The value of the routine pre-operative chest radiograph (POCR) has been widely debated in the literature [1-11]. Pre-operative chest radiographs are usually obtained to evaluate the evolution of a previously known disease, to detect unsuspected abnormalities that could influence the anaesthetic management and/or surgical plans , for medicolegal reasons , or institutional requirements and cost-effectiveness purposes .
Some studies showed that, in the absence of a medical history or clinical examination suggesting some intrathoracic disease, the clinical usefulness of the POCR was very low, that significant unexpected abnormalities were rarely demonstrated , and that they seldom influenced anaesthetic management or surgery [2,9,14,15], nor did they affect the final outcome . Moreover, selection of subjects for POCR might reduce costs, hospital stay, and related adverse effects due to additional unnecessary diagnostic procedures . These issues have encouraged physicians to obtain selective POCR in accordance with different criteria such as age [7,13, 18-24], sex , clinical status , the severity or the number of risk factors [13,21], or a history of cardiovascular or pulmonary diseases [11,13,18,21]. However, the precise threshold for each criterion has not been defined completely.
The aims of this study were: (i) to investigate the influence of the routine POCR on anaesthetic management and (ii) to characterise which patients might benefit from a POCR.
From May to September 1996, a prospective multicentre pilot study was conducted in 20 Italian hospitals of the Friuli Venezia Giulia Regional Health Service. In this area of about 1200 000 inhabitants, about 70 000 anaesthetic procedures are performed each year in over 24 hospitals: 17 general hospitals (two University hospitals), one children's hospital, one hospital for oncology, one orthopaedic hospital and four private hospitals. Twenty-one departments of anaesthesia and 22 departments of radiology in 20 hospitals were involved eventually. These included 16 general hospitals, one hospital for oncology, and three private hospitals.
The study was carried out by the Anaesthetist and Radiologist V.E.R.O. (Value and Effectiveness of the Routine PreOperative) chest radiograph study group. At each centre, two physicians (an anaesthetist and a radiologist) co-ordinated the study. The co-ordinator anaesthetist was responsible for the organization of the study within his own department, whereas the anaesthetist in charge of each patient was requested to fill in a questionnaire.
Data collection and definitions
After approval of the institutional ethics committee, all patients scheduled for elective surgery and submitted for a routine POCR because of an established pre-operative protocol, were enrolled in the study. The protocol was maintained unchanged in all centres during the study period. Each patient was seen by the anaesthetist on the day before surgery, after POCR had previously been requested by the surgeon. Routine pre-operative informed consent was requested. Patients who underwent 'selective' POCR as a result of the pre-anaesthetic examination, were excluded from the study.
All radiographs were acquired in analogue format. Radiographs were interpreted by the same general co-ordinator radiologist at each institution. Both the radiographic report and the film were available to the anaesthetist.
Demographic data, type of surgery, classification of surgical interventions and physical status, co-existing diseases, chest X-ray abnormalities, anaesthetic procedures and the influence of POCR on anaesthetic management were included in the questionnaire (Appendix 1). Surgical interventions were divided according to the type of surgery, and categorized as minor, standard and major . A minor operation was defined as a surgical procedure lasting less than 30 min, without tracheal intubation, craniotomy, thoracotomy or laparotomy. A standard operation was defined as a surgical procedure lasting less than 3 h with blood loss less than 10% of the estimated blood volume (including general, spinal or epidural anaesthesia). A major operation was defined as a surgical procedure lasting more than 3 h or with blood loss greater than 10% of estimated blood volume, or involving the central nervous system, lung, or cardiovascular system. The patient's pre-operative clinical condition was assessed using the American Society of Anesthesiologists' (ASA) physical status classification [27,28].
Pre-operative chest radiographies were classified into normal and abnormal by the attending anaesthetist. Abnormal findings included abnormalities of the chest wall (e.g. fractures, kyphosis, scoliosis, fibrothorax), heart and pulmonary vasculature (e.g. cardiomegaly, enlarged left ventricle, pulmonary congestion), mediastinum (e.g. deviated trachea, goitre, mediastinal mass), pleural space (e.g. pleural effusion, pneumothorax, blunted costophrenic angle), diaphragm (e.g. elevated hemidiaphragm), lung fields (e.g. infiltrates, pneumonia, atelectasis, fibrosis, infarction, metastases), and hyperaerated lung fields (e.g. emphysema).
At each centre, the usefulness of POCR for the anaesthetic management was assessed by the anaesthetist in charge of each patient and finally approved by the co-ordinator anaesthetist during a weekly meeting. Any revision of the anaesthetic plan resulting from the POCR, but not to the patient's history or physical examination, was carefully determined. Useful POCR was recorded if the patient's anaesthetist answered the question 'did the POCR alter the anaesthetic management of the patient?' in the affirmative. Changes in anaesthetic management included modifications of anaesthesia (e.g. general vs. regional), introduction of specific monitoring (e.g. cardiovascular or respiratory), or other strategies (e.g. avoidance of nitrous oxide). Additionally, it was noted if the POCR resulted in a delay in surgery due to further laboratory tests.
Continuous variables were categorized in order to improve readability and interpretability of the model. Age has been coded as ≤60 or >60, and ASA score as ≤2 or ≥3. Cut-offs roughly corresponded to the median and were shown to be sensitive in the literature [7,13,18,29]. The 60-year threshold was preferred to, at least theoretically, more sensitive cutoffs , because it was more conservative in evaluating as potentially useful all those POCRs obtained in patients older than 60 years. Co-existing diseases were divided into five groups: absent, cardiac, respiratory, other, two or more.
The preliminary analysis of the relations between useful POCR and the other variables were performed using the Pearson's χ2-test statistics and the likelihood ratio test (LRT). A P value less than 0.05 was considered to be significant.
Useful POCR was coded as a dichotomous response taking values 0 or 1 (i.e. not useful and useful, respectively). The probability of observing a success, i.e. a useful POCR, was modelled according to a random mixed effect logit model [30-32]. The use of a random effect model is motivated by the heterogeneity induced in the distribution of the response variable, because observations are clustered in hospitals. This mechanism was well described in a similar context by Griffiths  and with reference to a linear model by Feldman . Random effects were not limited to the intercept, but also to the slopes of the covariates. This implies that the overall contribution (in terms of explanatory power) of a specific covariate in determining the probability of a useful POCR can be split into fixed effects and random effects identified by the symbol β and b, respectively. The fixed effect is also known as the population average effect, indicating the usefulness of stratification for subjects belonging to the various factor classes: it remains constant for all units in the population belonging to the same stratum. The random effect represents how much each hospital increases (i.e. positive b coefficient) or decreases (i.e. negative b coefficient) the population averaged probability of a useful POCR according to the stratification factors introduced in the random part of the equation. The final model was selected using a LRT test employing the algorithm suggested by Lindstrom and Bates . Wald-type P values for each fixed effect were also computed. The analytical form of the equation associated with the random mixed effect logit model fitted on the data was presented (Appendix 2) and the probabilities for specific cases were computed. For predictive purposes, when -as in the present study -the random effects are themselves the focus and inferences are more dependent on the assumptions about their distribution, we checked the gaussian assumption for the random effect using the graphical method of Lange and Ryan . No model showed a systematic departure from these assumptions. The models were fitted using the NLME library  in S-plus .
Overall, 6111 patients undergoing elective surgery and submitted to routine POCR were enrolled in the study. Table 1 summarizes the demographic and general characteristics of the patients. An abnormal POCR was reported in 1116 patients (18.3%). The POCR influenced the anaesthetic management in 313 patients (5.1%): it changed the anaesthetic management in 226 patients (72.2%) and required a further evaluation in 80 patients (25.6%).
Table 2 shows the distribution of useful POCRs according to the different centres. The incidence of useful POCRs varied among hospitals.
Useful POCR occurred in 8.8% of patients with age >60 years, 6.8% of males, 15.5% of subjects with ASA classes 3-5, 10% of patients submitted to major surgery and 20.8% of subjects with coexisting respiratory diseases (P<0.01). Moreover, useful POCRs were shown in 32.3% of patients undergoing thoracic surgery, 11.1% of those undergoing cardiac surgery, and 9.8% of patients submitted to vascular surgery (Table 3).
At multivariate analysis, age >60 years, male sex, ASA classes 3-5, co-existing respiratory diseases and the presence of more than one comorbidity were significantly related with the probability of a useful POCR (P<0.01) (Table 4).
'Hospital effects' are clearly demonstrated in Table 5. Hospitals 1, 7, 10, 18 and 20 showed quite different b0 values when compared with hospitals 5, 9, 12, 17, and 21. The other relevant variable in the statistical model was represented by ASA classes ≥3 (i.e. b1): the weight ascribed to the same ASA class varied among hospitals (e.g. hospital 1 and hospital 4 ascribed b1 values with an opposite sign).
The equation (Appendix 2), including both fixed and random effects of the studied variables, has been applied to estimate the probability of a useful POCR. Table 6 shows the effects of different variables in different hospitals. The base-line probability of a useful POCR included all patients with variables categorized as female, age ≤60, standard intervention, ASA classes ≤2, and no comorbidities: it increased from 0.2% to 3.5%, according to the hospital. In the following columns, the effects of age and/or gender are shown in patients with ASA classes ≤2, no comorbidities and submitted to standard intervention: the same patient demonstrated different probabilities of a useful POCR moving along columns from hospital to hospital, reaching values higher than 5%. The last three columns illustrate the effects of gender and/or comorbidities in subjects with age >60 years, ASA classes ≥3, and submitted for major surgery. The probability of a useful POCR increased from 3.8% up to 81%, depending on the variables examined: pronounced 'hospital effects' were always well established. Moreover, the presence of b1 values (i.e. ASA ≥3) was very evident in the last three columns: in fact, percentages were not ranked in increasing order owing to different positive or negative coefficients ascribed to ASA class ≥3. Finally, in the last row of Table 6, we calculated the population average effects of the variables considered, assuming the study population to be homogeneous (i.e. b0 and b1 values equal zero). The base-line probability of a useful POCR was 0.9%, but increased up to 48% according to the type of variables included.
Abnormal pre-operative chest radiograph
Chest X-ray is a categorical variable that is difficult to quantify but simple to classify as 'normal' or 'abnormal'. In this study, 18.3% of patients showed an abnormal POCR. In previous studies, POCR abnormalities ranged from 0.6%  to 47% . The frequency of abnormal findings increases with age and male sex [7,11,13,20,21,24,25], and unsuspected abnormalities are more common in elderly than in young subjects . Abnormal POCRs are frequently related to chronic obstructive pulmonary disease (COPD) [7,20]: in our experience, hyperaerated lung fields were shown in 34.3% of abnormal POCRs. In a French study , the frequency of abnormal POCRs increased from 6.2% in subjects without risk factors to 72.5% in patients with three risk factors. Moreover, the frequency of abnormalities may increase when POCRs are ordered more selectively (e.g. after the pre-anaesthetic examination, or in cardiac or respiratory diseases) [8,14-16], or in populations with increased occurrence of pulmonary diseases (e.g. tuberculosis), or when the patient's history is difficult to obtain due to cultural or linguistic differences .
The frequency of abnormal POCR may depend on radiographic techniques, may be related to the professional expertise of the radiologist and may be influenced by setting the limits of the normal range. Moreover, small nodular or pleural calcifications or negligible pleural synechiae can be reported as abnormal findings, but they may have an insignificant impact on anaesthetic management. Conversely, the presence of a severe POCR abnormality does not always demand the need for a change in anaesthetic plans. Additionally, as stated by Charpak et al., some aspects of the evaluation of usefulness are difficult to explain, such as the importance of normal findings.
In general, we could easily infer that POCR abnormalities are not really helpful in the evaluation of usefulness. Pre-operative chest radiographic abnormalities are obviously discovered after a POCR has been obtained; thus, they cannot represent predictive criteria to select which patient might be submitted to a POCR. For these reasons, POCR abnormalities, even though recorded as variables, have not been considered in this analysis.
Usefulness of POCR for anaesthetic management
In this study, a useful POCR was reported in 5.1% of patients. The anaesthetic management was changed in 72.2% of these subjects. The multicentre trial of the Royal College of Radiologists carried on 10 619 elective noncardiopulmonary surgical patients concluded that POCR did not influence the decision to use inhalation anaesthesia . McKee and Scott , in a series of 400 patients, showed that the management was changed in 0.013% of subjects. In the study by Perez et al., management was altered in only 0.56% of patients: this may be related to the particular group of selected subjects, namely ASA classes 1 and 2. In a French study , surgical and anaesthetic procedures were modified in 0.5% of patients, ranging from 0.1% in the no risk group to 1.4% in the high-risk group. Similar results were observed in a recent prospective multicentre study conducted in general and gastrointestinal surgery . Charpak et al. showed that POCR led to modifications of medical management in 5% of patients: anaesthetic management was influenced in 2.7%, but anaesthetists considered that 15% of POCRs were helpful at any time during the hospital stay. A meta-analysis of 21 studies showed that peri-operative management was altered in 0.1% of unsuspected POCR abnormalities . However, the majority of published studies [1,13,14,18,20,21,23,41] were not specifically designed to assess the usefulness of POCR for anaesthetic management. Some studies were methodologically weak, contradictory or inconclusive .
In this study, univariate analysis showed that a useful POCR was significantly related to several variables including hospital, type of surgery, age, sex, ASA classes, classification of the intervention and co-existing diseases. Using multivariate analysis, only male sex, age >60 years, ASA classes ≥3, and co-existing respiratory diseases were significantly related to the probability of a useful POCR. Pre-operative chest radiography is commonly ordered in patients with cardiovascular diseases [6,8]. However, the present study demonstrated that neither cardiac diseases and other diseases except respiratory, nor the classification of surgical interventions produced a significant effect in determining the probability of a useful POCR. However, to maintain a high degree of generality, we kept all the above variables as part of the definitive statistical model.
Hospitals and ASA classes ≥3 represented two important variables of the random effect model, influencing the probability of a useful POCR. Wide variations of b0 and b1 values, with opposite sign, have been encountered among hospitals. This suggests that the same patient, admitted to different hospitals, may have quite different probabilities of a useful POCR. According to the proposed equation, the base-line probability of a useful POCR can vary ten fold depending on the admitting hospital: this corresponds to a healthy (ASA classes ≤2), young (≤60 years), female subject submitted for a standard operation ('base-line' patient). Moreover, we could easily conclude that the probability of a useful POCR is still under the high 5% threshold, and therefore, all hospitals agree with the low utility of the POCR in the above mentioned group of patients. In any case, assuming a homogeneous population, the base-line probability of a useful POCR is roughly 1%. Conversely, in changing the patient's characteristics, the probability of a useful POCR was modified in all hospitals with wide differences among hospitals and patients.
Three main factors may explain the heterogeneity encountered in the present study. First, the outcome measure (i.e. usefulness) of this study is a subjective decision by the anaesthetist. Second, ASA physical status classification represents a subjective evaluation of the patient's condition and it is not sufficiently precise to ensure that all anaesthetists will classify the same patient correctly . This can be especially applied to values close to the cutoff point, so that patients could be included in ASA class 2 instead of class 3 or vice versa. Third, poor anaesthetic practice, lack of anaesthesiological culture or knowledge and availability of high technologies may interfere with the outcome (e.g. avoidance of nitrous oxide in bullous emphysema, endtidal CO2 monitoring). However, when heterogeneity is discovered, as in the present study, the random effect model has been properly applied.
We acknowledge important limitations in this study. By design, the specific reasons for the POCR causing an alteration in anaesthetic management were not analysed and the clinical impact of the usefulness was not studied. This pilot study has been specifically undertaken to assess the impact of POCR on patient care (i.e. intermediate outcome). The outcome measure in this study was the decision of the anaesthetist rather than a 'true' outcome measure such as postoperative morbidity, mortality, costs or length of hospital stay. The ideal study would have been a randomized controlled trial in which a group of patients were not submitted for a POCR: this approach would not be approved ethically .
In conclusion, this pilot study confirmed some of the previous findings in an audit of local practice and provided further evidence against the habit of indiscriminately ordering POCRs. We defined the impact of some risk factors on the usefulness of the routine POCR for the anaesthetic management, and we demonstrated specific hospital effects that modified the outcome. In general, in healthy (ASA class ≤2), male or female, young or elderly patients submitted for standard surgery, the probability of a useful POCR is mostly lower than 5%. This probability increased up to 81% in male or elderly subjects, or in the presence of a respiratory disease, or ASA classes ≥3.
The authors wish to thank Professor Alan R. Aitkenhead for reviewing the manuscript.
The following additional investigators participated in the V.E.R.O. chest radiograph study group. The study would not have been possible without their enthusiastic support. They collaborated in all phases of the study, attending a meeting where preliminary results were presented and discussed, and reviewing earlier drafts of the manuscript. Ospedale di Cattinara, Trieste: G. Propedo; Ospedale di San Daniele del Friuli: A. Facin, L. Minin; Ospedale di Latisana: F. Marraro, R. Ricci; Centro di Riferimento Oncologico, Aviano: F. Fabiani, F. Coran; Ospedale S.Maria della Misericordia, Udine: M. Bertolissi, M. Licari, P. Raneri, A. Modesto, G. Tomasini; Ospedale di Pordenone: G. Nadalin, G. Girelli; Ospedale di San Vito al Tagliamento: G. Zannier, T. Veronese;. Casa di Cura Salus, Trieste: A. Grube, V. Silvano; Ospedale Maggiore, Trieste: P. Zanei, P.Bortolotto; Policlinico Universitario, Udine: A.M. Vecellio, A. Zingarelli: Ospedale di Monfalcone: A. Paravano, P. Cassetti; Policlinico Città di Udine, Udine: E. Spadacci; Ospedale di Palmanova: M. Dragani, P. Pellegrini;. Ospedale di Gorizia: M.P. Del Litto, M. Abbona; Ospedale di Maniago: S. Fai; Ospedale di Tolmezzo: A. Bassini; Casa di Cura Sanatorio Triestino, Trieste: L. ladanza, P. de Morpurgo; Ospedale di Gemona: G. Degano; Ospedale di Cividale: M.G. Fabiani, F. Lombardo; Ospedale di Spilimbergo: C. Carini.
1 Perez A, Planell J, Bacardaz C et al.
Value of routine preoperative tests: a multicentre study in four general hospitals. Br J Anaesth
2 Bouillot JL, Paquet JC, Coggia M. La radiografie thoracique preoperatoire systématique en chirurgie générale est-elle utile? Ann Fr Anesth Rean
3 Macpherson DS. Preoperative laboratory testing: should any tests be 'routine' before surgery? Med Clin North Am
4 Tsai TW, Gallagher FJ, Lombardi G, Gennis P, Carter W. Guidelines for selective ordering of admission chest radiography
in adult obstructive airway disease. Ann Emerg Med
5 Archer C, Levy AR, McGregor M. Value of routine preoperative chest x-rays: a meta-analysis. Can J Anaesth
6 Roizen MF. Cost-effective preoperatory laboratory testing. JAMA
7 Gagner M, Chiasson A. Preoperative chest X-ray films in elective surgery: a valid screening tool. Can J Surg
8 Charpak Y, Blery C, Chastang C, Szatan M, Fourgeaux B. Prospective assessment of a protocol for selective ordering of preoperative chest x-rays. Can J Anaesth
9 Royal College of Radiologists. Preoperative chest radiology. Lancet
10 Lurie P. Toward optimal use of radiographs. Should admission and preoperative chest films be routine. Postgrad Med
11 Tape TG, Mushlin AL. The utility of routine chest radiographs. Ann Intern Med
12 Bass EB, Steinberg EP, Luthra R et al.
Do ophthalmologists, anesthesiologists, and internists agree about preoperative testing in healthy patients undergoing cataract surgery? Arch Ophthalmol
13 Escolano F, Alonso J, Gomar C, Sierra P, Castillo J, Castano J. Utilidad de la radiografia preoperatoria de torax en cirurgia electiva. Rev Esp Anestesiol Reanim
14 Wienceck RG, Weaver DW, Bouwman DL, Sachs RJ. Usefulness of selective preoperative chest X-ray films. A prospective study. Am Surg
15 Umbach GE, Zubek S, Deck H-J, Buhl R, Bender HG, Jungblut RM. The value of preoperative chest X-rays in gynaecological patients. Arch Gynecol Obstet
16 Tape TG, Mushlin AI. How useful are routine chest X-rays of preoperative patients at risk for postoperative chest disease? J Gen Intern Med
17 Rutten CL, Post D, Smelt WL. Het poliklinische preoperatieve onderzoek door de anesthesioloog. I. Minder verrichtingen en preoperatieve opnamedagen. Ned Tijdschr Geneeskd
18 Panich A, Panjasawadwong Y. Routine preoperative chest roentgenogram and its impact on intraoperative complications. J Med Assoc Thai
19 Golub R, Cantu R, Sorrento JJ, Stein HD. Efficacy of preadmission testing in ambulatory surgical patients. Am J Surg
20 Boghosian SG, Mooradian AD. Usefulness of routine preoperative chest roentgenograms in elderly patients. J Am Geriatr Soc
21 Rucker L, Frye EB, Staten MA. Usefulness of screening chest roentgenograms in preoperative patients. JAMA
22 McKee RF, Scott EM. The value of routine preoperative investigation. Ann R Coll Surg Engl
23 Wood RA, Hoekelman RA. Value of the chest X-ray as a screening test for elective surgery in children. Pediatrics
24 Tornebrandt K, Fletcher R. Pre-operative chest x-rays in elderly patients. Anaesthesia
25 Rees AM, Robert CJ, Bligh AS, Evans KT. Routine preoperative chest radiography
in non-cardiopulmonary surgery. Br Med J
26 Sykes MK. Essential monitoring. Br J Anaesth
27 Dripps RD, Lamont A, Eckenhoff JE. The role of anesthesia in surgical mortality. JAMA
28 Owens WD, Felts JA, Spitznagel EL. ASA physical status classifications: a study of consistency of ratings. Anesthesiology
29 Larson CP Jr Evaluation of the patient and preoperative preparation. In: Barash PG, Cullen BF, Stoelting RK.eds. Clinical Anesthesia
2nd edn. Philadelphia: Lippincott Company, 1992: 545-562.
30 Karim MR. Generalized linear models with random effects: a Gibbs sampling approach.
PhD Thesis, Baltimore, Maryland, The John Hopkins University, Department of Biostatistics 1991.
31 Breslow NE, Clayton DG. Approximate inference in generalized linear mixed models. JASA
32 Lindstrom MJ, Bates DM. Nonlinear mixed effects models for repeated measures data. Biometrics
33 Griffiths DA. Maximum likelihood estimation for betabinomial distribution and an application to the household distribution of the total number of cases of a disease. Biometrics
34 Feldman HA. Families of lines: random effects in linear regression analysis. J Appl Physiol
35 Lange N, Ryan L. Assessing normality in random effect models. Ann Stat
36 Pinheiro JC, Bates DM. LME and NLME, Mixed Effects Models Methods and Classes for S and S-plus, Version 1.2.
Madison: University of Wisconsin, 1995.
37 Statistical Science, S-PLUS Guide to Statistical and Mathematical Analysis, Version 3.3.
Seattle: Statsci, Mathsoft Inc, 1995.
38 Lamers RJS, van Engelshoven JMA, Pfaff A. Nogmaals de routinematige preoperatieve thoraxfoto. Ned Tijdschr Geneeskd
39 Taylor HG, Stein CM. Clinical effect of admission chest X-ray in Zimbabwe. Lancet
40 Buillot JL, Fingerhut A, Paquet JC, Hay JM, Coggia M. Are routine preoperative chest radiographs useful in general surgery? A prospective, multicentre study in 3959 patients. Eur J Surg
41 Ugomori S, Hayasaka S, Noda S, Setogawa T. Chest X-rays before surgery. Ann Ophthalmol
42 Dalla Palma L, Dixon AK, Durand-Zaleski I, Reiser M, Soimakallio S, On behalf of the EAR Working Group on Cost-Effectiveness in Radiology. An overview of cost-effective radiology. Eur Radiol
Appendix 2 Calculation of the probability of a useful POCR.
Where: x=β0+b0+β1+β2+(β3 or β4) + (β6 or β7 or β8 or β9)+(β5+b1)
P is the probability that POCR influences anaesthetic management, i.e. useful POCR; e denotes Exp (Nepero's number = 2.718); β0-9 and b0-b1 values are obtained from Tables 4 and 5, respectively. β1-β9 values equal zero if the patient is female, with age ≤60, submitted to standard intervention, with ASA ≤2, and without comorbidities.