–15) v. 7 (6–8·5) d, P< 0·001).
Admission
Length of stay in relation to nutritional risk and status methods
The ability of the various methods to predict hospital LOS was the first step in the analyses (Table 2).
Table 2 Median length of stay (LOS) according to nutritional risk and status at admission (Number of patients and percentages; medians and interquartile ranges (IQR))
NRS-2002, Nutritional Risk Screening-2002; MUST, Malnutrition Universal Screening Tool; SGA, Subjective Global Assessment.
Statistical analysis was performed with non-parametric * Wilcoxon W test or non-parametric; † Median T test as appropriate, for median values comparison. P values express statistical differences between LOS in each nutritional risk and status category.
Nutritional Risk Screening-2002 and Malnutrition Universal Screening Tool
Patients identified as at nutritional risk by either NRS-2002 or MUST categories had a longer median LOS (risk v. low risk for NRS-2002 and moderate+high risk v. low risk for MUST, P< 0·001.
Percentage weight loss, Subjective Global Assessment and BMI
Patients who reported ≥ 5 % weight loss at admission were hospitalised for a longer period (P< 0·001). According to SGA, LOS was significantly longer in severe and moderate undernutrition v. well-nourished patients (P< 0·001). Using BMI, hospital LOS was longer in underweight than in regular-weight patients (P< 0·05), in the latter median LOS was not significantly different from overweight/obese.
Anthropometry
Using the previous anthropometric percentile criteria mentioned in the Methods section defining undernutrition(Reference Bloch, Shills, Bloch and Shills21), there was a lower prevalence of undernourished patients when compared with SGA; 69–75 % of patients identified as undernourished by SGA were classified as well nourished by anthropometric criteria, which may signify a misclassification by anthropometry. Moreover, κ coefficient agreement between the standard thresholds of TSF/MAC/MAMA v. SGA was low (κ = 0·15–0·18, P< 0·01). Based on the poor performance of the standard percentile thresholds (15th and 85th percentiles) in predicting hospital LOS, we thus tested a unique and simpler percentile value (50th percentile) in order to evaluate its clinical feasibility (Table 3).
Table 3 Median length of stay (LOS) using standard percentiles (≤15th, 15th–85th, ≥85th) v. a proposed threshold (<50th, ≥50th percentile) for anthropometry at admission and median variation of triceps skinfold (TSF), mid-arm circumference (MAC) and mid-arm muscle area (MAMA) during hospital stay (Number of patients and percentages; medians and interquartile ranges (IQR))
Statistical analysis was performed by non-parametric * Wilcoxon W test or non-parametric; † Median T test as appropriate, for comparison of median values.
‡ P values express statistical differences between median LOS by category of TSF, MAC, MAMA, at admission.
§ P values express statistical differences between median variation of TSF, MAC and MAMA during LOS by category of TSF, MAC, MAMA, respectively.
∥ Standard percentiles.
¶ Suggested percentiles ( < 50th and ≥ 50th percentile).
Using the standard percentiles, a TSF ≤ 15th did not predict a longer LOS v. a TSF within 15th–85th percentiles; however, patients with a TSF ≥ 85th percentile stayed longer in-hospital (P< 0·05) than patients with a TSF between the 15th and 85th percentiles. Conversely, a MAC or MAMA ≤ 15th percentile predicted a longer LOS (P< 0·001), whereas patients with MAMA ≥ 85th percentile had a shorter hospital stay (P< 0·05). Furthermore, the simultaneous presence of TSF+MAC+MAMA ≤ 15th percentile did not predict a longer LOS v. patients with TSF+MAC+MAMA >15th percentile (12 (9–22·5) v. 10 (7–13) d, NS).
Using the 50th percentile threshold, a TSF ≤ 50th or >50th percentile did not significantly affect the median LOS. Nonetheless, a MAC or MAMA ≤ 50th percentile predicted longer median LOS (P< 0·001); patients with TSF+MAC+MAMA ≤ 50th percentile also had significantly longer LOS (Table 3).
Comparison between the different nutritional risk and status methods in predicting a longer length of stay
In this section, we first undertook an analysis on the agreement between methods' results and LOS, by calculating the κ concordance coefficient (Table 4). Indeed, we did find a fair and significant κ concordance coefficient between LOS and the methods NRS-2002, MUST, SGA, recent weight loss and MAC (κ = 0·237–0·359, P< 0·001); on the other hand, although significant, the κ concordance coefficient between LOS and BMI/MAMA was low (κ = 0·069–0·163, P< 0·05) and no agreement was found between LOS and TSF (NS).
Table 4 Concordance between length of stay (LOS) and nutritional risk and status methods
NRS-2002, Nutritional Risk Screening-2002; MUST, Malnutrition Universal Screening Tool; SGA, Subjective Global Assessment; TSF, triceps skinfold; MAC, mid-arm circumference; MAMA, mid-arm muscle area.
* P <0·05, ** P <0·01, *** P <0·001.
† In order to enable concordance calculation between TSF/MAC/MAMA and LOS, the suggested percentiles categorisation ( < 50th and ≥ 50th percentile) was used.
Subsequently, to evaluate and confirm the value of the different methods tested in predicting a longer LOS, sensitivity, specificity and predictive values were calculated (Table 5). NRS-2002, MUST, recent weight loss, SGA, MAC and MAMA presented a high sensitivity value, thus revealing a strong capacity in identify patients who had a longer LOS as at risk/undernourished; on the other hand, BMI and TSF had a low capacity to detect patients who had a longer LOS as underweight/undernourished. Specificity values were higher for SGA, MAC and MAMA, thus revealing a strong capacity to identify patients who had a shorter LOS as well nourished; NRS-2002, MUST and recent weight loss had moderate specificity values, while BMI and TSF had low specificity values, thus revealing a weak capacity in detecting patients with a shorter LOS as underweight/undernourished.
Table 5 Nutritional risk and status tools and length of stay (LOS): sensitivity, specificity, positive and negative predictive values (Percentages and 95 % confidence intervals)
PPV, positive predictive value; NPV, negative predictive value; NRS-2002, Nutritional Risk Screening-2002; MUST, Malnutrition Universal Screening Tool; SGA, Subjective Global Assessment; TSF, triceps skinfold; MAC, mid-arm circumference; MAMA, mid-arm muscle area.
* Statistically significant.
Hospital stay and discharge
Length of stay in relation to nutritional status changes during hospital stay
At discharge, the various methods showed an overall increase in the prevalence of undernutrition (Tables 3 and 6).
Table 6 Comparison of median hospital length of stay (LOS), according to nutritional status categories of % weight loss and BMI at discharge (Number of patients and percentages; medians and interquartile ranges (IQR))
Statistical analysis was performed by non-parametric * Wilcoxon W test or non-parametric; † Median T test as appropriate, for median values comparison. P values express statistical differences between median LOS by category of nutritional risk and status.
Percentage weight loss and BMI
Adding the in-hospital weight loss, there was an increase in the prevalence of patients who lost ≥ 5 % of their usual weight. Moreover, twenty-nine (25 %) patients who at admission had reported < 5 % weight loss, lost weight during hospital stay. In fact, a median 3·3 (1·9–4·9) % weight loss was accounted for in 264 (87 %) patients; of these, fifty-nine (22 %) lost ≥ 5 % weight. It is worth mentioning that patients admitted for either non-elective surgery or GI surgery presented a greater in-hospital % weight loss (P< 0·01). There was a positive and significant correlation between cumulative % weight loss and LOS (0·43, P< 0·001). In addition, patients who lost ≥ 5 % of their weight in the hospital had a significantly longer LOS than those who lost < 5 % (14 (11–23·5) v. 9 (7–12) d, P< 0·001).
It is noteworthy that while in hospital, weight loss occurred in thirteen of the seventeen already underweight patients, in 169/191 (89 %) with regular BMI and in 82/90 (91 %) with overweight/obesity. Thereby, there was an increase in the undernutrition prevalence after admission: seventeen patients with regular BMI became underweight and twenty-five overweight/obese patients presented a regular BMI at discharge. Reinforcing admission data, underweight patients had a more prolonged LOS v. those with regular BMI or overweight/obesity (P< 0·001).
Anthropometry
Again, both standard percentiles(Reference Bloch, Shills, Bloch and Shills21) and the 50th percentile threshold were tested in association with LOS (Table 3). A TSF ≤ 15th or ≤ 50th percentile did not affect LOS (NS); conversely, patients with a MAC or MAMA ≤ 15th or ≤ 50th percentiles had significantly longer LOS v. those with a MAC or MAMA ≥ 15th or >50th percentiles.
TSF at discharge showed a median negative variation of 0·3 (interquartile range: − 0.7, − 0.2)mm in 216 (72 %), thus expressing a reduction in fat stores; patients with a negative TSF variation had longer LOS than those who maintained/increased their TSF (8 (6–12) v. 10 (8–14) d, P< 0·001). There was a negative correlation between TSF variation and LOS ( − 0·25, P< 0·001), though TSF variation during hospital stay was not associated with diagnoses or type of surgery.
Concerning MAC, 185 (62 %) patients had a median negative variation of − 0·5 (interquartile range − 1·0; − 0·2) mm; these had a longer LOS v. patients who maintained/increased MAC (11 (8–15) v. 8 (7–11) d, P< 0·001). MAC variation and LOS were negatively correlated ( − 0·35, P< 0·001); patients admitted for GI surgery had a greater negative MAC decrease (P< 0·001).
At discharge, MAMA had decreased in 189 (63 %) patients with a median negative variation of − 1·5 (interquartile range − 3·0; − 0·6). Similarly to TSF and MAC, the median LOS of patients with a negative MAMA variation was longer than in patients who maintained/increased MAMA (11 (8–12) v. 8 (7–11·5) d, P< 0·001) and there was a negative correlation between MAMA variation and longer LOS ( − 0·27, P< 0·001). By analysing the combination of TSF+MAC+MAMA, patients with a simultaneous negative variation in all three criteria (n 158, 53 %) had a longer LOS than in patients with TSF+MAC+MAMA positive variation, (11 (8–15) v. 8 (7–12) d, P< 0·001).
Nutritional risk at discharge
Given the results on nutritional deterioration during hospital stay, nutritional risk at discharge and LOS were assessed. Patients were at risk if % weight loss was ≥ 5 %. Patients at risk at discharge had a longer LOS then patients with no risk (P< 0·001). Moreover, patients at risk presented a higher % weight loss variation, when compared to patients without risk (3·6 (2·1–5·4) v. 1·6 (0·8–2·9) %, P< 0·001). In what concerns anthropometric data, patients at risk had a greater TSF, MAC and MAMA negative variation (P< 0·001).
In order to evaluate the relation between nutritional risk at discharge by ≥ 5 % weight loss and patients' nutritional status, assessed by TSF, MAC and MAMA, a frequency analysis was performed. Results revealed a significant association between ≥ 5 % weight loss v. MAC and MAMA percentile categories (P< 0·001). Most patients with MAC or MAMA ≤ 15th percentile were identified as being at nutritional risk; however, by using the standard percentile categories, patients at risk and without risk were more likely to present a MAC or MAMA between the 15th and 85th percentiles. On the other hand, using the 50th percentile as a cut-off, most patients identified at nutritional risk had a MAC or MAMA < 50th percentile, whereas patients without nutritional risk were more likely to have a MAC or MAMA ≥ 50th percentile (P< 0·001). No association was found between nutritional risk at discharge and TSF percentile categories (NS).
Discussion
Undernutrition at admission and nutritional deterioration during hospitalisation have a detrimental effect on the clinical outcome of surgical patients, leading to an increased risk of complications and prolonged hospital LOS(Reference Shils, Shike and Ross23). The present study explored the ability of various methods to assess nutritional risk and status at admission, as well as nutritional changes throughout hospital stay and their association with LOS. For anthropometry (TSF, MAC, MAMA), the 50th percentile as a unique cut-off for undernutrition was tested in addition to the standard percentiles(Reference Bloch, Shills, Bloch and Shills21). All tested methods but TSF and BMI, as well as the 50th percentile threshold for MAC and MAMA, were found to predict a longer LOS in at-risk or already undernourished surgical patients.
The present data in surgical patients showed that poorer scores at admission of NRS-2002 or MUST as well as worse SGA and greater % weight loss predicted longer LOS; these results agree with other studies(Reference Kyle, Kossovsky and Karsegard9, Reference Raslan, Gonzalez and Dias10, Reference Sorensen, Kondrup and Prokopowicz24). Underweight patients by BMI also had longer LOS, however, with limited clinical value since it failed to recognise many undernourished patients; moreover, BMI had low sensitivity, specificity and predictive values in predicting LOS. BMI does not value changes in weight or in body compartments(Reference Kyle, Morabia and Slosman25); we have already demonstrated that when compared with standard methods, BMI has low sensitivity, specificity, positive and negative predictive values(Reference Almeida, Correia and Camilo2, Reference Boléo-Tomé, Monteiro Grillo and Camilo26).
Body composition assessment provides more accurate information on nutritional status(Reference Kyle, Morabia and Slosman25); anthropometry is a simple, non-invasive and affordable method to assess muscle and fat depletion(Reference McWhirter and Pennington14), and thus more likely to mirror changes in body composition. The current anthropometric percentiles easily identified severe undernutrition(Reference Mourão, Amado and Ravasco4); however, they underestimate the prevalence of undernutrition(Reference Corish, Flood and Mulligan15, Reference Fettes, Davidson and Richardson16). Indeed, in our study, two-thirds of undernourished patients identified by SGA were classified as well nourished by anthropometric methods. In clinical practice, appropriate thresholds are due to identify undernutrition; we did test the 50th percentile as the cut-off and a significantly higher number of undernourished patients were identified in concordance with the other methods; in addition, a MAC/MAMA ≤ 50th percentile did effectively predict a longer LOS, alike the ≤ 15th percentile. Only muscle mass deterioration stood out as clinically significant, which concurs with a study that used bioelectrical impedance and found an association between muscle mass depletion at admission and longer LOS(Reference Pichard, Kyle and Morabia27). The 15th or 50th TSF percentiles were never associated with LOS, while a TSF indicative of excess fat mass ( ≥ 85th percentile) predicted longer LOS; excess of body fat has been recently associated with poorer outcomes(Reference Kyle, Pirlich and Lochs28).
The prevalence of weight loss in the hospital was extremely high, 87 %; weight loss during hospitalisation has been reported(Reference Brunn, Bosaeus and Bergstad1, Reference McWhirter and Pennington14, Reference Corish, Flood and Mulligan15, Reference Caccialanza, Klersy and Cereda29). At discharge, our data showed that a greater % weight loss predicted longer LOS and was associated with depletion of muscle and fat. Most patients were already undernourished at admission, and a continued muscle mass deterioration associated with weight loss resulted in a higher prevalence of undernutrition at discharge. Several factors have been identified as potential contributors to in-hospital weight loss, namely post-surgery catabolic stress response, underlying disease, insufficient oral intake due to anorexia, impaired cognitive function or patients' dissatisfaction with hospital meals, and inadequate perioperative nutritional care, such as prolonged and unjustified pre- and post-operative fasting and/or insufficient energy and protein supply(Reference Caccialanza, Klersy and Cereda29, Reference Hiesmayr, Schindler and Pernicka30).
Clinicians must realise that both undernutrition at admission and nutritional deterioration in the hospital may weaken postoperative and postpone full recovery; to our knowledge, there are no studies evaluating the impact of undernutrition on after-discharge recovery of surgical patients.
Nutritional screening is clearly established as the first step in providing quality of nutritional care in hospitals(8, 31). Our results in surgical patients show that NRS-2002/MUST indicative of nutritional risk, a % weight loss/SGA indicative of undernutrition, or the 50th percentile for MAC or MAMA effectively predicted longer LOS. Nutritional status deterioration during hospitalisation is worrying, yet preventable: the implementation of adequate perioperative nutritional care, such as the Enhanced Recovery After Surgery programme, minimises catabolism and supports anabolism, thus reducing morbidity and LOS(Reference Weimann, Braga and Harsanyi13, Reference Teeuwen, Bleichrodt and Strik32). Indeed, the high prevalence of patients at risk/undernutrition at admission/discharge stresses the urgent integration of screening for all surgical patients, enabling an adequate referral to nutrition professionals for a comprehensive assessment and individualised intervention integrated in the overall management; this is the only effective approach, allowing for an adequate nutritional intake before and after discharge, and potentially promoting a faster recovery.
Acknowledgements
The present study was partially supported by a Grant from the ‘Fundação para a Ciência e Tecnologia’ (RUN 437). All authors hereby declare that there were no financial and personal relationships with other people or organisations that could inappropriately influence their work. A. I. A. was responsible for the data analysis, interpretation and writing of the manuscript. M. Correia was responsible for data collection. M. Camilo was responsible for the writing and final review of the manuscript as well as data interpretation. P. R. was responsible for the writing and final review of the manuscript, data interpretation and data analysis.
