中断的时间序列 - 数据转换 - SQL Server

Question

我正在使用 Microsoft SQL Server 2012，如果您能提供帮助，我有 2 个问题。

我有 2 个主表：

Individual  Age  Gender
    1       66     1
    2       33     0
    3       72     1

Individual Appointment_Date Clinic
    1         10-12-2018      A 
    1         05-03-2019      A  
    1         31-08-2019      A 
    2         02-12-2018      A  
    2         16-03-2019      A  
    2         12-04-2019      A   
    2         16-05-2019      A 
    3         07-12-2018      B  
    3         12-05-2019      B 
    3         19-06-2019      B 

我想运行中断时间序列分析，以了解干预对结果指标的影响。我没有这个结果指标，我需要每周使用 Total_Appointments 来估计它。我计算了总时间中每个人的 Total_Appointments，如下所示：

SELECT T1.Individual, COUNT(DISTINCT T2.Appointment_Date) AS [TOTAL_APPOINTMENTS]
FROM T1 LEFT JOIN T2
     ON T1.Individual = T2.Individual AND
        T2.APPOINTMENT_DATE > '2018-12-01' AND
        T2.APPOINTMENT_DATE < '2019-08-31'

问题 1：我如何估算每个人和每周的 Total_Appointments？这是我需要执行的中间步骤，但我最终会得到一个在临床层面汇总的结果指标。

然后，我还需要将其他变量（Avg_Age、Avg_Gender）转换为固定的时间间隔，例如：

Date_start   Date_end   Weeks_passed Clinic Outcome Avg_Age Avg_Gender Intervention_occurs
2018-12-02   2018-12-08       1         A      0.1      63      0.7           0
2018-12-02   2018-12-08       1         B      0.3      66      0.5           0
2018-12-02   2018-12-08       1         C      0.2      67      0.6           0
2018-12-09   2018-12-15       2         A      0.2      64      0.7           0
2018-12-09   2018-12-15       2         B      0.4      65      0.6           0
2018-12-09   2018-12-15       2         C      0.3      66      0.6           0
2018-12-16   2018-12-22       3         A      0.3      64      0.7           0
2018-12-16   2018-12-22       3         B      0.3      65      0.6           0
2018-12-16   2018-12-22       3         C      0.4      66      0.6           0
2018-12-23   2018-12-29       4         A      0.5      64      0.7           1
2018-12-23   2018-12-29       4         B      0.2      65      0.6           1
2018-12-23   2018-12-29       4         C      0.3      66      0.6           1
2018-12-30   2019-01-05       5         A      0.6      64      0.7           1
2018-12-30   2019-01-05       5         B      0.5      65      0.6           1
2018-12-30   2019-01-05       5         C      0.3      66      0.6           1
2019-01-06   2019-01-12       6         A      0.6      64      0.7           1
2019-01-06   2019-01-12       6         B      0.5      65      0.6           1
2019-01-06   2019-01-12       6         C      0.3      66      0.6           1

这项研究从 2018 年 12 月 2 日（星期日）开始，直到 2019 年 8 月 31 日（星期六）。

问题 2： 您能告诉我如何在 SQL Server 中创建这样的时间序列表吗？然后我会将其导入 R 并在那里运行实际分析。

Answer 1

对于第 1 部分，这样做：

with t as (
select 1 individual, cast('20181210' as date) date, 'A' clinic union all
select 1 individual, cast('20190305' as date) date, 'A' clinic union all
select 1 individual, cast('20190831' as date) date, 'A' clinic union all
select 2 individual, cast('20181202' as date) date, 'A' clinic union all
select 2 individual, cast('20190316' as date) date, 'A' clinic union all
select 2 individual, cast('20190412' as date) date, 'A' clinic union all
select 2 individual, cast('20190516' as date) date, 'A' clinic union all
select 3 individual, cast('20181207' as date) date, 'B' clinic union all
select 3 individual, cast('20190512' as date) date, 'B' clinic union all
select 3 individual, cast('20190619' as date) date, 'B' clinic 
), t1 as (
select *, cast(datepart(year, date) as char(4)) + cast(datepart(week, date) as char(2)) yearweek from t
)
select count(distinct date) cnt, individual, yearweek from t1 group by individual, yearweek

它创建一个使用datepart 的列yearweek，以便能够按周分组。

现在我可以使用分析函数而不是 group by 创建整个列表。有一个小问题，因为 Microsoft Sql 不允许 count(distinct column) over (partition by ...) - 我为此使用了一种解决方法，使用了两次 dense_rank() over (...)。我还添加了一周开始专栏和一周结束专栏。 和以前一样，我使用 CTE 来首先创建示例数据，然后执行所需的修改 ：

with t as (
select 1 individual, cast('20181210' as date) date, 'A' clinic union all
select 1 individual, cast('20190305' as date) date, 'A' clinic union all
select 1 individual, cast('20190831' as date) date, 'A' clinic union all
select 2 individual, cast('20181202' as date) date, 'A' clinic union all
select 2 individual, cast('20190316' as date) date, 'A' clinic union all
select 2 individual, cast('20190412' as date) date, 'A' clinic union all
select 2 individual, cast('20190516' as date) date, 'A' clinic union all
select 2 individual, cast('20190514' as date) date, 'A' clinic union all--new row to demonstrate count-distinct
select 3 individual, cast('20181207' as date) date, 'B' clinic union all
select 3 individual, cast('20190512' as date) date, 'B' clinic union all
select 3 individual, cast('20190619' as date) date, 'B' clinic 
), t1 as (
select *, cast(datepart(year, date) as char(4)) + cast(datepart(week, date) as char(2)) yearweek,
dateadd(day,   -((5 + DATEPART(dw, date) + @@DATEFIRST) % 7), date) start_of_week,
dateadd(day,  7-((5 + DATEPART(dw, date) + @@DATEFIRST) % 7), date) end_of_week
 from t
)
select dense_rank() over (partition by individual, yearweek order by date) + dense_rank() over (partition by individual, yearweek order by date desc) - 1 cnt_distinct, 
individual, yearweek, start_of_week, date, end_of_week
from t1 
order by yearweek, individual