Background Serial measurements of peak expiratory flow (PEF) are a recommended method for confirming a diagnosis of occupational asthma and are the only available method for low molecular weight agents available to many non-specialists. There is a tradeoff between accepting only measurements fulfilling quality standards with reduced data quantity and accepting all measurements irrespective of quality. We have investigated the effect of systematically reducing quality or quantity on the diagnostic sensitivity and specificity of these records in the diagnosis of occupational asthma using the Oasys system.

Methods Serial PEF measurements from 36 specific inhalation challenge positive occupational asthmatics and 44 non-occupational asthmatics were used. Records contained 4 weeks of ≥ 4 PEF readings/day for 75% of days. PEFs were measured on metres without any data quality requirements. Data was corrupted in 2 ways: 1) Each PEF measurement was randomly changed to be up to +50 L/min or -50 L/min from the original value in increments of 10 L/min. Records were randomised 3 times and the sensitivity and specificity compared at each randomisation to the original using the Oasys score, area between curves (ABC) score and timepoint analysis. 2) Independently, the number of readings per day were reduced sequentially from ≥7 readings per day to 2 readings per day. The sensitivity and specificity of the Oasys score, area between curves (ABC) score and timepoint analysis were compared after each reduction.

Results Random alteration of individual readings had small effects on sensitivity and specificity at each randomisation (Table 1). When the number of readings were reduced, the sensitivity of the Oasys score and ABC score was extremely robust in all reductions down to 3 readings per day. The sensitivity of the timepoint analysis was more affected. With only 2 readings, sensitivity was reduced for all scores. Specificity was unaffected by data reduction (Table 1).

Conclusion Specificity was not reduced by adding random errors to the peak flow measurements nor through data reduction. Sensitivity was reduced, relatively more for the timepoint analysis, but in 2/3 randomisations it was preserved for the Oasys and ABC systems. Oasys analysis is robust despite decreasing data quality and quantity.