Kaplan-Meier curves

Protobi can create Kaplan-Meier survival curves based on date/time data. These are useful to describe distributions of time intervals, particularly when the intervals may be ongoing.

For example, below shows the distribution of time on first line regimen from a patient chart review. The challenge is that many patients who started first line therapy are still on their first line therapy, and the exact duration is unknown and increasing.

Why use survival analysis?

In terms closer to home, this is like estimating the life of light bulbs when some bulbs have burned out while others are still shining and may do so for some time yet. Consider five light bulbs that started shining in January:

Three bulbs burned out, two are still shining.

• We can't just look at the average life of bubs that burned out (3, 6, 9 => 18/3 = 6.0 months) as that ignores the two still shining after 12 months.
• Nor can we just average using the time that the last two bulbs have shined so far as they may shine longer yet (3, 6, 9, 12, 12 => 42 /5 = 8.4 months)

Rather we need a method that lets consider the life span of the all the bulbs, both burned and still shining.

Kaplan Meier Survival Chart

Enter Kaplan Meier survival analysis. This was original designed for clinical trials to consider patient survival times in a simple unbiased way. But it applies to pretty much any time intervals we might elicit in surveys.

The Kaplan-Meier Survival estimator is a non-parametric statistic that handles open time intervals. The term "non-parametric" is statistical jargon that means "just the facts, no modeling."

The key information required is two fields, start date and end date, plus a boolean field indicating if the end date is closed  or "open" (i.e., still continuing at the time).

The figure above shows the median time on 1L therapy is 26 months, considering patients who have and have not actually finished.

Create a Kaplan Meier Survival chart in Protobi

To create a Kaplan-Meier curve in Protobi, set the chartType to "KaplanMeier" and include an attribute "intervals" which defines an array of relevant time intervals.

Each interval is an object specifying its key ,title, fields that define the start date and end date, and a boolean field indicating whether the interval is censored.

"chartType": "KaplanMeier",
    "intervals": [
        {
            "censoredRight": "censored_reg1",
            "start": "dt_start_1L",
            "end": "dt_end_1L",
            "title": "1L",
            "key": "1L"
        }
    ],

These charts can also be cross-tabbed to compare differences by another variable like setting or location or market segment. Drag a variable title to the header and drop to create a crosstab: