Subjects and Exacerbations
Subjects were recruited from the London COPD cohort, first established in 1995 to investigate the causes and mechanisms of exacerbation. The methodology has been previously described [6]. In brief, patients attend for regular clinical assessments to comprehensively phenotype their disease, and they complete daily diary cards. The diary cards include assessment of PEF (best of three attempts) and change in respiratory symptoms. Symptoms are only recorded when they are new or worse than usual. The symptoms are classified as major (breathlessness, sputum volume and sputum purulence) and minor (cough, wheeze, sore-throat and coryza). In this way we can monitor PEF and symptoms prior to, at the onset of, and during exacerbations.
Exacerbations were defined using criteria that we have previously validated against health-status [9], airway inflammatory markers [10], lung function decline [11] and physical activity [12]. As in these previous studies, exacerbations were defined as two or more days of ≥2 new or worsening symptoms, at least one of which must have been a major symptom. The first of these two days was the day of onset. An exacerbation was considered to have recovered on the first of two consecutive symptom free days. This methodology also allowed us to calculate each patient's exacerbation frequency over the 12-month period prior to recording pulse oximetry. The number of major and minor symptoms recorded on each exacerbation day was summed to give a 'Symptom Count', an assessment of exacerbation severity which can vary between zero and seven. As patients only record symptoms when they are new or worse than normal, the Symptom Count is usually zero in stable COPD. Days on which symptoms were recorded, but which did not reach our criteria for exacerbation (and were not during an exacerbation) were termed 'symptom days', representing day-to-day symptom variation. Days on which no additional symptoms had been recorded, the majority, were termed 'symptom free days'.
Patients were originally recruited from primary and secondary care and inclusion criteria included a post-bronchodilator Forced Expiratory Volume in 1 second (FEV1) <80% predicted, FEV1/FVC (Forced Vital Capacity) ratio <0.7, minimal or no reversibility to β2-agonists (<200 ml and/or 15%) and the absence of other significant respiratory disease. At recruitment, a full clinical assessment was performed, together with spirometry (volumetric storage spirometer 122, Vitalograph 2160, Maids Moreton, Buckingham, UK). The St. George's Respiratory Questionnaire (SGRQ) was used to assess health-status [13].
The study was approved by Royal Free Hampstead NHS Trust ethics committee, and all patients provided written informed consent.
Domiciliary Pulse Oximetry
Twenty-eight Nonin 9500 Onyx™ pulse oximeters (Plymouth, Minnesota, US) were rotated through 40 patients between September 2008 and February 2010 (a convenience sample of those attending for routine clinic appointments). The equipment complies with ISO 10993-1 (Biological Evaluation of Medical Devices) and has a manufacturer-stated accuracy of +/-3 min-1 within the range 20-250 min-1 and +/-2% SpO2 (compared to invasive assessment) within the range 70-100%. All the patients were in sinus rhythm, none had respiratory failure, and none were using supplementary oxygen or ventilatory support.
Patients were asked to record their daily heart rate and SpO2 on a modified version of our standard diary card (in addition to changes in symptoms and PEF). Patients were given written instructions reminding them to take the oximetry reading at the same time each morning, before morning medication, before performing the PEF test, from the same finger, and after sitting for ten minutes at rest. Patients were asked to record the highest value obtained during each measurement.
Statistical Analysis
Data were analysed using SPSS version 14.0 http://www.spss.com. Normally distributed data are reported as mean and standard deviation (SD), non-parametric data as median and interquartile range (IQR). Comparisons between groups employed t-test or ANOVA as described in the Results. Correlations employed Pearson and Spearman techniques.
We first assessed the extent of day-to-day variation in heart rate, SpO2 and PEF. To do this, we calculated the individual mean and SD of these variables in the stable (symptom free) state. After initial distribution of the oximeters to each patient, the first seven days of data were ignored, to allow each patient to develop an oximetry measurement routine. The mean and SD of the heart rate, SpO2 and PEF were then calculated for the next 30 symptom free days (these data were normally distributed). For this analysis, data from seven days preceding any exacerbation to 14 days following exacerbation recovery were excluded. Symptom days were also ignored. On 34 otherwise stable days (3.7%), in 30 of the 31 patients, data were incomplete necessitating the inclusion of additional days such that each patient had thirty stable days monitoring of all three variables (the final patient had a prolonged period in which oxygen saturation was not recorded, prior to re-education). A period of 30 days was chosen after initial inspection of the data revealed stabilisation of the SD took up to three weeks. This methodology implies that on subsequent baseline (symptom free) days, 95% of values would be expected to lie between the mean +/- 1.96 (approximately 2) SD.
We next examined the time course of changes in heart rate, SpO2 and PEF in prospectively observed exacerbations, in those patients with stable baseline data. Data were examined from seven days prior, to seven days following the onset of exacerbation symptoms. There were 13 exacerbations in 13 subjects for this analysis, which is therefore based on 195 patient-days of exacerbation monitoring.
The data in individual patients were first expressed as multiples of that patient's SD. This is calculated using the formula (VD-VM)/VSD, where the measurement on that day, stable mean and stable standard deviation of each variable V (heart rate, oxygen saturation and PEF) are indicated by VD, VM and VSD respectively. For example, if a patient's mean (SD) heart rate when stable was 75 (5) min-1, and the heart rate on the day of onset was 85 min-1, the value recorded would be (85-75)/5 = +2. If the patient's heart rate had been 70 min-1, the value recorded would be (70-75)/5 = -1.
Next, we wished to explore whether changes in oximetry variables with or without PEF could reliably distinguish exacerbation onset from other days, particularly day-to-day symptom variation (symptom days). Given that exacerbations were typically associated with falls (rather than rises) in SpO2 and PEF, and a rise (rather than fall) in heart rate, we hypothesised that a composite score might perform better than the individual measurements alone. In preliminary analysis, we found no evidence for a significant correlation between SpO2 and heart rate in the stable state and therefore considered these variables independent. There was also no significant correlation between the greatest change in heart rate and greatest change in oxygen saturation over the first two exacerbation days, suggesting these variables were indeed independent (r = -0.32, p = 0.28). We therefore calculated an Oximetry Score (positive magnitude in SD units of the fall in SpO2 and rise in heart rate, /2) and an Oximetry-PEF Score (positive magnitude in SD units of the fall in SpO2, fall in PEF and rise in heart rate, /3).
We next assessed the ability of the Oximetry and Oximetry-PEF scores to differentiate exacerbation onset days from other days, in particular days on which symptom variation had been recorded (symptom days). To do this, we used all the remaining data that had not formed part of the above analyses (see Figure 1). Days were classified as symptom free, symptom, or exacerbation onset as described above. At exacerbation onset, the day with the higher score of the first two days was selected. All other exacerbation days were ignored. This analysis is based on 1580 days of follow-up (1469 symptom free days, 104 symptom days, and seven exacerbation onset days).
Receiver-operating characteristic (ROC) curves were constructed to assess the performance of oximetry indices and PEF in differentiating exacerbation onset from other days, and day-to-day symptom variation (symptom days). The test with the greatest area under the ROC curve was taken to have the best performance. We report sensitivity and specificity, but not positive predictive values as these would be dependent on the prevalence of the outcome (exacerbation) in this particular dataset.